ProPublica releases controversial report on formaldehyde safety standards; report faces criticism from multiple federal agencies including EPA, Department of Defense, USDA

A recent report from ProPublica irresponsibly seeks to mislead and scare the public about formaldehyde in the air around us.

Formaldehyde is a natural compound made of carbon, hydrogen, and oxygen. It is found in every living system – from plants to animals to humans. In fact, humans produce about 1.5 ounces of formaldehyde every day as part of our normal metabolic process. Formaldehyde is extensively regulated to protect human health and the environment, and decades of scientific evidence support safe levels of formaldehyde exposure at current regulatory standards.

However, ProPublica misleads its readers into thinking this ubiquitous, natural chemical is a health threat to Americans. ProPublica’s irresponsible reporting ignores a number of critical factors, including:

• That EPA’s approach to formaldehyde is out of step with the rest of the world, ignoring the state of the science on certain cancers and suggesting health effects at levels below natural concentrations in the environment, or background levels. This ignores recent conclusions by the World Health Organization and European Chemicals agency. In fact, EPA’s suggested workplace limits for formaldehyde are 30 times below the recently updated European Union occupational levels.
• Multiple EPA peer review bodies, including the National Academy of Sciences, EPA Human Studies Review Board, and EPA TSCA Scientific Advisory Committee on Chemicals, have strongly criticized EPA’s approach to formaldehyde toxicity as well as EPA conclusions on certain types of cancer.
• EPA’s approach to formaldehyde has been strongly criticized by other federal agencies, including Department of Defense, USDA, CDC, NASA, and the White House Office of Management and Budget.
• ProPublica ignores numerous EPA data limitations for AirToxScreen, including why people should not use this data to consider individual or local risks.
• Comments on EPA’s proposed Air Emissions Reporting Requirements rule outline how EPA’s National Emissions Inventory, AirToxScreen, and past determinations, as well as state environmental comments, support the overwhelming influence of biogenic, fire, and secondary sources (and the insignificance of industrial sources) for ambient formaldehyde concentrations. These comments also note studies that find disagreement between ambient formaldehyde concentrations through ground-based observations, satellite measurements, and modeled results. ¹
• Nearly all parts of the U.S. are far below EPA’s long-standing approach to unacceptable risks that need to be addressed under the Clean Air Act.
• Formaldehyde dissipates quickly in sunlight and EPA data fully acknowledges that nearly all formaldehyde in the outdoor air come from natural/biogenic sources and fires (primarily wildfires). For example, EPA’s most recent National Emissions Inventory shows that sources other than industry represent over 90 percent of emissions and natural/biogenic emissions contribute roughly 2,500 times more formaldehyde than industrial processes associated with chemical manufacturing. EPA data also shows that other sources like residential wood burning (more than 20 times more formaldehyde than chemical manufacturing) and commercial cooking (more than 7 times more than chemical manufacturing) are more significant sources.
• EPA’s AirToxScreen makes clear that virtually all formaldehyde modeled in this program comes from background and secondary formation.
• EPA’s draft TSCA risk evaluationPUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1 March 2024 2 United States Office of Chemical Safety and 3 Environmental Protection Agency Pollution Prevention 4 5 6 7 8 9 Draft Ambient Air Exposure Assessment for Formaldehyde 10 11 CASRN 50-00-0 12 13 14 15 16 17 18 19 20 21 22 23 24 March 2024 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 25 TABLE OF CONTENTS 26 EXECUTIVE SUMMARY ...................................................................................................................... 5 27 1 INTRODUCTION......................................................................................................................... 7 28 1.1 Risk Evaluation Scope .................................................................................................................. 7 29 1.2 Summary of the Chemistry, Fate, and Transport Assessment ...................................................... 8 30 1.3 Conceptual Exposure Model ........................................................................................................ 8 31 2 APPROACH AND METHODOLOGY .................................................................................... 10 32 2.1 Modeling ..................................................................................................................................... 10 33 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) ................................................... 10 34 2.1.1.1 Environmental Releases Evaluated ................................................................................ 11 35 2.1.1.2 Exposure Scenario .......................................................................................................... 12 36 2.1.1.3 IIOAC Model Output Values ......................................................................................... 13 37 2.1.1.3.1 Estimated Short-Term Formaldehyde Concentrations ............................................. 14 38 2.1.1.3.2 Estimated Long-Term Formaldehyde Concentrations ............................................. 14 39 2.1.1.4 Exposure Calculations .................................................................................................... 14 40 2.1.1.4.1 Acute Concentrations (AC) ...................................................................................... 14 41 2.1.1.4.2 Average Daily Concentration (ADC) ...................................................................... 15 42 2.1.1.4.3 Lifetime Daily Average Concentration (LADC) ..................................................... 15 43 Human Exposure Model (HEM) ........................................................................................... 16 44 Air Toxics Screening Assessment Tool (AirToxScreen) ...................................................... 17 45 2.2 Monitoring .................................................................................................................................. 18 46 Ambient Monitoring Technology Information Center (AMTIC) .......................................... 18 47 3 RESULTS .................................................................................................................................... 19 48 3.1 Modeling ..................................................................................................................................... 19 49 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) ................................................... 19 50 AirToxScreen ......................................................................................................................... 22 51 Human Exposure Model (HEM) ........................................................................................... 23 52 3.2 Monitoring .................................................................................................................................. 25 53 Ambient Monitoring Technology Information Center (AMTIC) .......................................... 25 54 3.3 Data Integration of Various Sources of Formaldehyde .............................................................. 28 55 4 STRENGTHS AND LIMITATIONS, ASSUMPTIONS, UNCERTAINTY, AND 56 CONFIDENCE STATEMENT ................................................................................................. 31 57 4.1 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) ........................................................ 31 58 4.2 AirToxScreen ............................................................................................................................. 31 59 4.3 Human Exposure Model (HEM) ................................................................................................ 32 60 4.4 Ambient Monitoring Technology Information Center (AMTIC) Dataset ................................. 32 61 4.5 Weight of Scientific Evidence .................................................................................................... 32 62 5 REFERENCES ............................................................................................................................ 34 63 APPENDICES ......................................................................................................................................... 35 64 Appendix A LIST OF SUPPLEMENTAL DOCUMENTS ............................................................ 35 65 Appendix B DATA SOURCES AND INFORMATION ................................................................ 36 66 B.1 Sensitivity Analyses Conducted to Inform Modeling for Formaldehyde .................................. 36 Page 2 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 67 Appendix C IIOAC RESULTS BY COU ........................................................................................ 40 68 69 LIST OF TABLES 70 Table 2-1. IIOAC Input Parameters for Stack and Fugitive Air Releases ................................................ 12 71 Table 2-2. HEM Input Parameters for Stack and Fugitive Air Releases .................................................. 16 72 Table 3-1. Overall Monitored Method Detection Limits (MDL) of Formaldehyde from AMTIC 73 Dataset (2015 to 2020) ...................................................................................................... 26 74 75 LIST OF FIGURES 76 Figure 1-1. Risk Assessment Document Map Summary ............................................................................ 7 77 Figure 1-2. Industrial Releases to the Environment and Pathways by Which Exposures of the General 78 Population to Formaldehyde May Occur ............................................................................ 9 79 Figure 3-1. Exposure Concentrations by TSCA COU for the 95th Percentile Release Scenario and 80 95th Percentile Modeled Concentration between 100 and 1,000 m from Industrial 81 Facilities Releasing Formaldehyde to the Ambient Air .................................................... 20 82 Figure 3-2. Distribution of Exposure Concentrations Modeled by IIOAC .............................................. 20 83 Figure 3-3. 2019 AirToxScreen Results ................................................................................................... 22 84 Figure 3-4. Map of Contiguous United States with HEM Model Results for TRI Releases Aggregated 85 and Summarized by Census Block ................................................................................... 24 86 Figure 3-5. Median and Maximum Downwind Concentrations (Fugitive, Stack, and Total Emissions) 87 across the 11 Discrete Distance Rings Modeled in HEM ................................................. 25 88 Figure 3-6. Histograms of Ambient Air Concentrations (µg/m3) of Formaldehyde across Contiguous 89 United States from 2015 to 2020 ...................................................................................... 27 90 Figure 3-7. Map of Monitoring Sites for Formaldehyde across the Contiguous United States................ 28 91 Figure 3-8. Distributions of AMTIC Monitoring Data, IIOAC Modeled Data, and AirToxScreen 92 Modeled Data .................................................................................................................... 29 93 Figure 3-9. Ambient Air Formaldehyde Concentrations by Sample Duration and Sampling Technique 30 94 95 LIST OF APPENDIX TABLES 96 Table_Apx C-1. Crosswalk of Industry Sector to COU along with Modeled Exposure Concentrations 97 by Industry Sector for 95th Percentile Release Scenario and 95th Percentile Modeled 98 Concentration at 100 to 1,000 m from Industrial Facilities Releasing Formaldehyde to 99 Ambient Air ...................................................................................................................... 40 100 Table_Apx C-2. Modeled Exposure Concentrations by COU for 95th Percentile Release Scenario and 101 95th Percentile Modeled Concentration at 100 to 1,000 m from Industrial Facilities 102 Releasing Formaldehyde to Ambient Air ......................................................................... 45 103 104 LIST OF APPENDIX FIGURES 105 Figure_Apx B-1. Sensitivity Analysis Conceptual Model for Exposure Scenarios Modeled for Max 106 and Mean Release Using IIOAC Model ........................................................................... 37 107 Figure_Apx B-2. Median and 95th Percentile Concentrations (Fugitive, Stack, and Total Emissions) 108 across the 11 Discrete Distance Rings Modeled in HEM ................................................. 39 109 Page 3 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Key Points: Ambient Air Exposure Assessment The following bullets summarize the key points of this draft ambient air exposure assessment: • Formaldehyde is ubiquitous and consistently present in the ambient air and therefore EPA quantitatively assesses human exposure to formaldehyde via the ambient air pathway. • Recognizing the ubiquity of formaldehyde in ambient air is due to contributions from many different sources, EPA considers and presents measured and modeled concentrations of formaldehyde from multiple lines of evidence, data, and analyses in this ambient air exposure assessment to evaluate and contextualize formaldehyde exposures in ambient air due to Toxic Substances Control Act (TSCA) conditions of use (COU). • EPA considered the last five years (2015 to 2020) of monitored formaldehyde concentrations extracted from EPA’s Ambient Monitoring Technology Information Center (AMTIC) dataset for this exposure assessment. EPA acknowledges this ambient monitoring data is generally representative of a total aggregate concentration from all formaldehyde sources (both TSCA and non-TSCA sources) but taken together with other data sources allows EPA to contextualize modeled concentrations from IIOAC relative to the monitored concentrations from AMTIC. • Additional analyses considered or conducted for this exposure assessment include 2019 AirToxScreen data [a screening analysis representing formaldehyde contributions from 37 different sources (including biogenic sources and secondary formation)], and additional modeling using EPA’s Human Exposure Model (HEM) [to evaluate national scale population impacts of and exposures to industrial releases of formaldehyde] • Acute and chronic exposures from industrial releases of formaldehyde which can be attributed to TSCA COUs based on the IIOAC modeling range from 0.0001 to 5.75 micrograms per cubic meter (µg/m3) with a median of 0.62 µg/m3 within the 100 to 1,000 m area distance evaluated. • Monitored formaldehyde concentrations extracted from AMTIC between 2015 and 2020 ranged from 0 to 60.1 µg/m3 with a median of 1.6 µg/m3. • When contextualizing modeled concentrations from EPA’s IIOAC modeling in relation to the other data sources considered, EPA found modeled concentrations generally fall within the lower quartile range of the AMTIC monitoring data and within the same range of concentrations attributed to secondary formation sources from the 2019 AirToxScreen dataset. Contributions from biogenic sources of formaldehyde from the 2019 AirToxScreen dataset generally fall within the lower quartile range of the IIOAC modeled concentrations. • Overall, EPA’s IIOAC results show 21 of 29 TSCA COUs have modeled concentrations greater than the 95th percentile concentration of 0.28 ug/m3 attributable to biogenic/natural sources (from 2019 AirToxScreen data). Eighteen TSCA COUs are greater than five times the 95th percentile concentration attributable to biogenic sources and seven TSCA COUs are greater than ten times the 95th percentile concentration attributable to biogenic sources of formaldehyde. • EPA is confident in the characterization of exposures to formaldehyde via the ambient air pathway-inhalation route in this ambient air assessment resulting from industrial facilities releasing formaldehyde to the ambient air. The greatest uncertainty is associated with the contribution of formaldehyde to the total ambient monitoring data due to secondary formation and as a byproduct of combustion. 110 Page 4 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 111 EXECUTIVE SUMMARY 112 Environmental releases of formaldehyde are reported to occur into the ambient air (U.S. EPA, 2024d). 113 Although subject to direct and indirect photolysis in the ambient air, formaldehyde is ubiquitous and has 114 consistently been found to be present in air based on testing and ambient monitoring implemented under 115 multiple EPA programs. 116 117 EPA’s assessment of the ambient air pathway considers both measured (monitored) and modeled 118 formaldehyde concentrations in ambient air. Modeling for this ambient air exposure assessment used 119 EPA’s Integrated Indoor/Outdoor Air Calculator (IIOAC) model to estimate air concentrations to which 120 people, including locally impacted and more sensitive populations, are exposed. The modeling for this 121 ambient air assessment focused specifically on exposures to individuals living nearby (within 100 to 122 1,000 meters (0.062 to 0.62 miles)) industrial facilities reporting releases of formaldehyde to the 123 ambient air which are associated with TSCA COUs. Exposures presented in this ambient air exposure 124 assessment were quantified from Toxic Release Inventory (TRI) reported release data across 35 industry 125 sectors representing 29 TSCA COUs. Exposures were also quantified from National Emissions 126 Inventory (NEI) reported release data and, while not summarized in this ambient air exposure 127 assessment, are included in the Draft IIOAC Assessment Results and Risk Calcs Supplement A to this 128 ambient air exposure assessment. 129 130 The Agency used monitoring data extracted from EPA’s AMTIC (U.S. EPA, 2022a) from 2015 through 131 2021 to contextualize modeled values as well as characterize total aggregate exposures to formaldehyde 132 from all possible contributing sources—including sources associated with TSCA COUs and other 133 sources out of scope for this assessment and not associated with TSCA COUs (e.g., biogenic sources 134 (decay of organic matter), secondary formation, combustion byproduct formation, other byproduct 135 formation, mobile sources, and others). EPA also considered existing modeled data from 2019 136 AirToxScreen to show how different sources (e.g., point sources, biogenic sources, and secondary 137 formation) contribute to total ambient air concentrations of formaldehyde. The Agency also used the 138 2019 AirToxScreen data to contextualize how formaldehyde concentrations in the ambient air associated 139 directly with TSCA COUs compare to contributions from other sources not associated with TSCA 140 COUs like biogenic-sourced formaldehyde. 141 142 Modeled results for acute and chronic exposures to populations living near industrial facilities releasing 143 formaldehyde to the ambient air ranged from 0.0001 to 5.75 µg/m3. Monitored formaldehyde 144 concentrations extracted from AMTIC (2015 to 2020) ranged from 0 to 60.1 µg/m3 with a median of 1.6 145 µg/m3 across more than 300,000 monitored values. Monitoring data from select monitoring sites located 146 near industrial releasing facilities showed comparable formaldehyde concentrations to modeled 147 concentrations from those same releasing facilities (within the same order of magnitude). Formaldehyde 148 concentrations attributable to biogenic or secondary production based on AirToxScreen range from 0.13 149 to 1.8 µg/m3. 150 151 Taken together, these data and results show modeled formaldehyde concentrations using IIOAC are 152 generally in the same range as monitored formaldehyde concentrations in ambient air as well as 153 contributions of formaldehyde to the ambient air from biogenic and secondary production. However, to 154 characterize exposures associated with TSCA COUs it is necessary to recognize modeled concentrations 155 may contribute to some part of the total concentrations captured by monitored data but are independent 156 and exclusive of non-TSCA sources like biogenic sources or secondary formation. Therefore, while 157 modeled concentrations are in the same range as biogenic and secondary formation, the exposures 158 resulting from TSCA COUs are representative of actual exposures from TSCA COUs alone, and not a Page 5 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 159 subset of the biogenic or secondary formation data. To contextualize this, actual exposures occurring at 160 a given location from both TSCA COUs and biogenic or secondary formation would be an additive 161 exposure receiving the full contribution of formaldehyde from TSCA COUs as well as the full 162 contribution of formaldehyde from biogenic sources. 163 164 Based on the modeled concentrations using IIOAC, the following TSCA COUs result in the highest 165 formaldehyde concentrations: 166 • Processing – incorporation into an article-adhesives and sealant chemicals; 167 • Processing as a reactant – intermediate; 168 • Processing – incorporation into a formulation, mixture, or reaction product-intermediate; 169 • Processing – incorporation into article-finishing agent; 170 • Processing – incorporation into a formulation, mixture, or reaction product-bleaching agents; 171 • Processing – incorporation into an article-paint additives and coating additives; and 172 • Industrial use – chemical substances in industrial products-paints and coatings; adhesives and 173 sealants, lubricants. 174 When IIOAC modeled concentrations associated with TSCA COUs are compared to the 95th percentile 175 concentration attributable to biogenic sources in the ambient air, EPA found 21 of 29 TSCA COUs are 176 greater than the 95th percentile concentration attributable to biogenic sources. Eighteen TSCA COUs are 177 greater than five times the 95th percentile concentration attributable to biogenic sources and 7 TSCA 178 COUs are greater than ten times the 95th percentile concentration attributable to biogenic sources, 179 180 EPA has high confidence in the overall characterization of exposures for this ambient air exposure 181 assessment. Exposure results relied upon direct reported releases, which can be tied to TSCA COUs and 182 peer-reviewed models to derive exposure concentrations at distances from releasing facilities where 183 individuals within the general population typically reside for many years. Additionally, monitoring data 184 from locations near industrial releasing facilities show comparable formaldehyde concentrations in the 185 ambient air and therefore provide added confidence to EPA’s use of modeled results to characterize 186 human exposures to formaldehyde via the ambient air pathway. 187 Page 6 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 188 1 INTRODUCTION 189 Formaldehyde is a naturally occurring aldehyde produced during combustion, decomposition of organic 190 matter, and in the human body as a normal part of metabolism. Formaldehyde is also released into the 191 ambient air by industrial operations involved with manufacturing, processing, formulation, disposal, and 192 other practices (U.S. EPA, 2024d). It may be distributed as a mixture known as formalin or as a solid 193 known as paraformaldehyde. This assessment focuses on formaldehyde after it has been released to air 194 as a gas (U.S. EPA, 2024b). Due to the previously mentioned natural occurrences and continuous 195 releases from industrial facilities, formaldehyde is ubiquitous in the outdoor environment. 196 197 Formaldehyde is a high priority chemical undergoing the TSCA risk evaluation process for existing 198 chemicals following passage of the Frank R. Lautenberg Chemical Safety for the 21st Century Act in 199 2016. It is concurrently undergoing a risk assessment under the Federal Insecticide, Fungicide, 200 Rodenticide Act (FIFRA) and EPA’s Integrated Risk Information System (IRIS) programs. This 201 ambient air exposure assessment considers TSCA COUs, as defined by the TSCA section 3(2) definition 202 of “chemical substance.” This TSCA-specific document serves to support risk management needs by the 203 Office of Pollution Prevention and Toxics (OPPT) and is one of many documents included within the 204 Draft Formaldehyde Risk Evaluation package. 205 1.1 Risk Evaluation Scope 206 The TSCA risk evaluation of formaldehyde comprises several human health, environmental, fate, and 207 exposure assessment modules and two risk assessment documents—the environmental risk assessment 208 and the human health risk assessment. A diagram showing the layout of these modular assessments and 209 the relationships between assessments is provided in Figure 1-1. This ambient air exposure assessment is 210 shaded blue. In some cases, individual assessments were completed jointly under TSCA and FIFRA. 211 These modules are shown in dark gray. 212 213 214 Figure 1-1. Risk Assessment Document Map Summary 215 Page 7 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 216 Environmental releases of formaldehyde are reported for ambient air in the Draft Environmental Release 217 Assessment for Formaldehyde (U.S. EPA, 2024d). The Draft Chemistry, Fate, and Transport 218 Assessment for Formaldehyde (U.S. EPA, 2024b) and available monitoring data from EPA’s AMTIC 219 indicate formaldehyde is ubiquitous and consistently present in the ambient air. Additional modeling and 220 data from the 2019 AirToxScreen support the ubiquity of formaldehyde in ambient air from multiple 221 sources, including non-TSCA sources. Considering these lines of evidence and the ubiquity of 222 formaldehyde in ambient air, EPA expects human exposure to formaldehyde via the ambient air to be 223 common and therefore quantitatively assesses human exposure to formaldehyde via the ambient air 224 pathway. 225 226 The scope of this ambient air exposure assessment focuses on human exposures to formaldehyde 227 resulting from industrial releases of formaldehyde to the ambient air that are associated with TSCA 228 COUs. Detailed descriptions of TSCA COUs considered are included in the Conditions of Use for the 229 Draft Formaldehyde Risk Evaluation (U.S. EPA, 2024c). Certain uses are considered out of scope for 230 this human health risk assessment, including combustion byproducts, mobile sources, biogenic sources, 231 secondary formation, and other non-TSCA sources. The Draft Conditions of Use Module also includes 232 discussion around those COUs that are considered out of scope. 233 234 Considering the ubiquity of formaldehyde in ambient air—while biogenic sources, secondary formation, 235 and other non-TSCA sources are out of scope—EPA considers the contributions of these sources to 236 ambient concentrations of formaldehyde to contextualize the relative contributions resulting from TSCA 237 COUs. However, for TSCA purposes, the Agency can only regulate formaldehyde resulting from TSCA 238 COUs, and to do so in a manner that could be complied with and enforced, needs to distinguish exposure 239 resulting from TSCA COUs from exposures resulting from other non-TSCA sources. 240 1.2 Summary of the Chemistry, Fate, and Transport Assessment 241 Formaldehyde is a colorless, flammable gas at room temperature and has a strong odor. As noted in the 242 Draft Chemistry, Fate, and Transport Assessment for Formaldehyde (U.S. EPA, 2024b), formaldehyde 243 is subject to direct and indirect photolysis and chemical transformation processes in the ambient air. In 244 direct sunlight, studies indicate formaldehyde has a half-life up to 4 hours, although in the absence of 245 direct sunlight formaldehyde has a half-life up to 114 hours. Formaldehyde transformations in the 246 ambient air include, for example, hydrolysis in humid or moist air to formic acid. Formaldehyde also 247 reacts with ozone, nitrates, and hydroxy radicals to form other chemicals in the ambient air. 248 Formaldehyde can also be formed in ambient air from parent compounds such as 1,3-butadiene, or 249 through other secondary formation paths as described in “Sensitivity of Ambient Atmospheric 250 Formaldehyde and Ozone to Precursor Species and Source Types Across the United States” (Luecken et 251 al., 2018). 252 1.3 Conceptual Exposure Model 253 EPA expects the ambient air pathway to be the predominant human exposure pathway to formaldehyde 254 in the outdoor environment as shown in Figure 1-2. In summary, formaldehyde is released from 255 industrial facilities as uncontrolled fugitive releases (e.g., process equipment leaks, process vents, 256 building windows, building doors, roof vents) and stack releases that may be either uncontrolled (e.g., 257 direct releases out a stack) or controlled with pollution control device (e.g., baghouse, scrubber, thermal 258 oxidizer). Once released to the ambient air, formaldehyde may move off-site into the surrounding areas 259 where humans may be exposed through inhalation. This draft ambient air exposure assessment focuses 260 on exposures to a subset of the general population living nearby industrial facilities releasing 261 formaldehyde to the ambient air. Page 8 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 262 263 264 Figure 1-2. Industrial Releases to the Environment and Pathways by Which Exposures of the 265 General Population to Formaldehyde May Occur Page 9 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 266 2 APPROACH AND METHODOLOGY 267 EPA considered both modeled and monitored formaldehyde concentrations in the ambient air for this 268 draft ambient air exposure assessment. EPA estimated both short-term (daily-averaged) and long-term 269 (annual-averaged) formaldehyde concentrations in ambient air for purposes of characterizing exposures 270 in this ambient air exposure assessment and deriving acute and chronic risk estimates for comparison to 271 human health hazard data in the Draft Human Health Risk Assessment for Formaldehyde (U.S. EPA, 272 2024e). 273 274 Given the complexities of the exposure assessment of formaldehyde in ambient air as previously 275 described, multiple yet complimentary lines of evidence were considered to understand and 276 contextualize the ambient air concentrations of formaldehyde resulting from TSCA COUs. These 277 evidence streams are summarized below and detailed in the following subsections. 278 1. Estimated Formaldehyde Concentration from Formaldehyde TSCA COUs: This draft 279 assessment uses EPA’s Integrated Indoor/Outdoor Air Calculator Model (IIOAC)1 to estimate 280 formaldehyde concentrations near releasing facilities based on reported formaldehyde release 281 data for TSCA COUs. In addition, EPA’s Human Exposure Model (HEM v4.2) is used to 282 estimate geographically specific formaldehyde concentrations based on site-specific reported 283 formaldehyde release information for formaldehyde TSCA COUs. 284 2. Relative Contributions of Formaldehyde Concentrations in Ambient Air: This draft assessment 285 uses the 2019 Air Toxics Screening Assessment Tool (AirToxScreen) to contextualize 286 formaldehyde concentrations in ambient air resulting from all known sources of hazardous air 287 pollutants (including formaldehyde). While there are multiple sources of formaldehyde to the 288 ambient air, this ambient air exposure assessment includes consideration of several larger 289 sources contributing formaldehyde to the ambient air including biogenic sources (natural 290 production), secondary formation (formed through chemical transformations like breakdown of 291 isoprene to formaldehyde), and point sources (stationary sources including industrial facilities 292 with releases of formaldehyde that can be attributed to TSCA-COUs). 293 3. Measured Formaldehyde Concentrations: This assessment summarizes monitoring data from 294 EPA’s AMTIC (U.S. EPA, 2022a) to understand aggregate or total formaldehyde concentrations 295 in ambient air and to characterize modeled concentrations of formaldehyde. 296 2.1 Modeling 297 As described above, EPA uses multiple approaches to understand formaldehyde concentrations as well 298 as the relative contributions of those concentrations in ambient air. 299 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) 300 EPA used the IIOAC Model to estimate daily-averaged and annual-averaged formaldehyde 301 concentrations for a suite of exposure scenarios at three predefined distances from a facility releasing 302 formaldehyde to the ambient air. EPA’s modeling evaluated industrial releases of formaldehyde which 303 are associated with TSCA COUs from two separate databases: Toxic Release Inventory (TRI) and 304 National Emissions Inventory (NEI). While releases from both databases were evaluated, EPA 305 compared releases and modeled concentrations from the two databases and found results were similar 306 (i.e., fell within the same estimated distribution range). Therefore, to provide a clearer picture of 307 findings, EPA only presents results from the TRI dataset in this ambient air exposure assessment. 1 For further information see the IIOAC homepage. Page 10 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 308 Nonetheless, results from all exposure scenarios and datasets evaluated are provided in the “Draft 309 Ambient Air Exposure Assessment Results and Risk Calcs Supplement A.” 310 311 IIOAC is a spreadsheet-based tool that estimates indoor and outdoor air concentrations using pre-run 312 results from a suite of dispersion scenarios in a variety of meteorological and land-use settings within 313 EPA’s American Meteorological Society/Environmental Protection Agency Regulatory Model 314 (AERMOD). 2 As such, IIOAC is limited by the parameterizations utilized for the pre-run scenarios 315 within AERMOD (meteorologic data, stack heights, distances, exposed population, etc.). Additional 316 information on IIOAC can be found in the user guide (U.S. EPA, 2019b). 317 2.1.1.1 Environmental Releases Evaluated 318 As further discussed in Draft Environmental Release Assessment for Formaldehyde (U.S. EPA, 2024d), 319 EPA developed the air release estimates included in this assessment using the 2016 through 2021 320 reporting years for TRI (U.S. EPA, 2022b) and 2017 NEI (U.S. EPA, 2019a). These databases were 321 queried in 2022 and include the most recent 6 years available at that time in the TRI database and the 322 most recent reporting year available for NEI (2017) and received a high-quality rating under EPA’s 323 systematic review process. NEI provides emission data at the process unit level and incorporates a larger 324 database of site emission data. In total, EPA identified more than 150,000 separate emission data records 325 (including unit-level estimates) for formaldehyde across both databases. 326 327 Due to the substantial available release information, EPA used site-level emission data and developed a 328 new industrial-sector approach to categorize the release data. The Agency uses the primary North 329 American Industry Classification System (NAICS) codes reported by the reporting facility to group 330 facilities by their primary industry sector. By taking an industry sector approach in this TSCA ambient 331 air exposure assessment, EPA can more directly map exposures to industrial TSCA COUs (U.S. EPA, 332 2024e). This, in turn, informed which TSCA COUs are contributing to human exposures to 333 formaldehyde via the ambient air. From these industry sectors, EPA calculated from the distribution 334 (non-zero) of site-specific releases from each database across all years (e.g., 2015 to 2020 for TRI data 335 and separately 2017 for NEI data): and extracted: (1) the maximum reported release, (2) the 95th 336 percentile release, (3) the median release, and (4) the minimum release for each industrial sector. These 337 industrial sectors are then used to characterize TSCA COUs. The emissions described herein were used 338 as direct inputs to EPA’s IIOAC Model to estimate ambient air concentrations of formaldehyde for each 339 sector from each database. 340 341 The Draft Environmental Release Assessment for Formaldehyde (U.S. EPA, 2024d) and associated 342 supplemental files provide detailed descriptions of the methods used to extract release data, calculate the 343 emission statistics described above, and all releases considered for this ambient air exposure assessment. 344 To summarize, the TRI database included 810 reporting facilities across the 6 years of reporting data 345 queried. Fugitive releases ranged from 4.5×10−5 kg/year (minimum) to 14,272 kg/year (maximum), 346 while stack releases ranged from 7.1×10−5 kg/year (minimum) to 158,757 kg/year (maximum). The NEI 347 database included nearly 150,000 reported releases from the 2017 reporting year. Fugitive releases 348 ranged from 1×10−14 kg/year (minimum) to 138,205 kg/year (maximum), while stack releases ranged 349 from 1×10−11 kg/year (minimum) to 1,412,022 kg/year (maximum). Although the formaldehyde release 350 amounts reported in both databases vary, the variability within an individual database is due to facility 351 size, process rates, and several other factors, while variability across the two databases is due to 352 differences in the criteria for reporting between the programs. Neither is reflective of uncertainty in the 353 reported releases themselves. 2 See AERMOD for further information. Page 11 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 354 2.1.1.2 Exposure Scenario 355 IIOAC can model a variety of user defined input parameters and exposure scenarios, including varying 356 release scenarios/patterns, release types, release durations, urban and rural settings (topography), and 357 meteorological conditions. EPA modeled stack and fugitive releases separately, each using the default 358 release parameters integrated within the IIOAC Model along with a user-defined length and width for 359 fugitive releases as listed in Table 2-1 360 361 Table 2-1. IIOAC Input Parameters for Stack and 362 Fugitive Air Releases Stack Release Parameters Value Stack height (m) 10 Stack diameter (m) 2 Exit velocity (m/sec) 5 Exit temperature (°K) 300 Fugitive Release Parameters Value Length (m) 10 Width (m) 10 Angle (degrees) 0 Release height (m) 3.05 363 364 For this ambient air exposure assessment, EPA categorized reporting industrial facilities into 35 industry 365 sectors (associated with 29 TSCA COUs) for the TRI dataset and 46 industry sectors for the NEI dataset. 366 For each industry sector, results presented in this draft ambient air exposure assessment are based on 367 1. The 95th percentile release value reported by industry sector (mapped to respective TSCA COUs 368 for the TRI dataset) assuming the total annual reported releases are continuous and equally 369 distributed across all days of operation. 370 2. The 95th percentile modeled daily- and annual-averaged air concentrations from the IIOAC 371 output files at a distance of 100 to 1,000 m from the release point to characterize exposures. 372 3. An operating scenario representing industrial facilities releasing formaldehyde to the ambient air 373 operating 24 hours/day, 7 days/week, 365 days/year. 374 This exposure scenario was selected to represent a national level exposure estimate inclusive of sensitive 375 and locally impacted populations. The 95th percentile release scenario and modeled concentrations were 376 used to represent a more national level exposure estimate based on actual reported releases. The 100 to 377 1,000 m area distance is selected because it represents an area distance where a larger community may 378 reside and be exposed rather than a smaller number of individuals who may reside at a given finite 379 distance ring. The operating scenario used for this assessment was selected because it is representative 380 of typical operating conditions under which industrial facilities involved with formaldehyde 381 manufacturing, processing, etc. operate. While this scenario is representative of a high-end exposure 382 scenario which is inclusive of more sensitive and locally impacted populations, it is not a maximum 383 worst-case exposure scenario and thus considered more representative of an overall community or 384 nationally representative exposure scenario. 385 Page 12 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 386 The scenario further represents exposed residents experiencing a lifetime continuous exposure to the 387 total formaldehyde release based on reasonably conservative assumptions about release and stack 388 parameters based on two separate sets of sensitivity analyses described in Appendix B.1. Because of the 389 exposure scenario used, the daily-averaged modeled concentration and annual-averaged modeled 390 concentration output values from the IIOAC model are the same. Results from this exposure scenario are 391 summarily presented independently in the “Draft Ambient Air Exposure Assessment Results and Risk 392 Calcs Supplement B.” 393 394 In addition to the exposure scenario described above, EPA assessed a suite of additional exposure 395 scenarios including the consideration of stack and fugitive releases, urban and rural topography, and a 396 second industrial operating scenarios (250 days/year, 8 hours/day, 5 days/week) for both the TRI and 397 NEI datasets. All exposure scenarios modeled used the south coastal meteorological data, which 398 represents a more conservative meteorological region within the IIOAC model in relation to the other 13 399 regional meteorological stations included within the IIOAC model. In total, the IIOAC output results 400 represent 32 exposure scenarios for each industry sector for both the TRI and NEI datasets. While not 401 presented directly in this ambient air exposure assessment, all results for these additional exposure 402 scenarios are included in the Draft Ambient Air Exposure Assessment Results and Risk Calcs 403 Supplement A. 404 2.1.1.3 IIOAC Model Output Values 405 The IIOAC Model provides multiple output values including daily-averaged and annual-averaged high- 406 end (95th percentile) and mean air concentrations of formaldehyde at three pre-defined distances (100, 407 100 to 1,000, and 1,000 m) from releasing facilities. Daily-averaged concentrations are the average of all 408 modeled hourly concentrations at each modeled exposure point for each day of operation modeled 409 within IIOAC. For the two finite distances (100 and 1,000 m), IIOAC considers a polar grid of 16 410 modeled exposure points placed equidistance around each distance ring and 5 years of meteorological 411 data. Since the exposure scenario modeled assumes 365 days of operation each year (366 for a leap 412 year), this results in a total of 29,216 separate daily-averaged concentrations at each of the three 413 distances evaluated. The 95th percentile daily-averaged concentration for the 100 and 1,000 meter 414 distances therefore represents the 95th percentile daily-averaged concentration across the entire 415 distribution of 29,216 daily-averages modeled within IIOAC. The 95th percentile generally represents a 416 downwind concentration at a modeled exposure point where a resident lives downwind of the industrial 417 release, taking into account a non, site-specific estimate for dispersion and dilution. 418 419 For the area distance of 100 to 1,000 m, IIOAC considers a cartesian grid of 456 receptors placed at 100 420 m intervals across the entire area distance. The daily-averaged concentrations then represent the 421 arithmetic average of all daily-averaged concentrations across all 456 receptors for each day. Since 422 IIOAC considers 5 years of meteorological data and our exposure scenario assumes 365 days of 423 operation each year (366 for a leap year), this results in a total of 2,191 separate daily-averaged 424 concentrations. The 95th percentile daily-averaged concentration for the 100 to 1,000 m area distance 425 therefore represents the 95th percentile daily-averaged concentration across the entire distribution of 426 2,191 daily-averaged concentrations modeled within IIOAC. 427 428 Annual-averaged concentrations are the 5-year average of all modeled daily-averaged concentrations at 429 each modeled exposure point within IIOAC. Since IIOAC considers 16 modeled exposure points and 430 one 5-year average modeled concentration, 16 annual-averaged concentrations are produced at each 431 distance evaluated for the 100 and 1,000 m finite distances. For the 100 to 1,000 m area distance, IIOAC 432 considers the arithmetic average of all modeled daily-averaged concentrations and a single modeled 433 exposure point, the 95th percentile annual-averaged concentration is the 95th percentile modeled Page 13 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 434 concentration across the entire distribution of 2,191 modeled daily-averaged concentrations within 435 IIOAC and generally represents a downwind concentration at a modeled exposure point where to which 436 a resident living approximately downwind of the industrial release, taking into account a non, site- 437 specific estimate for dispersion and dilution. 438 439 Based on the IIOAC users guide, the Fenceline average represents the daily-averaged and annual- 440 averaged concentrations at 100 m distance from releasing facilities. The community average represents 441 the daily-averaged and annual-averaged concentrations within the area distance of 100 to 1,000 m from 442 releasing facilities. Lastly, the outer-boundary average represents the daily- and annual-averaged 443 concentrations at 1,000 m distance from releasing facilities. 444 2.1.1.3.1 Estimated Short-Term Formaldehyde Concentrations 445 EPA uses the downwind daily-averaged concentration (95th percentile) output values from the IIOAC 446 model to evaluate short-term exposures to formaldehyde in this ambient air assessment. These 447 downwind daily-averaged air concentrations are the basis for estimating acute exposures (acute 448 concentrations) as described in Section 2.1.1.4. 449 2.1.1.3.2 Estimated Long-Term Formaldehyde Concentrations 450 EPA uses the downwind annual-averaged air concentration (95th percentile) output values from the 451 IIOAC model to evaluate long-term exposure to formaldehyde in this ambient air assessment. These 452 downwind annual-averaged air concentrations are the basis for calculating chronic exposure 453 concentrations (average daily concentrations [ADC] and lifetime average daily concentrations [LADC]) 454 as described in Section 2.1.1.4. 455 2.1.1.4 Exposure Calculations 456 To compare the modeled air concentrations from the IIOAC Model, outputs to the human health hazard 457 data (U.S. EPA, 2024e) for purposes of deriving risk estimates, the modeled air concentration outputs 458 from IIOAC need to be adjusted to reflect the exposure evaluated. Because available formaldehyde 459 hazard data include acute, chronic, and cancer inhalation hazard values, EPA adjusted modeled air 460 concentrations from IIOAC for both acute exposure concentrations (AC) and chronic (non-cancer 461 [ADC] and cancer [LADC]) exposures as described in Sections 2.1.1.4.1 and 2.1.1.4.3. 462 2.1.1.4.1 Acute Concentrations (AC) 463 The AC is defined as the daily air concentration adjusted for exposure duration and averaging time. For 464 purposes of this ambient air exposure assessment, EPA uses the downwind daily-averaged air 465 concentration output from IIOAC and adjusts it for a 24-hour exposure duration and a 24-hour averaging 466 time to calculate the AC using Equation 2-1. 467 468 Equation 2-1. (𝐶 × 𝐸𝐷) 469 AC = 𝐴𝑇 470 Where: 471 AC = Acute concentration (µg/m3) 472 C = Daily-averaged air concentration (µg/m3) [IIOAC output] 473 ED = Exposure duration (24 hours) 474 AT = Averaging time (24 hours) 475 Page 14 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 476 The calculated AC is used for comparison with acute human hazard data3 to calculate acute non-cancer 477 risk estimates as described in the Draft Human Health Risk Assessment for Formaldehyde (U.S. EPA, 478 2024e). 479 2.1.1.4.2 Average Daily Concentration (ADC) 480 The ADC is defined as the mean amount of an agent to which an individual is exposed on a daily basis 481 often averaged over a definitive period of time. For purposes of this ambient air exposure assessment, 482 EPA uses the downwind annual-averaged air concentration output from IIOAC and adjusts it for 483 exposure time, exposure frequency, exposure duration, and averaging time associated with the exposure 484 scenario to calculate the ADC using Equation 2-2. 485 486 Equation 2-2. (𝐶 × 𝐸𝑇 × 𝐸𝐹 × 𝐸𝐷) 487 ADC = 𝐴𝑇 488 489 Where: 490 ADC = Average daily concentration (µg/m3) 491 C = Annual-averaged air concentration (µg/m3) 492 ET = Exposure time (24 h/day) 493 EF = Exposure frequency (365 days/yr) 494 ED = Exposure duration (1 yr) 495 AT = Averaging Time, 1 yr × 365 days/yr × 24 hr/day 496 497 The calculated ADC is used for comparison with chronic human hazard data to calculate chronic non- 498 cancer risk estimates as described in the Draft Human Health Risk Assessment for Formaldehyde (U.S. 499 EPA, 2024e). 500 2.1.1.4.3 Lifetime Daily Average Concentration (LADC) 501 The LADC is defined as the average daily concentration adjusted for an individuals expected residency 502 time and average lifetime. For purposes of this ambient air exposure assessment, EPA uses the 503 downwind annual-averaged air concentration output from IIOAC and adjusts it for exposure time, 504 exposure frequency, exposure duration, and averaging time associated with the exposure scenario to 505 calculate the LADC using Equation 2-3. 506 507 Equation 2-3. (𝐶 × 𝐸𝑇 × 𝐸𝐹 × 𝐸𝐷) 508 LADC = 𝐴𝑇 509 510 Where: 511 LADC = Lifetime average daily concentration (µg/m3) 512 C = Annual-averaged air concentration (µg/m3) 513 ET = Exposure time (24 h/day) 514 EF = Exposure frequency (365 days/yr) 515 ED = Exposure duration (78-year resident); number of years resident assumed to reside 516 in a residential location where exposure occurs 517 AT = Averaging Time (78-year × 365 days/yr × 24 hr/day); number of years a receptor 3 For context, while EPA uses the daily-averaged air concentration (24-hour average) to calculate the AC, the acute inhalation hazard value available for formaldehyde reflects a 15-minute peak exposure window. Page 15 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 518 is assumed to live (years) (from EPA’s Exposure Factors Handbook) (U.S. EPA, 519 2011, 7485096) 520 521 The calculated LADC is used for comparison with cancer human hazard data to calculate cancer risk 522 estimates as described in the Draft Human Health Risk Assessment for Formaldehyde (U.S. EPA, 523 2024e). The LADC is intended to represent the population within the community that experiences 524 exposure over a full 78-year lifetime. Since formaldehyde is ubiquitous in the ambient air, EPA assumes 525 a residency time of 78-years because even if a resident moves to another location, exposure to 526 formaldehyde is expected to continuously occur. 527 Human Exposure Model (HEM) 528 EPA used the Human Exposure Model (HEM 4.2) to estimate formaldehyde concentrations on a site- 529 specific basis at multiple distances from releasing facilities. HEM 4.2 has two components: (1) an 530 atmospheric dispersion model, AERMOD, with included regional meteorological data; and (2) U.S. 531 Census Bureau population data at the Census block level. The current HEM version utilizes 2020 532 Census data—including all 50 states, the District of Columbia, Puerto Rico, and the U.S. Virgin Islands. 533 AERMOD estimates the magnitude and distribution of chemicals concentrations in ambient air in the 534 vicinity of each releasing facility within a user-defined radial distances out to 50 km (about 30 miles). 535 HEM also provides chemical concentrations in ambient air at the centroid of over 8 million census 536 blocks across the United States. The model is also able to combine the estimated chemical’s 537 concentrations with hazard data to estimate cancer risks and noncancer hazards, and the population data 538 to inform cancer incidence, and other risk measures. HEM automatically utilizes regional meteorological 539 data for each release point, as well as local topographic information, to inform the release dispersion 540 model. Refer to the HEM v4.2 User Guide for more details about these and other capabilities. 541 EPA evaluated site-specific releases from 810 TRI facilities directly reporting to TRI with Form R using 542 HEM v4.2. EPA expects TRI emissions to be a sub-set of the larger NEI point source database and 543 incorporate the larger release sites associated with TSCA COUs. Facilities must meet TRI reporting 544 criteria for the number of full-time employees, specific NAICS codes and a chemical threshold of 545 manufacturing and processing (>25,000 lb) or otherwise using formaldehyde (>10,000 lb). A bulk run of 546 all facilities reporting air releases of formaldehyde to TRI was conducted to obtain aggregated location- 547 specific air concentrations at a national scale. Stack and fugitive releases were modeled as distinct 548 sources, each using a generic set of parameters (Table 2-2). 549 Table 2-2. HEM Input Parameters for Stack 550 and Fugitive Air Releases Stack Release Parameters Value Stack height (m) 10 Stack diameter (m) 2 Exit velocity (m/sec) 5 Exit temperature (°K) 300 Fugitive Release Parameters Value Length (m) 10 Width (m) 10 Page 16 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Stack Release Parameters Value Angle (degrees) 0 Release height (m) 3.05 551 552 The exposure scenario modeled with HEM is identical to the scenario modeled with IIOAC and 553 assumed each facility operates 24 hours/day, 7 days/week and 365 days/year. However, for the HEM 554 modeling, EPA utilized the site-specific maximum annual release reported to TRI from 2016 to 2021 for 555 modeling. 556 557 HEM was run in a configuration with 11 rings of receptors placed at varying radial distances from the 558 facility center: 10, 30, 60, and 100 m; 1, 2.5, 5, 10, 15, 25, and 50 km. Each ring is made up of 16 evenly 559 spaced modeled exposure points. The HEM results were applied to consider concentrations at discrete 560 distances, compare the impact of fugitive and stack releases at discrete distances, and to compare with 561 IIOAC results. HEM calculates resulting annual average concentrations at each modeled exposure point 562 among the rings, and then processes the results to aggregate concentrations at a Census block scale. The 563 resulting output data is both tabular and geographic, capable of being mapped to physical locations. The 564 model results associated with Census geometry can then be compared to populations and demographics 565 associated with ranges of modeled ambient air conditions. 566 567 An additional analysis was conducted to evaluate the locations and populations where modeled facility 568 releases resulted in concentrations above the 95th percentile biogenic formaldehyde concentrations in 569 AirToxScreen modeling data (see Section 3.2). HEM calculates an aggregated risk value, called the 570 maximum individual risk (MIR) for each Census block within the model domain. This risk value is 571 calculated by multiplying the aggregate Census block concentration by the inhalation unit risk (IUR). 572 An IUR of 6.4×10−6 m³/µg was applied for formaldehyde in this modeling approach. Because this 573 aggregated result across Census blocks is only presented as a risk quotient, the bulk MIR output by 574 Census block was converted to concentrations by dividing by the IUR. The resulting aggregate mean 575 annual concentrations were then mapped to visualize the spatial distribution of modeled concentrations. 576 These aggregated concentrations are the summed stack and fugitive release concentrations, which can 577 include the summation of multiple adjacent facilities, at specific locations. The site-specific 578 concentration results represent the expected annual average ambient air concentration attributable from 579 all modeled TRI releases of TSCA COUs, in some census blocks accounting for concentrations from 580 multiple releasing facilities. The 2020 Census block population estimates included in the HEM Census 581 database for modeled blocks were applied to evaluate the magnitude of the exposed population to 582 various levels of concentrations. 583 Air Toxics Screening Assessment Tool (AirToxScreen) 584 AirToxScreen uses the chemical transport model (CMAQ) and the dispersion model (AERMOD) to 585 estimate average annual outdoor ambient air concentrations across the United States. EPA used data 586 from the 2019 AirToxScreen to contextualize the relative relationship of formaldehyde concentrations in 587 ambient air resulting from all known source of hazardous air pollutants in ambient air.The tool uses 588 release data from the NEI database. These AirToxScreen data allow EPA to differentiate among 589 modeled emissions from various source categories (e.g., point sources, biogenic sources, and secondary 590 formation. 591 592 In this draft assessment, EPA used data from AirToxScreen to estimate a 95th percentile concentration 593 of census tract concentrations of formaldehyde from all modeled biogenic sources. This estimate Page 17 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 594 captures concentrations that are reasonably expected to occur without human contributions. EPA used 595 this estimate for comparison to concentrations from other formaldehyde sources, including those that are 596 expected from formaldehyde TSCA COUs. While this value is a percentile derived from the entirety of 597 AirToxScreen data, it should be noted there may be locations where biogenic sources are not prevalent. 598 As such, in some locations which may be highly industrialized (with industrial facilities releasing 599 formaldehyde into the ambient air), this value may overestimate the true contribution of biogenic 600 sources to the ambient concentrations of formaldehyde. 601 2.2 Monitoring 602 EPA identified and summarized monitoring data for formaldehyde from EPA’s AMTIC (U.S. EPA, 603 2022a). EPA also identified and summarized monitoring data that were included in EPA’s systematic 604 review process (U.S. EPA, 2023). 605 Ambient Monitoring Technology Information Center (AMTIC) 606 AMTIC oversees ambient air quality monitoring data collected for high priority air pollutants overseen 607 by the EPA Ambient Air Monitoring Group (AAMG). AMTIC maintains a database of submitted air 608 quality monitoring data that meet the collection and quality assurance criteria set by AMTIC. 609 610 Ambient air concentration data was pulled in July of 2023 capturing monitoring data from 2015 to 2020. 611 The formaldehyde AMTIC monitoring data used comprised 234,000 entries by 20 monitoring programs 612 covering 36 states and 187 census tracts. Dates of sample collection ranged from January 2015 through 613 December 2020. Samples were collected using the Fluxsense sampling system (83% of samples), DNPH 614 silica cartridge (17%), or by pressurized canister (<1%). Samples collected using Flux sense had a 5- 615 minute collection period while DNPH silica included 3-, 6-, 8-, and 24-hour collection durations. The 616 resulting formaldehyde concentrations were converted to µg/m3. Samples with concentrations below the 617 reported limit of detection were filtered out (15 percent of samples). The remaining concentrations 618 ranged from 0.0015 to 60.1 µg/m3 with a 95th percentile ranging from 0.382 to 6.2 µg/m3 and median 619 concentration of 1.88 µg/m3. 620 621 EPA used the AMTIC monitoring data for formaldehyde to assess the geographic distribution of 622 formaldehyde in ambient air across the United States. These data allowed the EPA to identify the 623 influence of specific facilities reporting the emissions of formaldehyde falling under the TSCA 624 conditions of use on formaldehyde concentrations in air. The data were also used to identify regions 625 with elevated ambient formaldehyde concentrations for more in-depth investigations. These data were 626 also used to compare the results of the AirToxScreen and IIOAC models to field monitoring data for 627 model validation and quality assurance. Page 18 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 628 3 RESULTS 629 630 3.1 Modeling 631 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) 632 As noted in Section 2.1.1.2, EPA modeled a suite of exposure scenarios with IIOAC including releases 633 from two separate datasets (TRI and NEI). These results are provided in the “Draft Ambient Air 634 Exposure Assessment Results and Risk Calcs Supplement A” to this ambient air exposure assessment. 635 636 As described in Section 2.1.1.3, for this ambient air exposure assessment, EPA utilized the 95th 637 percentile of annual release by industry sector reported to TRI, and the downwind modeled daily- 638 averaged and annual-averaged air concentrations from the IIOAC output file at an area distance of 100 639 to 1,000 m from the releasing facility, to calculate exposure concentrations (see Section 2.1.1.4) for each 640 industry sector. The industry sector results are then mapped to associated TSCA COUs for purposes of 641 presenting the data in this ambient air exposure assessment. Use of these TRI-based results represent 642 exposures resulting from actual reported releases tied to TSCA COUs. 643 644 Results for acute and chronic exposure concentrations across all industry sectors and associated COUs 645 ranged from 0.0001 to 5.7 µg/m3 for the exposure scenario described above. Results are presented for 646 each TSCA COU in Figure 3-1; the distribution is shown in Figure 3-2. These results represent the 647 highest exposure concentration across each industry sector associated with the respective formaldehyde 648 TSCA COU. A Table summarizing all industry sectors, the associated mapping to TSCA COU, and the 649 modeled exposure concentration for the scenario presented and summarized in this ambient air exposure 650 assessment is provided in Appendix C. Page 19 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 651 652 Figure 3-1. Exposure Concentrations by TSCA COU for the 95th Percentile Release Scenario and 653 95th Percentile Modeled Concentration between 100 and 1,000 m from Industrial Facilities 654 Releasing Formaldehyde to the Ambient Air 655 656 657 Figure 3-2. Distribution of Exposure Concentrations Modeled by IIOAC 658 659 A total of 810 TRI reporting facilities reported releases of formaldehyde to the ambient air. These 660 facilities represented 35 industry sectors, which were evaluated in this ambient air exposure assessment 661 and are associated with 29 formaldehyde TSCA COUs, as shown in the Figure 3-1. Across all 29 TSCA 662 COUs evaluated, 19 (65.5 percent) have calculated exposure concentrations greater than 1 µg/m3; 7 Page 20 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 663 (24.1 percent) have calculated exposure concentrations between 0.1 and 1 µg/m3; 2 (6.9 percent) have 664 calculated exposure concentrations between 0.01 and 0.1 µg/m3; and 1 has a calculated exposure 665 concentration of 0.0001 µg/m3. 666 667 Recognizing the ubiquity of formaldehyde in ambient air occurs from multiple sources including non- 668 TSCA sources like biogenic/natural sources and secondary formation, EPA compared modeled 669 concentrations from IIOAC to the 95th percentile concentration of 0.28 µg/m3 attributable to biogenic 670 sources (from the 2019 AirToxScreen data described in Sections 3.1.2). Across all 29 TSCA COUs 671 evaluated, 21 TSCA COUs have IIOAC modeled concentrations (ranging from 0.3 to 5.75 µg/m3) 672 greater than the 95th percentile concentration attributable to biogenic sources. Twenty TSCA COUs 673 have IIOAC modeled concentrations (ranging from 0.7 to 5.75 µg/m3) greater than twenty percent of the 674 95th percentile concentration attributable to biogenic sources (0.34 µg/m3). Eighteen TSCA COUs have 675 IIOAC modeled concentrations (ranging from 1.65 to 5.75 µg/m3) greater than five times the 95th 676 percentile concentration attributable to biogenic sources (1.4 µg/m3). Seven TSCA COUs have IIOAC 677 modeled concentrations (ranging from 3.4 to 5.75 µg/m3) greater than ten times the 95th percentile 678 concentration attributable to biogenic sources (2.8 µg/m3). Eight TSCA COUs have IIOAC modeled 679 concentrations less than the 95th percentile concentration attributable to biogenic sources. 680 681 The following three industry sectors had the highest calculated exposure concentrations relative to the 682 other calculated exposure concentrations evaluated: 683 • Non-metallic Mineral Product Manufacturing (5.7 µg/m3); 684 • Textiles, Apparel, and Leather Product Manufacturing (4.5 µg/m3); and 685 • Transportation Equipment Manufacturing (3.4 µg/m3). 686 Together, these three industry sectors are associated with seven formaldehyde TSCA COUs: 687 • Processing – incorporation into an article-adhesives and sealant chemicals (5.7 µg/m3); 688 • Processing as a reactant – intermediate (5.7 µg/m3); 689 • Processing – incorporation into a formulation, mixture, or reaction product-intermediate (5.7 690 µg/m3); 691 • Processing – incorporation into article-finishing agent (4.5 µg/m3); 692 • Processing – incorporation into a formulation, mixture, or reaction product-bleaching agents (4.5 693 µg/m3); 694 • Processing – incorporation into an article-paint additives and coating additives (3.4 µg/m3); and 695 • Industrial use – chemical substances in industrial products-paints and coatings; adhesives and 696 sealants, lubricants (3.4 µg/m3). 697 Although a single industry sector can represent multiple formaldehyde TSCA COUs (as shown above 698 where Non-metallic Mineral Product Manufacturing is associated with the first three TSCA COUs), it is 699 also possible that multiple industry sectors are associated with an individual formaldehyde TSCA COU. 700 Therefore, in addition to the maximum values plotted in Figure 3-1, there may be additional exposure 701 estimates for each formaldehyde TSCA COU that better represent individual industry processes. For 702 example, the exposure concentration for the TSCA COU “Industrial use – non-incorporative activities – 703 oxidizing/reducing agent; processing aids, not otherwise listed” is plotted as 0.20 µg/m3; however, the 704 exposure concentrations for this COU ranged from 0.09 to 0.2 µg/m3 across four industry sectors as 705 listed below: 706 • Organic Fiber Manufacturing (0.2 µg/m3); 707 • Computer and Electronic Part Manufacturing (0.09 µg/m3); 708 • Fabricated Metal Product Manufacturing (0.19 µg/m3); and Page 21 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 709 • Primary Metal Manufacturing (0.12 µg/m3). 710 A table of results for all the industry sectors and the associated industrial TSCA COUs is provided in 711 Appendix C for reference. Some TSCA industrial COUs were associated with industry sectors not 712 captured within the TRI database, EPA used ambient air results from the NEI data in those cases. These 713 results are provided in the “Draft Ambient Air Exposure Assessment Results and Risk Calcs Supplement 714 A.” 715 AirToxScreen 716 EPA used 2019 AirToxScreen to understand the relative formaldehyde concentration contributions of 717 non-TSCA sources compared to TSCA-COUs. These data are shown in Figure 3-3, which presents the 718 range of concentrations across all sources of formaldehyde, as well as contributions from biogenic 719 sources, secondary sources, and point sources. Results from the 2019 AirToxScreen national-scale 720 analysis are attributable to 38 different emission source categories (including point source, secondary 721 production, biogenic, on/off-road, etc.). 722 723 724 Figure 3-3. 2019 AirToxScreen Results 725 726 Secondary production of formaldehyde is the largest contributor of formaldehyde to ambient air 727 concentrations ranging from 0.085 to 1.8 µg/m3 (mean ± 1 SD: 0.86 ± 0.25 µg/m3). Secondary 728 production is the atmospheric formation of formaldehyde from naturally and manmade compounds. This Page 22 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 729 can include the degradation of isoprene (a compound naturally produced by animals and plants) to 730 formaldehyde. AirToxScreen is not able to source apportion relative contributions from different 731 secondary sources. Biogenic sources also significantly contribute to total concentration with a range of 732 0.0014 to 0.67 µg/m3 (mean ± 1 SD: 0.13 ± 0.072 µg/m3) and a 95th percentile calculated concentration 733 of 0.28 µg/m3. Biogenic sources are those emissions from trees, plants, and soil microbes. Notably, these 734 data do not show any TSCA-specific data but do show relative distributions of various sources; 735 however, the point sources estimates are expected to include contributions from TSCA COUs. Point 736 sources contributions to total formaldehyde concentrations range from 0.0 to 0.88 µg/m3 (mean ± 1SD: 737 0.0070 ± 0.014 µg/m3). Although Figure 3-3 does not include AirToxScreen data for on-road sources, 738 near-road sources, off-road sources, wildfire sources, etc., these sources would be captured in the results 739 shown for all sources. 740 Human Exposure Model (HEM) 741 Concentrations resulting from TRI facility releases modeled by HEM were aggregated and summarized 742 at the Census block level, allowing visualization of the geographic distribution of results (Figure 3-4). 743 Resulting concentrations ranged from 0 to 8.9 µg/m³, with the greatest concentrations nearby industrial 744 facilities. Census blocks with modeled total concentrations below a concentration of 0.28 µg/m³ 745 associated with biogenic/natural sources of formaldehyde are presented in gray. Blue dots show Census 746 blocks with concentrations ranging from 1 to 5 times 0.28 µg/m³, purple dots show concentrations from 747 5 to 10 times 0.28 µg/m³, and pink dots show values greater than 10 times 0.28 µg/m³. Across the 748 nation, a total population of 105,463 people (based on 2020 Census data) live in the Census blocks 749 shown with ambient are concentrations above 0.28 µg/m³. 750 Page 23 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 751 Figure 3-4. Map of Contiguous United States with HEM Model Results for TRI Releases 752 Aggregated and Summarized by Census Block 753 754 Elevated ambient air concentrations of formaldehyde from industrial releases appear most densely 755 concentrated in the southeastern United States. Census blocks with elevated concentrations are found 756 throughout the nation, with some regions showing fewer overall TRI facilities, and fewer releases 757 resulting in elevated air concentrations. 758 759 Patterns in the relative contribution of stack and fugitive releases, and the distribution of results at 760 varying radial distances from the releasing facility were examined (Figure 3-5). The concentration 761 results across all facilities and COUs were pooled for this analysis to visualize general trends across all 762 TRI facilities reporting formaldehyde releases. Each vertical bar and maximum line indicate the shape of 763 the distribution of concentrations by release type for individual facilities. These results indicate that 764 concentrations resulting from fugitive emissions are greater than those from stack emissions closer to the 765 releasing facility, but concentrations from stack emissions tend to become greater at further distances. 766 As many facilities report only a single release type (either fugitive or stack), the total concentration 767 distributions represent a greater number of facilities than the corresponding fugitive and stack 768 distributions and the median values tend to fall somewhere between the fugitive and stack values. Total 769 modeled concentrations tend to reach their maximum within 1,000 m of a facility. Values represented in 770 this analysis are directly modeled at the radial receptor points, rather than Census block centroids, and 771 can therefore be located much closer to the releasing facility and represent much higher concentrations. 772 These points are not associated with population estimates, and in some cases the modeled distances may 773 still be within a facility property boundary. 774 Page 24 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 775 776 Figure 3-5. Median and Maximum Downwind Concentrations (Fugitive, Stack, and Total 777 Emissions) across the 11 Discrete Distance Rings Modeled in HEM 778 3.2 Monitoring 779 Ambient Monitoring Technology Information Center (AMTIC) 780 EPA considered a total of 306,529 samples from the AMTIC database (U.S. EPA, 2024a). 781 The Agency computed summary statistics for all samples, as well as samples by state, census tract, 782 monitoring site, monitoring site and year, and monitoring site and year and quarter. Sample collection 783 durations ranged from 5 minutes to 24 hours using one of five approved collection methods. No data 784 was omitted based on collection duration or method. Entries with concentrations reported below the self- 785 reported limit of detection or contained invalid concentration data (i.e., NULL, NA) were omitted from 786 the final data set. Formaldehyde concentrations were converted to µg/m3 for consistency across sample 787 analysis methods but were not otherwise normalized by sample collection duration or methodology. 788 EPA used the overall statistics across all samples to characterize exposures, derive risk estimates, and 789 characterize risks to the general population (see Table 3-1). Histograms and summary statistics of annual 790 data are shown in Figure 3-6. 791 Page 25 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 792 Table 3-1. Overall Monitored Method Detection Limits (MDL) of Formaldehyde from AMTIC 793 Dataset (2015 to 2020) Monitored Concentration Statistics (µg/m3) Entry Non-zero Standard Group Minimum Median Mean Maximum Count Minimum Deviation Grouped by: all samples All 233,961 0 0.00012 1.6 2.1 2.2 60 Grouped by: collection duration description 12 Hours 340 0.50 0.50 3.6 3.8 1.7 9.0 24 Hours 39,288 0 0.0015 2.3 2.8 2.1 60 3 Hours 5,870 0 0.0083 3.7 4.4 3.3 45 5 Minutes 184,307 0 0.00012 1.3 1.8 2.0 49 6 Hours 1 3.4 3.4 3.4 3.4 – 3.4 8 Hours 4,155 0.0055 0.0055 3.6 4.1 2.8 24 Grouped by: collection method description 6-L Pressurized 67 3.5 3.5 11 14 7.9 42 Canister Cartridge Dnph On 6,671 0 0.020 2.3 2.7 1.8 46 Silica, Heated O3 Denuder Cartridge-Dnph-On- 10,115 0 0.024 3.1 3.7 2.6 60 Silica Fluxsense 184,307 0 0.00012 1.3 1.8 2.0 49 Silica-Dnph-Cart-Ki 32,801 0 0.0015 2.5 3.0 2.3 45 O3 Scrub 794 Page 26 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 795 796 Figure 3-6. Histograms of Ambient Air Concentrations (µg/m3) of Formaldehyde across 797 Contiguous United States from 2015 to 2020 Page 27 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 798 799 From the overall AMTIC dataset, samples were collected from June 01, 2015, through December 31, 800 2020. Within this dataset, EPA found 24 percent of entries lacked standardized concentration data. EPA 801 also found 15 percent of samples fall below the standard method detection limit (MDL) with a mean 802 standardized formaldehyde concentration of 2.1 ± 2.2 μg/m3. The overall monitoring dataset had 803 concentrations ranging from 0 to 60 μg/m3 with a median value of 1.6 μg/m3. Figure 3-7 shows the 804 location and relative concentration of formaldehyde at each formaldehyde monitoring site. 805 806 807 Figure 3-7. Map of Monitoring Sites for Formaldehyde across the Contiguous United States 808 3.3 Data Integration of Various Sources of Formaldehyde 809 Monitoring data from AMTIC, modeled exposures calculated from IIOAC by COU, and modeled data 810 from AirToxScreen were compiled to understand how TSCA COUs fit into the broader context of 811 available information on formaldehyde. Figure 3-8 shows the distributions of data from these datasets. 812 This image shows that the calculated concentrations of formaldehyde resulting from formaldehyde 813 TSCA COUs falls within the range of concentrations that are expected to occur from naturally occuring 814 sources including biogenic and secondary production of formaldehyde. Furthermore, modeled exposures 815 calculated from IIOAC fall within the range of measured concentration values. 816 Page 28 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 817 818 Figure 3-8. Distributions of AMTIC Monitoring Data, IIOAC Modeled Data, and AirToxScreen 819 Modeled Data 820 821 EPA recognizes that the different model estimates are not directly comparable. For example, the IIOAC 822 results represent a downwind annual average concentration between 100 to 1,000 m from the release 823 point. In contrast, AirToxScreen concentrations represent annual average concentrations averaged across 824 census tracts. Only point source data may include concentrations resulting from formaldehyde TSCA 825 COUs, but again AirToxScreen point source data is averaged across census tracts and not directly 826 comparable to the IIOAC results. Given the spatial scale difference, it is expected that AirToxScreen 827 results could underestimate concentrations on a smaller scale (i.e., near facilities) or have lower 828 concentration estimates than IIOAC because the IIOAC estimates represent distances closer to the 829 release point. None-the-less, considering these multiple data sources, even across mixed temporal data, 830 can help contextualize contributions of formaldehyde concentrations from TSCA COUs relative to other 831 non-TSCA sources. 832 833 Furthermore, the AMTIC data represent a range of formaldehyde samples collected at various locations 834 and represent a total exposure from all formaldehyde sources and cannot be attributed to TSCA COUs. 835 Additionally, collection durations for AMTIC are much shorter than a year (5 minutes to 24 hours). 836 Similarly, sampling method may significantly affect concentration (see Figure 3-9). Samples collected 837 using 6-liter pressurized canisters tend to be higher than all other sampling methods. Page 29 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 838 839 Figure 3-9. Ambient Air Formaldehyde Concentrations by Sample Duration and Sampling 840 Technique 841 842 Taken together, these data and results show modeled formaldehyde concentrations using IIOAC are 843 generally in the same range as monitored formaldehyde concentrations in ambient air as well as 844 contributions of formaldehyde to the ambient air from biogenic and secondary production. However, to 845 characterize exposures associated with TSCA COUs it is necessary to recognize modeled concentrations 846 may contribute to some part of the total concentrations captured by monitored data but are independent 847 and exclusive from non-TSCA sources. Therefore, while modeled concentrations are in the same range 848 as biogenic and secondary formation, the exposures resulting from TSCA COUs are representative of 849 actual exposures from TSCA COUs alone, and not a subset of the biogenic or secondary formation data. 850 To contextualize this, actual exposures occurring at a given location from both TSCA COUs and 851 biogenic or secondary formation would be an additive exposure receiving the full contribution of 852 formaldehyde from TSCA COUs as well as the full contribution of formaldehyde from biogenic sources. 853 Page 30 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 854 4 STRENGTHS AND LIMITATIONS, ASSUMPTIONS, 855 UNCERTAINTY, AND CONFIDENCE STATEMENT 856 4.1 Integrated Indoor/Outdoor Air Calculator Model (IIOAC) 857 The approach and methodology presented in this ambient air exposure assessment replicates previously 858 peer reviewed approaches and methods, and incorporates several additional components recommended 859 by peer reviewers to provide a more comprehensive exposure assessment. As such, EPA has high 860 confidence in the IIOAC modeling and use of the results to characterize exposures and derive risk 861 estimates for individuals living nearby releasing facilities. 862 863 Strengths of this ambient air exposure assessment include use of environmental release data from 864 multiple databases across multiple years which are reported by industry, as required by statute, and 865 undergoes repeatable quality assurance quality control reviews (U.S. EPA, 2024d). These release data 866 are used as direct inputs to EPA’s peer reviewed IIOAC to estimate concentrations at several distances 867 from releasing facilities where individuals reside for many years. Additionally, all reported releases 868 within each database are categorized by industry sector based on NAICS codes reported to the 869 respective database allowing for a more direct association of exposures from industrial releases to TSCA 870 COUs. 871 872 Use of additional peer-reviewed models (AirToxScreen and HEM), along with monitoring data 873 (AMTIC) to contextualize ambient air concentrations of formaldehyde, present a consistent picture of 874 exposures when compared to IIOAC results. They also provide added strength and confidence to the 875 approaches and methods used in this draft ambient air exposure assessment. 876 877 Limitations of the approaches and methods used for modeling are generally associated with overall 878 limitations of IIOAC. For example, IIOAC modeling is based on pre-run scenarios within AERMOD. 879 As such, default input parameters for IIOAC are confined to those input parameters utilized for those 880 pre-run AERMOD scenarios and cannot be changed, including default stack parameters, 2011 to 2015 881 meteorological data, and the lack of site-specific information like building dimensions, stack heights, 882 elevation, and land use. 883 884 However, to alleviate concerns around these limitations and impacts on the overall confidence in the 885 IIOAC modeled results, EPA conducted a series of sensitivity analyses to address some of the concerns. 886 These sensitivity analyses are described in Appendix B.1 and provide a brief discussion of EPA’s 887 findings. Generally, the Agency found that although the limitations identified above have some impact 888 on overall modeled results, the impact is not substantial in terms of overall modeled concentrations and 889 do not change the overall characterization of exposures from industrial releases. Additionally, EPA 890 found that while exposure scenarios and default stack parameters may represent a more conservative 891 estimate of exposures to individuals living near releasing facilities, the findings are not overly 892 conservative and generally represent a more health protective, but still realistic exposure. 893 4.2 AirToxScreen 894 AirToxScreen has been previously reviewed by EPA’s Science Advisory Board (SAB). As, such EPA 895 has high confidence in the modeled data. Similarly, these data are based on the NEI, which has been 896 rated as a high-quality data source according to the Draft Systematic Review Protocol Supporting TSCA 897 Risk Evaluations for Chemical Substances (U.S. EPA, 2021b). 898 Page 31 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 899 The strengths of these data are that they show the various potential sources of formaldehyde in the 900 conterminous United States. The limitations of these data are that they cannot be directly compared to 901 TSCA COUs as they are not representative of facility-scale releases and subsequent ambient air 902 concentrations. A key assumption of AirToxScreen is that it cannot provide a precise exposure and risk 903 for a specific person. Instead, these results are best applied to understand differences in potential sources 904 of formaldehyde. Thus, using summary statistics is an appropriate application. 905 4.3 Human Exposure Model (HEM) 906 The HEM 4.2 dispersion calculations are handled by EPA’s AERMOD, which has undergone review 907 and evaluation as part of the regulatory models process. A description of its promulgation as a 908 regulatory model is included in “Appendix W” to 40 CFR Part 51. Due to its regular application in 909 assessments to inform regulatory decisions, EPA has high confidence in the modeling results. A 910 limitation of the model is the exclusion of photodegradation processes, which may be relevant to 911 ambient air concentrations of formaldehyde. Without modeled decreases in concentration due to 912 photodegradation, the results presented in this document are more conservative (protective). 913 4.4 Ambient Monitoring Technology Information Center (AMTIC) 914 Dataset 915 AMTIC data has been previously reviewed and verified by the AMTICs Ambient Air Monitoring 916 Group, which has taken various quality assurance steps to ensure data quality and has been certified in 917 accordance to 40 CFR Part 58.15. EPA has high confidence in the AMTIC ambient air data set (U.S. 918 EPA, 2022a), which received a high-quality rating from EPA’s systematic review process. (U.S. EPA, 919 2021a). 920 921 The primary limitations of the AMTIC data are as presented in this ambient air exposure assessment, the 922 data has not been annualized and therefore represents a diverse collection of sampling durations (none of 923 which are annual averages) that are not directly comparable to either IIOAC or AirToxScreen data. 924 Additionally, because monitored data represents a total aggregate concentration from all sources of 925 formaldehyde contributing to ambient air concentrations, the AMTIC data cannot be associated with 926 TSCA COUs for purposes of characterizing exposures from TSCA COUS. 927 4.5 Weight of Scientific Evidence 928 EPA has high confidence in the overall characterization of exposures for this ambient air exposure 929 assessment as it relies upon direct reported releases from databases that received a high-quality rating 930 from EPA’s systematic review process and peer-reviewed models to derive exposure concentrations at 931 distances from releasing facilities where individuals reside for many years. Use of additional peer- 932 reviewed models (AirToxScreen and HEM) along with monitoring data (AMTIC) to further 933 contextualize ambient air concentrations of formaldehyde, which also present a consistent picture of 934 exposures when compared to IIOAC results, provide added strength and confidence to the approaches 935 and methods used in this draft ambient air exposure assessment. This confidence statement is consistent 936 with the Draft Systematic Review Protocol Supporting TSCA Risk Evaluations for Chemical Substances 937 (U.S. EPA, 2021b). 938 939 The use of release data from multiple databases across multiple years of data provide a more 940 comprehensive ambient air exposure assessment and ensure higher release years are not missed. 941 Furthermore, use of actual reported releases minimizes uncertainties around estimated releases using 942 theoretical distributions and provides added confidence that modeled concentrations and exposures are 943 real and not hypothetical. Page 32 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 944 Additionally, while the approaches and methods used in this ambient air exposure assessment use more 945 conservative assumptions and default model inputs may be viewed as a limitation and uncertainty, 946 sensitivity analyses around several of these inputs found that while they have some impact on overall 947 modeled results, the impact is (1) not substantial in terms of overall modeled concentrations, and (2) 948 does not change the overall characterization of exposures from industrial releases. Additionally, EPA 949 found that although exposure scenarios and default stack parameters may represent a more conservative 950 estimate of exposures to individuals living near releasing facilities, the findings are not overly 951 conservative and generally represent a more health protective, yet still realistic exposure. 952 Page 33 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 953 5 REFERENCES 954 Luecken, DJ; Napelenok, SL; Strum, M; Scheffe, R; Phillips, S. (2018). Sensitivity of ambient 955 atmospheric formaldehyde and ozone to precursor species and source types across the United 956 States. Environ Sci Technol 52: 4668-4675. http://dx.doi.org/10.1021/acs.est.7b05509 957 U.S. EPA. (2019a). National Emissions Inventory (NEI) [database]: CASRNs 79-00-5, 75-34-3, 107-06- 958 2, 78-87-5, 84-61-7, 106-99-0, 106-93-4, 50-00-0, 85-44-9, 106-46-7, 85-68-7, 84-74-2, and 959 115-86-6 [Database]. Washington, DC. Retrieved from https://www.epa.gov/air-emissions- 960 inventories/national-emissions-inventory-nei 961 U.S. EPA. (2019b). User's Guide: Integrated Indoor-Outdoor Air Calculator (IIOAC). Washington, DC: 962 U.S. EPA. 963 U.S. EPA. (2021a). Best practices for review and validation of ambient air monitoring data. (EPA- 964 454/B-21-007). 965 U.S. EPA. (2021b). Draft systematic review protocol supporting TSCA risk evaluations for chemical 966 substances, Version 1.0: A generic TSCA systematic review protocol with chemical-specific 967 methodologies. (EPA Document #EPA-D-20-031). Washington, DC: Office of Chemical Safety 968 and Pollution Prevention. https://www.regulations.gov/document/EPA-HQ-OPPT-2021-0414- 969 0005 970 U.S. EPA. (2022a). Ambient Monitoring Technology Information Center (AMTIC) - Ambient 971 Monitoring Archive for HAPs [Database]. Washington, DC. Retrieved from 972 https://www.epa.gov/amtic/amtic-ambient-monitoring-archive-haps 973 U.S. EPA. (2022b). National Analysis TRI dataset (TRI): Formaldehyde (CAS RN: 50-00-0), reporting 974 years 2016-2021 [Database]. Washington, DC. Retrieved from https://www.epa.gov/toxics- 975 release-inventory-tri-program/tri-listed-chemicals 976 U.S. EPA. (2023). Draft Risk Evaluation for Formaldehyde – Systematic Review Protocol. Washington, 977 DC: Office of Pollution Prevention and Toxics, Office of Chemical Safety and Pollution 978 Prevention. 979 U.S. EPA. (2024a). Draft Ambient Air Exposure Assessment for the Formaldehyde Risk Evaluation. 980 Washington, DC: U.S. Environmental Protection Agency, Office of Pollution Prevention and 981 Toxics. 982 U.S. EPA. (2024b). Draft Chemistry, Fate, and Transport Assessment for Formaldehyde. Washington, 983 DC: U.S. Environmental Protection Agency, Office of Pollution Prevention and Toxics. 984 U.S. EPA. (2024c). Draft Conditions of Use for the Formaldehyde Risk Evaluation. Washington, DC: 985 U.S. Environmental Protection Agency, Office of Pollution Prevention and Toxics. 986 U.S. EPA. (2024d). Draft Environmental Release Assessment for Formaldehyde. Washington, DC: U.S. 987 Environmental Protection Agency, Office of Pollution Prevention and Toxics. 988 U.S. EPA. (2024e). Draft Human Health Risk Assessment Characterization of Formaldehyde. 989 Washington, DC: U.S. Environmental Protection Agency, Office of Pollution Prevention and 990 Toxics. 991 Page 34 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 992 APPENDICES 993 994 Appendix A LIST OF SUPPLEMENTAL DOCUMENTS 995 Draft IIOAC Assessment Results and Risk Calcs for Formaldehyde Supplement A 996 Draft IIOAC Assessment Results and Risk Calcs for Formaldehyde Supplement B Page 35 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 997 Appendix B DATA SOURCES AND INFORMATION 998 999 B.1 Sensitivity Analyses Conducted to Inform Modeling for Formaldehyde 1000 EPA conducted a series of model sensitivity analyses to identify some key input parameters to be 1001 considered for this ambient air exposure assessment along with impact of select parameters on the 1002 overall modeling results. 1003 1004 Compare IIOAC to HEM 1005 Both IIOAC and HEM rely upon EPA’s AERMOD as the base model from which estimated ambient air 1006 concentrations are derived. Although both IIOAC and HEM use the same underlying model, slight 1007 differences in inputs, capabilities, and outputs warrant a sensitivity analysis to determine overall 1008 comparability of the modeled results. EPA conducted a sensitivity analysis using both models and 1009 identical input and exposure scenarios and found estimated concentrations, associated exposures, and 1010 associated risks were generally well within a magnitude of each other across multiple chemicals. 1011 1012 Based on these findings, EPA has high confidence in the modeled exposure concentrations from each 1013 model. As such, EPA uses both models in this general population risk assessment to inform exposures 1014 and take advantage of certain model capabilities to better characterize exposures for TSCA COUs, 1015 varying inputs, and other fit-for-purpose needs for this general population risk assessment. IIOAC is 1016 used for the screening and national level analyses as it is easier to use and faster to run. HEM has added 1017 flexibility to consider more than three pre-defined distances, additional meteorological stations, and 1018 other factors so HEM is used for the site-specific analysis to both target local population impacts as well 1019 as estimate concentrations at 11 finite distances away from each releasing facility, including 100 and 1020 1,000 m that can be compared to outputs from IIOAC. HEM is not readily set up to consider area 1021 distances, so the area distance between 100 and 1,000 m is not evaluated with HEM. EPA takes 1022 advantage of HEM’s flexibilities allowing user defined inputs to characterize findings for sensitivity 1023 analyses related to impact on modeled concentrations from different stack heights and different 1024 distances. 1025 1026 Identifying HE and CT Met Stations in IIOAC 1027 IIOAC includes 14 pre-defined climate regions (each with a surface station and upper-air station). Since 1028 release data used for the screening and national level analyses are not location-specific, EPA conducted 1029 a sensitivity analysis to identify 2 of the 14 climate regions within IIOAC which represent a central 1030 tendency and high-end climate region. This analysis looked at the average concentration and deposition 1031 predictions from each of the 14 climate regions under a set of identical release and exposure scenarios 1032 using 5 years of meteorological data (2011 to 2015) for all source types. EPA then ranked the modeled 1033 results from largest to smallest and found the highest air concentration estimate (considered high-end for 1034 this sensitivity analysis) occurred with the South (Coastal) climate region and refers to the Lake Charles, 1035 Louisiana, surface station within IIOAC. The 6th highest air concentration estimate (considered central 1036 tendency for this sensitivity analysis) occurred with the West North Central climate region and refers to 1037 the Sioux Falls, South Dakota, surface station within IIOAC. 1038 1039 Identifying HE Exposure Scenario in IIOAC 1040 IIOAC is capable of modeling a variety of release types, topography, meteorological conditions, and 1041 release scenarios. Since release data used for the screening and national level analyses are not location- 1042 specific, EPA previously developed and conducted a sensitivity analysis using IIOAC across multiple 1043 chemicals to evaluate a series of exposure scenarios presented in Figure_Apx B-1. The goal of this 1044 sensitivity analysis was to identify which exposure scenario, of those evaluated, tended to result in Page 36 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1045 higher air concentration estimates relative to the other scenarios across multiple chemicals. The results 1046 of this sensitivity analysis found the scenario highlighted in orange in Figure_Apx B-1 tended to result 1047 in the higher concentration estimates relative to the other scenarios evaluated. 1048 1049 1050 Figure_Apx B-1. Sensitivity Analysis Conceptual Model for Exposure Scenarios Modeled for Max 1051 and Mean Release Using IIOAC Model 1052 1053 Impact of Different Years of Meteorologic Data 1054 IIOAC considers 5 years of meteorological data (2011 to 2015). EPA previously received comment 1055 around this being older data and recommendations to consider more recent years of meteorological data 1056 in our ambient air exposure assessments. To alleviate concerns about the use of older meteorological 1057 data, EPA conducted a sensitivity analysis across different years of meteorologic data within AERMOD 1058 to see what the impacts on the estimated concentrations are. Because AERMOD is the base model 1059 within which pre-run scenarios were run to develop IIOAC, any findings from this sensitivity analysis in 1060 AERMOD would extend to IIOAC. The results from this sensitivity analysis within found that, although 1061 different years of meteorological data may result in small differences in estimated concentrations, results 1062 are well within the same order of magnitude across different years of meteorological data, indicating 1063 minimal impact on the estimated concentrations. Therefore, these findings support EPA’s ongoing use 1064 of the current meteorological data within IIOAC. 1065 1066 Impact of Different Stack Heights in HEM 1067 IIOAC includes a default stack height of 10 m for a point-source stack release. This stack height is based 1068 on a national average stack height across the United States for processes that are not higher-temperature 1069 incinerators or hazardous waste incinerators. The default stack height of 10 m is inherent to the pre-run 1070 AERMOD scenarios from which IIOAC is built, are integrated into the IIOAC model directly, and 1071 cannot be changed. Page 37 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1072 Although 10 m represents a national average stack height, EPA recognizes actual stack heights may vary 1073 (higher or lower) from facility-to-facility. EPA also recognizes the 10-meter stack height, and other 1074 stack parameters integral to IIOAC, represent a low, slow moving, non-buoyant plume and therefore 1075 results in a more conservative concentration estimate at the distances evaluated. Additionally, EPA 1076 recognizes a higher stack height under normal conditions can provide for additional dispersion prior to a 1077 plume reaching the breathing level of individuals within the general population who are then exposed to 1078 pollutants within the plume. 1079 1080 EPA developed and conducted a sensitivity analysis to more thoroughly explore the impacts of different 1081 stack heights on modeled ambient air concentrations at multiple distances from releasing facilities. HEM 1082 is relied upon for this sensitivity analysis because of its added flexibilities to allow user defined stack 1083 parameters (including stack height), distances, and meteorological stations. This particular sensitivity 1084 analysis explored and compared modeled ambient air concentrations resulting from two stack heights 1085 (10- and 25-meters) at 11 finite distances under identical exposure scenarios. 1086 1087 As expected, EPA found the 25-meter stack height allowed for additional dispersion prior to a plume 1088 reaching the breathing zone of individuals at the distances evaluated and where exposure occurs. 1089 Generally, the 25-meter stack height resulted in slightly lower modeled concentrations at the distances 1090 evaluated when compared to the 10-meter stack height at the same distances. Additionally, the greatest 1091 impact from the 25-meter stack height generally occurred at the 1,000 m finite distance from the 1092 releasing facilities while the greatest impact from the 10-meter stack height occurred at the 100 m finite 1093 distance from the releasing facilities. Based on these findings, EPA determined that while there are 1094 differences between estimated concentrations from different stack heights, the impacts are minimal. 1095 Therefore, EPA retains use of the 10-meter stack height and relies upon the IIOAC default stack 1096 parameters to provide a more conservative concentration estimate for this ambient air exposure 1097 assessment. 1098 1099 Fugitive Impact Distances vs. Stack Impact Distances Using HEM 1100 Fugitive and stack type releases are modeled separately in air dispersion models like IIOAC and HEM. 1101 Both models model fugitive releases as an area source with a user defined “area of source” and stack 1102 releases as a point source. Each then provides source apportioned results of estimated concentrations for 1103 each release type at each distance evaluated. EPA utilized these source apportioned results from HEM in 1104 a sensitivity analysis designed to explore two concepts associated with exposures. 1105 1. Which release type has the greatest impact on exposures at each distance evaluated? 1106 2. At what distance, of those evaluated, does each release type have the greatest impact on 1107 exposure? 1108 Results from this sensitivity analysis are shown in Figure_Apx B-2 and can inform exposure, risk 1109 estimates, risk determinations, and risk management rulemaking decisions around a fit-for-purpose 1110 national level risk evaluations. Generally, EPA found fugitive releases have greater overall impacts on 1111 exposures at distances less than 100 m. At distances farther than 100 m, fugitive releases tend to have 1112 similar impacts to modeled concentrations as stack releases. Stack releases, in contrast, have greater 1113 overall impacts on exposures at distances at distances between 60 and 1,000 m followed by a moderate 1114 decline in modeled concentrations beyond 1,000 m. Although these findings align with statements made 1115 in previous work by EPA, this sensitivity analysis specifically explored these findings to inform the 1116 relative impacts of fugitive and stack releases on exposures to individuals residing near industrial 1117 facilities releasing formaldehyde to the ambient air. 1118 Page 38 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1119 1120 Figure_Apx B-2. Median and 95th Percentile Concentrations (Fugitive, Stack, and Total 1121 Emissions) across the 11 Discrete Distance Rings Modeled in HEM 1122 Page 39 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1123 Appendix C IIOAC RESULTS BY COU 1124 1125 Table_Apx C-1. Crosswalk of Industry Sector to COU along with Modeled Exposure Concentrations by Industry Sector for 95th 1126 Percentile Release Scenario and 95th Percentile Modeled Concentration at 100 to 1,000 m from Industrial Facilities Releasing 1127 Formaldehyde to Ambient Air Exposure Concentrations (µg/m3) Industry Sector Associated TSCA COU(s) AC ADC LADC Petroleum Refineries Commercial use 0.63 0.63 0.63 Utilities Commercial use 1.50 1.50 1.50 Industrial Gas Manufacturing Disposal 0.03 0.03 0.03 Services Disposal (services industrial sector includes waste 0.30 0.30 0.30 management industry) Manufacturing of formaldehyde Domestic manufacturing 1.93 1.93 1.93 Transportation Equipment Manufacturing Industrial use – chemical substances in industrial products- 3.40 3.40 3.40 paints and coatings; adhesives and sealants, lubricants Fabricated Metal Product Manufacturing Industrial use – chemical substances in industrial products – 0.19 0.19 0.19 paints and coatings; adhesives and sealants; lubricants Furniture and Related Product Manufacturing Industrial use – chemical substances in industrial products – 2.16 2.16 2.16 paints and coatings; adhesives and sealants; lubricants Organic Fiber Manufacturing Industrial use – non-incorporative activities – 0.20 0.20 0.20 oxidizing/reducing agent; processing aids, not otherwise listed Computer and Electronic Product Manufacturing Industrial use – non-incorporative activities – 0.09 0.09 0.09 oxidizing/reducing agent; processing aids, not otherwise listed (e.g., electroless copper plating) Fabricated Metal Product Manufacturing Industrial use – non-incorporative activities – 0.19 0.19 0.19 oxidizing/reducing agent; processing aids, not otherwise listed (e.g., electroless copper plating) Primary Metal Manufacturing Industrial use – non-incorporative activities – 0.12 0.12 0.12 oxidizing/reducing agent; processing aids, not otherwise listed (e.g., electroless copper plating) Furniture and Related Product Manufacturing Industrial use – non-incorporative activities – used in: 2.16 2.16 2.16 construction PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Exposure Concentrations (µg/m3) Industry Sector Associated TSCA COU(s) AC ADC LADC Wholesale and Retail Trade Manufacturing – importing 0.14 0.14 0.14 Food, beverage, and tobacco product Non-TSCA 4.27 4.27 4.27 manufacturing Pharmaceutical and Medicine Manufacturing Non-TSCA 1.79 1.79 1.79 All Other Chemical Product and Preparation Non-TSCA; processing – incorporation into a formulation, 0.21 0.21 0.21 Manufacturing mixture, or reaction product – other: preservative Custom Compounding of Purchased Resin Processing as a reactant – intermediate 0.20 0.20 0.20 Fabricated Metal Product Manufacturing Processing as a reactant – intermediate 0.19 0.19 0.19 Industrial Gas Manufacturing Processing as a reactant – intermediate 0.03 0.03 0.03 Nonmetallic Mineral Product Manufacturing Processing as a reactant – intermediate 5.75 5.75 5.75 (includes clay, glass, cement, concrete, lime, gypsum, and other nonmetallic mineral product manufacturing) Primary Metal Manufacturing Processing as a reactant – intermediate 0.12 0.12 0.12 Synthetic Dye and Pigment Manufacturing Processing as a reactant – intermediate 0.22 0.22 0.22 Machinery Manufacturing Processing – incorporation into a formulation, mixture, or 1.86 1.86 1.86 reaction product Petrochemical Manufacturing Processing – incorporation into a formulation, mixture, or 1.75 1.75 1.75 reaction product All Other Basic Organic Chemical Manufacturing Processing – incorporation into a formulation, mixture, or 0.24 0.24 0.24 reaction product Asphalt Paving, Roofing, and Coating Materials Processing – incorporation into a formulation, mixture, or 0.35 0.35 0.35 Manufacturing reaction product Soap, Cleaning Compound, and Toilet Preparation Processing – incorporation into a formulation, mixture, or 0.67 0.67 0.67 Manufacturing reaction product Synthetic Dye and Pigment Manufacturing Processing – incorporation into a formulation, mixture, or 0.22 0.22 0.22 reaction product Pesticide, Fertilizer, and Other Agricultural Processing – incorporation into a formulation, mixture, or 1.28 1.28 1.28 Chemical Manufacturing reaction product – agricultural chemicals (nonpesticidal) Page 41 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Exposure Concentrations (µg/m3) Industry Sector Associated TSCA COU(s) AC ADC LADC Plastic Material and Resin Manufacturing Processing – incorporation into a formulation, mixture, or 0.72 0.72 0.72 reaction product – intermediate Paint and Coating Manufacturing Processing – incorporation into a formulation, mixture, or 1.65 1.65 1.65 reaction product – paint additives and coating additives not described by other categories Plastic Material and Resin Manufacturing Processing – incorporation into a formulation, mixture, or 0.72 0.72 0.72 reaction product – paint additives and coating additives not described by other categories Paint and Coating Manufacturing Processing – incorporation into a formulation, mixture, or 1.65 1.65 1.65 reaction product – solvents (which become part of a product formulation or mixture) Plastic Material and Resin Manufacturing Processing – incorporation into a formulation, mixture, or 0.72 0.72 0.72 reaction product – surface active agents Adhesive Manufacturing Processing – incorporation into a formulation, mixture, or 0.05 0.05 0.05 reaction product – adhesives and sealants chemicals Textiles, apparel, and leather manufacturing Processing – incorporation into a formulation, mixture, or 4.53 4.53 4.53 reaction product – bleaching agents All Other Chemical Product and Preparation Processing – incorporation into a formulation, mixture, or 0.21 0.21 0.21 Manufacturing reaction product – intermediate Custom Compounding of Purchased Resin Processing – incorporation into a formulation, mixture, or 0.20 0.20 0.20 reaction product – intermediate Fabricated Metal Product Manufacturing Processing – incorporation into a formulation, mixture, or 0.19 0.19 0.19 reaction product – intermediate Nonmetallic Mineral Product Manufacturing Processing – incorporation into a formulation, mixture, or 5.75 5.75 5.75 (includes clay, glass, cement, concrete, reaction product – intermediate lime, gypsum, and other nonmetallic mineral product manufacturing) Wholesale and Retail Trade Processing – incorporation into a formulation, mixture, or 0.14 0.14 0.14 reaction product – intermediate Paint and Coating Manufacturing Processing – incorporation into a formulation, mixture, or 1.65 1.65 1.65 reaction product – ion exchange agents Page 42 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Exposure Concentrations (µg/m3) Industry Sector Associated TSCA COU(s) AC ADC LADC Adhesive Manufacturing Processing – incorporation into a formulation, mixture, or 0.05 0.05 0.05 reaction product – ion exchange agents Adhesive Manufacturing Processing – incorporation into a formulation, mixture, or 0.05 0.05 0.05 reaction product – lubricant and lubricant additive All Other Chemical Product and Preparation Processing – incorporation into a formulation, mixture, or 0.21 0.21 0.21 Manufacturing reaction product – plating agents and surface treating agents All Other Basic Inorganic Chemical Processing – incorporation into a formulation, mixture, or 0.26 0.26 0.26 Manufacturing reaction product – processing aids, specific to petroleum production All Other Chemical Product and Preparation Processing – incorporation into a formulation, mixture, or 0.21 0.21 0.21 Manufacturing reaction product – processing aids, specific to petroleum production Miscellaneous Manufacturing Processing – incorporation into a formulation, mixture, or 1.79 1.79 1.79 reaction product – solid separation agents Transportation Equipment Manufacturing Processing – incorporation into an article – paint additives 3.40 3.40 3.40 and coating additives Rubber Product Manufacturing Processing – incorporation into article – additive 0.0001 0.0001 0.0001 Nonmetallic Mineral Product Manufacturing Processing – incorporation into article – adhesive and 5.75 5.75 5.75 (includes clay, glass, cement, concrete, sealants lime, gypsum, and other nonmetallic mineral product manufacturing) Paper Manufacturing Processing – incorporation into article – adhesive and 0.79 0.79 0.79 sealants Wood Product Manufacturing Processing – incorporation into article – adhesive and 1.85 1.85 1.85 sealants Textiles, apparel, and leather manufacturing Processing – incorporation into article – finishing agents 4.53 4.53 4.53 Paint and Coating Manufacturing Processing – reactant – adhesive and sealant chemicals 1.65 1.65 1.65 All Other Basic Organic Chemical Manufacturing Processing – reactant – adhesive and sealant chemicals 0.24 0.24 0.24 Plastic Material and Resin Manufacturing Processing – reactant – adhesive and sealant chemicals 0.72 0.72 0.72 Wood Product Manufacturing Processing – reactant – adhesive and sealant chemicals 1.85 1.85 1.85 Page 43 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Exposure Concentrations (µg/m3) Industry Sector Associated TSCA COU(s) AC ADC LADC Wood Product Manufacturing Processing – reactant – bleaching agent 1.85 1.85 1.85 Paper Manufacturing Processing – reactant – intermediate 0.79 0.79 0.79 Pesticide, Fertilizer, and Other Agricultural Processing – reactant – intermediate 1.28 1.28 1.28 Chemical Manufacturing Paint and Coating Manufacturing Processing – reactant – intermediate 1.65 1.65 1.65 Adhesive Manufacturing Processing – reactant – intermediate 0.05 0.05 0.05 All Other Basic Organic Chemical Manufacturing Processing – reactant – intermediate 0.24 0.24 0.24 All Other Chemical Product and Preparation Processing – reactant – intermediate 0.21 0.21 0.21 Manufacturing Petrochemical Manufacturing Processing – reactant – intermediate 1.75 1.75 1.75 Plastic Material and Resin Manufacturing Processing – reactant – intermediate 0.72 0.72 0.72 Plastics Product Manufacturing Processing – reactant – intermediate 1.96 1.96 1.96 Rubber Product Manufacturing Processing – reactant – intermediate 0.0001 0.0001 0.0001 Soap, Cleaning Compound, and Toilet Preparation Processing – reactant – intermediate 0.67 0.67 0.67 Manufacturing Wood Product Manufacturing Processing – reactant – intermediate 1.85 1.85 1.85 Plastics Product Manufacturing Processing – recycling 1.96 1.96 1.96 Wood Product Manufacturing Processing – recycling 1.85 1.85 1.85 Paper Manufacturing Processing – recycling 0.79 0.79 0.79 Wholesale and Retail Trade Processing – repackaging 0.14 0.14 0.14 Mining (except Oil and Gas) and Support Activities 0.02 0.02 0.02 1128 Page 44 of 46 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 1129 Table_Apx C-2. Modeled Exposure Concentrations by COU for 95th Percentile Release Scenario 1130 and 95th Percentile Modeled Concentration at 100 to 1,000 m from Industrial Facilities Releasing 1131 Formaldehyde to Ambient Air Exposure Condition of Use (COU) Concentration (µg/m3) Disposal (services industrial sector includes waste management industry) 3.0E–01 Domestic manufacturing 1.9E00 Industrial use – non-incorporative activities – oxidizing/reducing agent; processing aids, not 2.0E–02 otherwise listed (e.g., electroless copper plating) Industrial use – chemical substances in industrial products – paints and coatings; adhesives 3.4E00 and sealants, lubricants Processing – incorporation into a formulation, mixture, or reaction product 1.9E00 Processing – incorporation into a formulation, mixture, or reaction product – adhesive and 5.4E–02 sealant chemicals Processing – incorporation into a formulation, mixture, or reaction product – agricultural 1.3E00 chemicals (nonpesticidal) Processing – incorporation into a formulation, mixture, or reaction product - intermediate 7.2E–01 Processing – incorporation into a formulation, mixture, or reaction product – ion exchange 1.7E00 agents Processing – incorporation into a formulation, mixture, or reaction product – lubricant and 5.4E–02 lubricant additive Processing – incorporation into a formulation, mixture, or reaction product – other: 2.1E–01 preservative Processing – incorporation into a formulation, mixture, or reaction product – paint additives 1.7E00 and coating additives not described by other categories Processing – incorporation into a formulation, mixture, or reaction product – plating agents 2.1E–01 and surface treating agents Processing – incorporation into a formulation, mixture, or reaction product – processing 2.6E–01 aids, specific to petroleum production Processing – incorporation into a formulation, mixture, or reaction product – solid 1.8E00 separation agents Processing – incorporation into a formulation, mixture, or reaction product – solvents 1.7E00 (which become part of a product formulation or mixture) Processing – incorporation into a formulation, mixture, or reaction product – surface active 7.2E–01 agents Processing – incorporation into a formulation, mixture, or reaction product – bleaching 4.5E00 agents Processing – incorporation into article – additive 1.1E–04 Processing – incorporation into article – finishing agents 4.5E00 Processing – incorporation into a formulation, mixture, or reaction product 2.4E–01 PUBLIC RELEASE – DO NOT CITE OR QUOTE March 2024 Exposure Condition of Use (COU) Concentration (µg/m3) Processing – incorporation into a formulation, mixture, or reaction product – intermediate 1.4E–01 Processing – incorporation into an article – paint additives and coating additives 3.4E00 Processing – incorporation into article –adhesive and sealants 1.9E00 Processing – reactant – adhesive and sealant chemicals 1.7E00 Processing – reactant – adhesives and sealants 1.9E00 Processing – reactant – bleaching agent 1.9E00 Processing – reactant – intermediate 2.0E00 Processing – repackaging 1.4E–01 1132 Page 46 of 46 acknowledges that ambient air monitoring data shows formaldehyde concentrations are below levels of concern (and frequently below detection limits).
• Even in EPA’s most precautionary approach to formaldehyde under the Integrated Risk Information (IRIS) and Toxic Substances Control Act (TSCA) programs, the numbers suggested by ProPublica do not add up. EPA only has sufficient data to quantify an association with very rare types of cancer.

Formaldehyde is one of the most studied and regulated chemical substances in commerce today. Federal standards and regulations are in place to limit formaldehyde emissions and minimize any potential human health risks. For example, the EPA’s regulations under TSCA Title VI, Formaldehyde Standards for Composite Wood Products, are the most stringent formaldehyde product emissions standards in the world today. Since the early 1980s, formaldehyde has been continuously studied and results show that current regulated safe exposure levels are protective. Given formaldehyde’s wide use, extensive product stewardship efforts have been implemented by industry, including state-of-the-art peer-reviewed and published scientific research, to improve understanding and characterization of potential human health risks associated with formaldehyde.

Even though formaldehyde is already highly regulated by numerous agencies and exposures are low, well-managed, and controlled, it is understandable that the public may still have questions about its safety. That is why formaldehyde manufacturers and users continue to take their responsibility seriously, work with regulatory authorities, and provide information on the safe use of formaldehyde and the products made from it.

Learn more about formaldehyde.

¹ See additional state and EPA documents on key caveats/limitations and background contributions: 
https://epd.georgia.gov/document/document/georgia-epd-comments-2017-airtoxscreen-pdf/download;
https://www.regulations.gov/comment/EPA-HQ-OAR-2004-0489-0263 (pg. 24-25);
https://www.epa.gov/system/files/documents/2023-02/AirToxScreen_2018%20TSD.pdfTechnical Support Document EPA's Air Toxics Screening Assessment 2018 AirToxScreen TSD EPA-452/B-22-002 August 2022 Technical Support Document EPA's Air Toxics Screening Assessment 2018 AirToxScreen TSD U.S. Environmental Protection Agency Office of Air Quality Planning and Standards Health and Environmental Impacts Division Air Quality Assessment Division Research Triangle Park, NC Contents Contents ....................................................................................................................................................... iv Tables ......................................................................................................................................................... viii Figures ........................................................................................................................................................ viii Common Acronyms and Abbreviations ........................................................................................................ x 1. Introduction .......................................................................................................................................... 1 1.1. Overview .......................................................................................................................................... 1 1.2. Purpose and Overview of AirToxScreen Steps ................................................................................. 2 1.3. What AirToxScreen Is ....................................................................................................................... 4 1.4. Air Toxics Screening Assessment History ......................................................................................... 5 1.5. How EPA and State, Local and Tribal Air Agencies Use AirToxScreen Results ................................. 6 1.6. How AirToxScreen Results Should Not Be Used ............................................................................... 7 1.7. The Risk Assessment Framework AirToxScreen Uses ...................................................................... 7 1.8. The Scope of AirToxScreen .............................................................................................................. 9 1.8.1. Sources of Air Toxic Emissions That AirToxScreen Addresses ..................................................... 9 1.8.2. Stressors that AirToxScreen Evaluates ...................................................................................... 11 1.8.3. Exposure Pathways, Routes and Time Frames for AirToxScreen .............................................. 11 1.8.4. Receptors that AirToxScreen Characterizes .............................................................................. 12 1.8.5. Endpoints and Measures: Results of AirToxScreen ................................................................... 12 1.9. Model Design ................................................................................................................................. 13 1.9.1. The Strengths and Limitations of the Model Design ................................................................. 15 2. Emissions ............................................................................................................................................. 17 2.1. Sources of Emissions ...................................................................................................................... 17 2.1.1. Pollutants and Pollutant Groups ............................................................................................... 20 2.1.2. Emissions Categorization: NEI and AirToxScreen ..................................................................... 23 2.1.3. Differences Between the current NEI and Emissions Used for AirToxScreen .......................... 25 2.1.4. Overview of Differences in Emissions for CMAQ and AERMOD ............................................... 27 2.2. Preparation of Emissions Inputs for CMAQ ................................................................................... 29 2.2.1. Sectors in the CMAQ AirToxScreen Platform ............................................................................ 29 2.2.2. Fires and Biogenics .................................................................................................................... 32 2.2.3. Speciation .................................................................................................................................. 33 2.2.4. Temporalization ........................................................................................................................ 33 2.2.5. Spatial Allocation ...................................................................................................................... 34 2.3. Preparation of Emissions Inputs for AERMOD ............................................................................... 35 2.3.1. Source and Run Groups – Overview ......................................................................................... 35 2.3.2. Point Sources Excluding Airports .............................................................................................. 43 2.3.3. Airport Point Sources ................................................................................................................ 47 2.3.4. Nonroad, On-road and Nonpoint – County-level Sources ........................................................ 50 2.3.5. Residential Wood Combustion ................................................................................................. 53 2.3.6. Oil and Gas ................................................................................................................................ 53 2.3.7. Agricultural Livestock ................................................................................................................ 54 AirToxScreen 2018 Documentation iv 2.3.8. Commercial Marine Vessels – County/Shape-level Sources ..................................................... 54 2.3.9. Urban/Rural Determination for All Emission Sources .............................................................. 56 3. Air Quality Modeling and Characterization ........................................................................................ 57 3.1. Modeling Overview ........................................................................................................................ 57 3.1.1. Photochemical Model Selection ............................................................................................... 57 3.1.2. Dispersion Model Selection ...................................................................................................... 59 3.2. Meteorological Data ...................................................................................................................... 59 3.2.1. Meteorological Data Inside the Contiguous States .................................................................. 59 3.2.2. Meteorological Data Outside the Contiguous States................................................................ 61 3.3. CMAQ Setup ................................................................................................................................... 63 3.3.1. Sources Modeled in CMAQ ....................................................................................................... 63 3.3.2. Boundary and Initial Conditions ................................................................................................ 68 3.4. AERMOD Setup .............................................................................................................................. 70 3.4.1. Sources Modeled in AERMOD ................................................................................................... 70 3.4.2. Receptor Placement .................................................................................................................. 70 3.4.3. Model Options .......................................................................................................................... 74 3.4.4. AERMOD Simulations ................................................................................................................ 74 3.4.5. Post-processing of AERMOD Results ........................................................................................ 75 3.5. Hybrid Modeling ............................................................................................................................ 76 3.5.1. Overview ................................................................................................................................... 77 3.5.2. Treatment of Species ................................................................................................................ 78 3.5.3. Ambient Monitoring Data Preparation ..................................................................................... 80 3.5.4. Model Performance Statistics ................................................................................................... 81 3.5.5. Hybrid Evaluation ...................................................................................................................... 83 3.5.6. Non-hybrid Evaluation .............................................................................................................. 86 4. Estimating Exposures for Populations ................................................................................................. 88 4.1. Estimating Exposure Concentrations ............................................................................................. 88 4.2. About HAPEM ................................................................................................................................ 88 4.3. HAPEM Inputs and Application ...................................................................................................... 89 4.3.1. Data on Ambient Air Concentrations ........................................................................................ 90 4.3.2. Population Demographic Data .................................................................................................. 90 4.3.3. Data on Population Activity ....................................................................................................... 90 4.3.4. Microenvironmental Data ......................................................................................................... 91 4.4. Exposure Factors ............................................................................................................................ 93 4.5. Evaluating Exposure Modeling ...................................................................................................... 94 4.6. Summary ........................................................................................................................................ 94 5. Characterizing Effects of Air Toxics ..................................................................................................... 95 5.1. Toxicity Values and Their Use in AirToxScreen .............................................................................. 95 5.2. Types of Toxicity Values ................................................................................................................. 96 5.2.1. Cancer URE ................................................................................................................................ 96 5.2.2. Noncancer Chronic RfC ............................................................................................................. 99 5.3. Data Sources for Toxicity Values .................................................................................................... 99 5.3.1. U.S. EPA Integrated Risk Information System ......................................................................... 100 AirToxScreen 2018 Documentation v 5.3.2. U.S. Department of Health and Human Services, Agency for Toxic Substances and Disease Registry ...................................................................................................................... 100 5.3.3. California Environmental Protection Agency Office of Environmental Health Hazard Assessment ............................................................................................................................. 100 5.3.4. U.S. EPA Health Effects Assessment Summary Tables ............................................................ 100 5.3.5. World Health Organization International Agency for Research on Cancer ............................. 101 5.4. Additional Toxicity Decisions for Some Chemicals ....................................................................... 101 5.4.1. Polycyclic Organic Matter ........................................................................................................ 101 5.4.2. Glycol Ethers ........................................................................................................................... 102 5.4.3. Acrolein ................................................................................................................................... 102 5.4.4. Metals ..................................................................................................................................... 102 5.4.5. Adjustment of Mutagen UREs to Account for Exposure During Childhood ............................ 103 5.4.6. Diesel Particulate Matter ........................................................................................................ 104 5.5. Summary ...................................................................................................................................... 105 6. Characterizing Risks and Hazards in AirToxScreen ............................................................................ 106 6.1. The Risk-characterization Questions AirToxScreen Addresses .................................................... 106 6.2. How Cancer Risk Is Estimated ...................................................................................................... 107 6.2.1. Individual Pollutant Risk .......................................................................................................... 107 6.2.2. Multiple-pollutant Risks .......................................................................................................... 107 6.3. How Noncancer Hazard is Estimated ........................................................................................... 108 6.3.1. Individual Pollutant Hazard ..................................................................................................... 108 6.3.2. Multiple-pollutant Hazard ....................................................................................................... 109 6.4. How Risk Estimates and Hazard Quotients Are Calculated for AirToxScreen at Tract, County and State Levels ............................................................................................................... 110 6.4.1. Model Results for Point Sources: Aggregation to Tract-level Results ..................................... 110 6.4.2. Background Concentrations and Secondary Pollutants: Interpolation to Tract-level Results ..................................................................................................................................... 110 6.4.3. Aggregation of Tract-level Results to Larger Spatial Units ...................................................... 110 6.5. The Risk Characterization Results That AirToxScreen Reports .................................................... 111 6.6. Summary ...................................................................................................................................... 113 7. Variability and Uncertainty Associated with AirToxScreen ............................................................... 114 7.1. Introduction ................................................................................................................................. 114 7.2. How AirToxScreen Addresses Variability ..................................................................................... 114 7.2.1. Components of Variability ....................................................................................................... 115 7.2.2. Quantifying Variability ............................................................................................................ 116 7.2.3. How Variability Affects Interpretation of AirToxScreen Results ............................................. 118 7.3. How AirToxScreen Addresses Uncertainty .................................................................................. 119 7.3.1. Components of Uncertainty .................................................................................................... 119 7.3.2. Components of Uncertainty Included in AirToxScreen ........................................................... 121 7.4. Summary of Limitations in AirToxScreen ..................................................................................... 125 AirToxScreen 2018 Documentation vi 8. References ........................................................................................................................................ 128 Appendix A. Glossary ............................................................................................................................. A-1 Appendix B. Air Toxics Modeled in AirToxScreen .................................................................................. B-1 Appendix C. Estimating Background Concentrations for AirToxScreen ................................................. C-2 Appendix D. Model Evaluation Summaries ........................................................................................... D-1 Appendix E. Exposure Factors for AirToxScreen .................................................................................... E-1 Introduction to HAPEM and its Use in AirToxScreen ................................................................................. E-2 AirToxScreen 2018 Documentation vii Tables Table 2-1. Summary of emissions sources in the 2018 NEI ................................................................................ 18 Table 2-2. Pollutant groups ................................................................................................................................. 20 Table 2-3. PAH/POM pollutants group ............................................................................................................... 21 Table 2-4. Metal speciation factors for NEI metal compounds .......................................................................... 23 Table 2-5. Map of NEI data categories to AirToxScreen categories ................................................................... 24 Table 2-6. Release parameter defaults/changes to the FF10 inventory files for point sources ......................... 26 Table 2-7. Differences in spatial characterization of sources between CMAQ and AERMOD ............................ 29 Table 2-8. Platform sectors for the 2018 emissions modeling platform ............................................................ 30 Table 2-9. Run groups for AERMOD .................................................................................................................... 36 Table 2-10. Resolution of the run groups modeled as gridded sources ............................................................. 37 Table 2-11. Non-gridded AERMOD run groups ................................................................................................... 37 Table 2-12. AERMOD gridded run groups ........................................................................................................... 38 Table 2-13. AirToxScreen source groups ............................................................................................................ 41 Table 2-14. Assignment of AERMOD source type for point sources .................................................................. 45 Table 2-15. Options for temporal variation specification in helper files ............................................................ 47 Table 2-16. Lead emissions (kg/yr) at SMO in 2008 by aircraft operation mode ............................................... 49 Table 2-17. Airport emissions file format ........................................................................................................... 49 Table 2-18. NEI SCCs covered in the CMV run group .......................................................................................... 54 Table 3-1. CMAQ HAPs ........................................................................................................................................ 57 Table 3-2. Vertical layer structure for WRF and CMAQ (heights are layer top) ................................................. 60 Table 3-3. WRF Options Used for the Alaska, Hawaii, and Puerto Rico/Virgin Islands domains. ....................... 62 Table 3-4. Boundary conditions from 2018 remote concentration estimates ................................................... 68 Table 3-5. CMAQ HAP boundary conditions applied as zero value .................................................................... 69 Table 3-9. List of hybrid HAPs evaluated ............................................................................................................ 83 Table 4-1. Key differences between recent versions of HAPEM ........................................................................ 89 Table 4-2. Microenvironments used in HAPEM modeling for AirToxScreen ...................................................... 93 Table 6-1. Criteria establishing AirToxScreen drivers and contributors of health effects for risk characterization .............................................................................................................................. 112 Figures Figure 1-1. AirToxScreen – basic steps ................................................................................................................. 2 Figure 1-2. Detailed steps and approach used in AirToxScreen ........................................................................... 4 Figure 1-3. The general air toxics risk assessment process .................................................................................. 8 Figure 1-4. Conceptual model for AirToxScreen ................................................................................................. 10 Figure 1-5. The AirToxScreen risk assessment process and corresponding sections of this TSD ....................... 14 AirToxScreen 2018 Documentation viii Figure 2-1. Example fugitive source characterization: NEI length = 1897 feet, width = 680 feet and angle = 22 ......................................................................................................................................... 46 Figure 2-2. Port shapes for Los Angeles and Long Beach, California .................................................................. 55 Figure 3-1(a). Map of the CMAQ modeling domain; the blue box denotes the 12 km CONUS modeling domain ............................................................................................................................................. 64 Figure 3-1(b). Map of the CMAQ modeling domain; the green box denotes the 9 km Alaska modeling domain ............................................................................................................................................. 65 Figure 3-1(c). Map of the CMAQ modeling domain; the red box denotes the 3 km Hawai’i modeling domain ............................................................................................................................................. 66 Figure 3-1(d). Map of the CMAQ modeling domain; the red box denotes the 3 km Puerto Rico/Virgin Islands modeling domain ................................................................................................................. 67 Figure 3-2. CBSAs exceeding 1 million people .................................................................................................... 72 Figure 3-3. Dense (left) and coarse (right) receptor grid layout in CMAQ Lambert Projection .......................... 73 Figure 3-4. Example grid cell with subgrid cells and census blocks .................................................................... 73 Figure 3-5. CMAQ domain with expanded cell showing hybrid receptors; colors indicate modeled concentrations; dots in inset show locations of receptors within a grid cell ................................... 77 AirToxScreen 2018 Documentation ix Common Acronyms and Abbreviations µg/m3 microgram(s)/cubic meter AERMOD atmospheric dispersion model developed by the American Meteorological Society and the U.S. Environmental Protection Agency’s Regulatory Model Improvement Committee ASPEN Assessment System for Population Exposure Nationwide ATSDR Agency for Toxic Substances and Disease Registry CAA Clean Air Act CHAD Consolidated Human Activity Database CAP criteria air pollutant CMAQ Community Multiscale Air Quality model CONUS Contiguous United States (modeling domain for CMAQ) EC exposure concentration EPA Environmental Protection Agency EGU electricity generating unit HAP hazardous air pollutant HAPEM Hazardous Air Pollutant Exposure Model HEM Human Exposure Model HI hazard index HQ hazard quotient IRIS Integrated Risk Information System ISC Industrial Source Complex MOVES Motor Vehicle Emissions Simulator NATA National Air Toxics Assessment NEI National Emissions Inventory NMIM National Mobile Inventory Model OAQPS Office of Air Quality Planning and Standards PAH polycyclic aromatic hydrocarbon PM particulate matter PM2.5 particulate matter less than 2.5 microns in diameter PM10 particulate matter less than 10 microns in diameter POM polycyclic organic matter RfC reference concentration RTR Risk and Technology Review SCC Source Classification Code S/L/T state, local or tribal agency URE unit risk estimate WRF Weather Research Forecasting model AirToxScreen 2018 Documentation x 1. Introduction 1.1. Overview The Air Toxics Data Update is the U.S. Environmental Protection Agency’s (EPA; referred to throughout this document as “we”) ongoing thorough evaluation of air toxics across the United States. A key part of the Air Toxics Data Update is the Air Toxics Screening Assessment, or AirToxScreen. EPA developed AirToxScreen as a state-of-the-science tool to inform both national and localized efforts to collect air toxics information, characterize emissions and help prioritize pollutants and areas of interest for further study to gain a better understanding of risks. AirToxScreen is the successor to the National Air Toxics Assessment, or NATA. The goal of AirToxScreen is to identify those air toxics which are of greatest potential concern in terms of contribution to population risk. Ambient and exposure concentrations and estimates of risk and hazard for air toxics in each state are typically generated at the census tract level. This AirToxScreen Technical Support Document (TSD) describes the data and approaches EPA used to conduct AirToxScreen, including descriptions of how we:  compiled emissions data and prepared them for use as model inputs (Section 2);  estimated ambient concentrations of air toxics (Section 3);  estimated exposures to air toxics for populations (Section 4);  selected toxicity values (Section 5);  characterized human-health risks and hazards (Section 6); and  addressed variability and uncertainty (Section 7). References to additional documents are included (Section 8) to facilitate access to more detailed technical information on the emissions inventories, dispersion modeling, photochemical modeling, exposure modeling and toxicity values. The TSD also includes the following appendixes:  Appendix A – a glossary of the key terms and their definitions  Appendix B – a list of air toxics included in AirToxScreen  Appendix C – procedures used to estimate AirToxScreen background concentrations  Appendix D – additional model evaluation summaries  Appendix E – documentation on HAPEM7 and its use in AirToxScreen AirToxScreen 2018 Documentation 1 We also provide a “Supplemental Data” folder with this document that contains the Microsoft® Access™ and Microsoft® Excel™ files referenced throughout this TSD. This TSD satisfies basic documentation protocol expected of EPA products and provides a resource for the technically oriented user community by summarizing the data sources, methods, models and assumptions used in AirToxScreen. This document does not provide quantitative results for AirToxScreen and thus presents no exposure or risk estimates. You can find results and other specific information for AirToxScreen on the AirToxScreen website (https://www.epa.gov/AirToxScreen). Links to previous assessments are also available on this site. 1.2. Purpose and Overview of AirToxScreen Steps AirToxScreen follows several basic steps to produce the final assessment. These AirToxScreen steps are depicted in Figure 1-1. Figure 1-1. AirToxScreen – basic steps The first and most time-consuming step in AirToxScreen is assembling the National Emissions Inventory (NEI), a detailed, nationwide inventory of air toxics emissions. The NEI includes emissions from point, nonpoint and mobile sources, as well as emissions from biogenic sources and fires. These source data are the foundation of AirToxScreen’s air quality modeling. Section 2 of this document details the steps EPA used to construct this extensive database. After preparing the NEI emissions and other needed data (e.g., meteorological data), we use these data as inputs to two air quality models used to estimate ambient air concentrations of air toxics: the American Meteorological Society/Environmental Protection Agency Regulatory Model (AERMOD) atmospheric dispersion model and the Community Multiscale Air Quality (CMAQ) photochemical model. AERMOD is used for all AirToxScreen air toxics modeled, and CMAQ is used for a list of 52 air toxics that are incorporated into CMAQ multipollutant version 5.2. CMAQ provides the overall mass, chemistry and formation for hazardous air pollutants (HAPs) formed secondarily in the atmosphere (e.g., formaldehyde, acetaldehyde and acrolein), whereas AERMOD provides spatial granularity and more detailed source attribution. CMAQ also provides the biogenic and fire concentrations, as these sources are not run in AERMOD. Only AERMOD is run in areas outside the lower 48, or contiguous, U.S. states (referred to as CONUS) that are included in AirToxScreen: Alaska, Hawaii, Puerto Rico and the Virgin Islands. Special steps are taken to estimate secondary HAPs, fires and biogenics in these areas. Section 3 of this document details the steps EPA used to model for AirToxScreen. AirToxScreen 2018 Documentation 2 For the HAPs modeled in both CMAQ and AERMOD, we combine the model-calculated annual average concentrations using a hybrid approach. We next use these concentrations, along with the other concentrations calculated by AERMOD, to prepare census tract-level concentrations of all modeled air toxics. Then, using the HAPEM7 exposure model, we account for human activity patterns and develop exposure concentrations, or ECs, for each census tract. Finally, we estimate census tract-level risks by applying health benchmark data to the ECs. Sections 4 through 6 of this TSD detail these steps. EPA’s state, local and tribal (S/L/T) air agency partners play an integral role in AirToxScreen. First, S/L/T specialists review early versions of NEI’s source data for their area, working with local industry and other emissions sources to develop and forward corrections to their area’s emissions data. They also review preliminary risk results during early stages of the AirToxScreen modeling process, which often reveals other inaccuracies in the data. This review and feedback process helps ensure that AirToxScreen’s input data are as accurate as possible in the final version. For the 2018 AirToxScreen, we preliminarily modeled using AERMOD the 2018 NEI, the results of which were used to prioritize review of emissions that might contribute to high risk in the final results. Our S/L/T partners reviewed the June 2020 version of the 2018 point source inventory (without airports and railyards) along with a 2018 risk estimates scaled from 2017 first-pass modeling and risk results. We reviewed and incorporated comments into the AirToxScreen modeling process and results. We then previewed these final results with S/L/T agencies before releasing the 2018 AirToxScreen in August 2022. Figure 1-2 provides a more detailed flowchart showing the emissions sources used in the air quality models and how the hybrid approach fits into the overall approach. AirToxScreen 2018 Documentation 3 Figure 1-2. Detailed steps and approach used in AirToxScreen 1.3. What AirToxScreen Is AirToxScreen is a first-pass, screening tool intended to evaluate the human-health risks posed by air toxics across the United States. AirToxScreen provides screening-level estimates of the risk of cancer and other potentially serious health effects from inhaling air toxics. AirToxScreen uses emissions data compiled for a single year as inputs to air quality models. The models use these source data along with meteorological data for the same year to estimate ambient air concentrations of certain air toxics. EPA then combines these modeled concentrations with census data and other information to calculate exposure concentrations of the air toxics. We also estimate cancer risks and potential noncancer health effects associated with chronic inhalation exposure to the toxics. EPA generates AirToxScreen results for each U.S. state at county and census tract levels. We also generate results for Puerto Rico, the Virgin Islands and the District of Columbia. These results help state, local and tribal agencies prioritize air toxics, emission sources and locations of interest for further study. They also help air agencies plan and implement national, regional and local efforts to reduce toxic air pollution. AirToxScreen provides a “snapshot” of outdoor air quality as it relates to air toxics. It also suggests the risks to human health if air toxic emission levels were to remain the same as those estimated for the assessment year. The estimates only reflect risks associated with chronic (assumed to be lifetime) AirToxScreen 2018 Documentation 4 inhalation exposures to air toxics at the population level. Assumptions and methods we use to complete the assessment limit the types of questions that AirToxScreen can answer reliably. You should consider these limitations, described throughout this document and summarized in Section 7, when interpreting the AirToxScreen results or when using them to address questions posed outside of AirToxScreen. AirToxScreen results can provide general answers to questions about emissions, ambient air concentrations and exposures and risks across broad geographic areas (such as counties, states and the nation) for the year modeled in the assessment. AirToxScreen can answer questions such as the following:  Which air toxics pose the greatest potential risk of cancer or adverse noncancer effects across the entire United States?  Which air toxics pose the greatest potential risk of cancer or adverse noncancer effects in certain areas of the United States?  Which air toxics pose less, but still significant, potential risk of cancer or adverse noncancer effects across the entire United States?  When risks from long-term inhalation exposures to all outdoor air toxics are considered together, how many people could experience a lifetime cancer risk greater than levels of concern (e.g., 1-in-1 million or 100-in-1 million)?  When considering potential adverse noncancer effects from long-term exposures to all outdoor air toxics together for a given target organ or system, how many people could experience exposures that exceed the reference levels intended to protect against those effects (i.e., a hazard quotient greater than 1)? 1.4. Air Toxics Screening Assessment History EPA’s first national-scale air toxics study was the Cumulative Exposure Project (Caldwell et al. 1998). EPA developed this project based on estimates of air toxics emissions present before the Clean Air Act (CAA) was amended in 1990. The Cumulative Exposure Project estimated outdoor air toxics concentrations in each contiguous-U.S. census tract. For the first National Air Toxics Assessment (NATA), EPA enhanced the Cumulative Exposure Project framework to include estimates of population exposure and health risk. The first NATA used a more refined inventory of air toxics emissions developed for 1996, known at that time as the National Toxics Inventory. EPA submitted this assessment for a technical peer review in January 2001 to a panel of EPA’s Science Advisory Board (EPA 2001b). The panel provided detailed comments later that year on the validity of the overall approach, the elements of the assessment (including the data, models and methods used) and the manner in which these components were integrated into a national-scale assessment (EPA 2001a). EPA incorporated many of the Science Advisory Board’s suggestions into NATA and published the results of that assessment in 2002. Since then, five assessments have been completed – representative of air toxic emissions in 1999, 2002, 2005, 2011 and 2014, respectively – based on significant triennial updates AirToxScreen 2018 Documentation 5 of the national emission inventories. In general, the scope of NATA progressively expanded with subsequent versions, and some methods were refined and improved. The AirToxScreen assessments, which started in 2017, use the same basic methods as those used in the 2014 NATA. 1.5. How EPA and State, Local and Tribal Air Agencies Use AirToxScreen Results We designed AirToxScreen to help guide efforts to reduce toxic air pollution and to provide information that can be used to further the already significant emissions reductions achieved in the United States since 1990. EPA and S/L/T air agencies use AirToxScreen to identify those air toxics and source sectors (e.g., point or mobile sources) having the highest exposures and health risks. The assessment results also help us identify geographic patterns and ranges of risks across the country and across individual states and county areas within states. Specifically, we use AirToxScreen results to:  identify pollutants and industrial source categories of greatest concern;  improve understanding of health risks posed by air toxics;  help set priorities for the collection of additional information;  set priorities for improving emission inventories;  expand and prioritize EPA’s network of air toxics monitors;  support communities in designing their own local assessments;  enhance targeted risk-reduction activities; and  provide a multiple-pollutant modeling framework linking air toxics to the Criteria Pollutant Program (EPA 2015c). Similarly, S/L/T air agencies use AirToxScreen to:  prioritize pollutants and emission source types;  identify places of interest for further study;  get a starting point for local assessments;  focus community efforts; and  inform monitoring programs. AirToxScreen 2018 Documentation 6 1.6. How AirToxScreen Results Should Not Be Used As described above, AirToxScreen is a screening-level assessment, designed to answer specific types of questions. The underlying assumptions of AirToxScreen and its methods limit the range of questions that can be answered reliably. AirToxScreen results should not be used:  as a definitive means to pinpoint specific risk values within a census tract;  to characterize or compare risks at local levels (such as between neighborhoods);  to characterize or compare risks between states;  as the sole basis for developing risk reduction plans or regulations;  as the sole basis for determining appropriate controls on specific sources or air toxics; or  as the sole basis to quantify benefits of reduced air toxic emissions. The limitations of the assessment methods prevent AirToxScreen from serving as a stand-alone tool. Furthermore, although EPA reports results at the census tract level in AirToxScreen, average risk estimates are far more uncertain at this level of spatial resolution than at the county or state level. To analyze air toxics in smaller areas, such as census blocks or in a suspected “hotspot,” other tools such as site-specific monitoring and local-scale assessments coupled with refined and localized data should be used. These caveats are integral to the proper interpretation of AirToxScreen results. You should use AirToxScreen results only to address those questions for which the assessment methods are suited. EPA does not use AirToxScreen as the sole source of information leading to regulations or guiding the enforcement of existing rules. Some of the methods used to conduct AirToxScreen are like those used in air-related risk assessments conducted under the CAA mandate (such as residual risk assessments of HAP emissions from point sources, or assessments of exposures to criteria air pollutants (CAPs) for evaluations of National Ambient Air Quality Standards), AirToxScreen fundamentally differs from such assessments in that it is not a regulatory program. 1.7. The Risk Assessment Framework AirToxScreen Uses In AirToxScreen we use methods consistent with the general risk assessment framework used throughout EPA. This section overviews EPA’s risk assessment framework and summarizes the AirToxScreen process. Later sections detail the analytical components of this process. EPA has published a series of guidelines (EPA 2015d) that establishes and explains the methods recommended for assessing human-health risks from environmental pollution. This series makes recommendations for carcinogen risk assessment, exposure assessment, chemical mixtures risk assessment and other major EPA-wide risk assessments. EPA has also developed the three-volume Air Toxics Risk Assessment (ATRA) Reference Library (EPA 2004a,b; EPA 2006a) as a reference for those conducting air toxics risk assessments. This library details the fundamental principles of risk-based AirToxScreen 2018 Documentation 7 assessment for air toxics, how to apply those principles in various settings, and strategies for reducing risk at the local level. EPA’s guidelines and methods are consistent with the National Research Council’s recommendations on conducting risk assessments (NRC 1983, 1994). As described in more detail in these guidelines and documents, EPA’s risk assessment process has three phases (Figure 1-3), the second of which has two parts.  The first phase (problem formulation) comprises the initial planning and scoping activities and definition of the problem, which results in the development of a conceptual model.  The second phase (analysis) includes two components:  Exposure assessment; and  Toxicity assessment.  The third phase is risk characterization, a synthesis of the outputs of the exposure and toxicity assessments to characterize health risks for the scenario described in the initial phase. Analysis Toxicity Assessment Exposure Assessment Risk Characterization Figure 1-3. The general air toxics risk assessment process Source: Adapted from EPA (2004a). An air toxics risk assessment starts with problem formulation. This first step begins with the systematic planning and scoping needed before any analyses are begun. This planning process helps ensure that the objectives of the assessment are met, resources are used efficiently and the overall effort is successful. One important product of the problem formulation step is a conceptual model that describes how releases of air toxics might pose risks to people. The conceptual model serves as a guide or “road map” to the assessment. It defines the physical boundaries, potential sources, emitted air toxics, potentially exposed populations, chemical fate and transport processes, expected routes of exposure and potential health effects. The planning and scoping activities and problem formulation we conduct before carrying out the analyses, are critical – they set the course for the assessment and inform EPA’s decisions AirToxScreen 2018 Documentation 8 regarding specific methods, models and data sources to use. The following section (1.8) describes the conceptual model developed for AirToxScreen – the product of the first phase. Meanwhile, the rest of this document is concerned primarily with describing the analysis phase of the general air toxics risk assessment process (and specifically with describing the analyses conducted for AirToxScreen). The analysis phase is the stage at which we use the risk assessment processes to evaluate the problem at hand. Section 1.9 outlines the analytical steps, with detailed descriptions of each step presented in later sections of this document. 1.8. The Scope of AirToxScreen The national-scale assessment described in this document is consistent with EPA’s definition of a cumulative risk assessment, as stated in EPA’s Framework for Cumulative Risk Assessment (EPA 2003, p. 6), as “an analysis, characterization, and possible quantification of the combined risks to health or the environment from multiple agents or stressors.” The Framework emphasizes that a conceptual model is an important output of the problem formulation phase of a cumulative risk assessment. The conceptual model defines the actual or predicted relationships among exposed individuals, populations or ecosystems and the chemicals or stressors to which they might be exposed. Specifically, the conceptual model lays out the sources, stressors, environmental media, routes of exposure, receptors and endpoints (i.e., measures of effects) relevant to the problem or situation that is being evaluated. This model takes the form of a written description and a visual representation of the relationships among these components. The conceptual model can sometimes include components that are not addressed specifically or quantitatively by an assessment, but that are nevertheless important to consider. Section 2.4 of the report for the 1996 NATA presented to EPA’s Science Advisory Board for review (EPA 2001b) included a conceptual model. Some of the specifics included in that conceptual model have since evolved as newer assessments have been completed (for example, the number of air toxics evaluated has increased substantially since the 1996 NATA). The fundamental components included in NATA and the relationships among them, however, have been generally consistent for all six NATAs and for AirToxScreen. Moreover, the conceptual model described in this document is very similar to the one presented in the documentation for the 1996 NATA. AirToxScreen is national in scope, covering the United States, Puerto Rico and the U.S. Virgin Islands. It focuses on long-term inhalation exposures to air toxics. In general, AirToxScreen is intended to provide EPA with the best possible national-scale population-level estimates of exposure to and risks associated with air toxics, considering data availability, technical capabilities and other potentially limiting factors. The conceptual model for the AirToxScreen is presented in Figure 1-4. Each component included in the model is described briefly in the sections that follow. 1.8.1. Sources of Air Toxic Emissions That AirToxScreen Addresses Sources of primary air toxic emissions included in AirToxScreen (i.e., the AirToxScreen categories) are point, nonpoint, mobile, biogenics and fires in the contiguous United States. Point, nonpoint and mobile sources are included from Alaska, Hawaii, Puerto Rico and the U.S. Virgin Island (with impacts from biogenics and fires estimated). Examples of point sources are large waste incinerators and factories. Nonpoint sources include residential wood combustion (RWC), commercial cooking, and consumer and AirToxScreen 2018 Documentation 9 commercial solvents. Mobile sources include vehicles found on roads and highways, such as cars and trucks, and nonroad equipment, including lawn mowers and construction equipment. Nonroad sources also include marine vessels, trains and aircraft. Receptors/Subpopulations Figure 1-4. Conceptual model for AirToxScreen Blue boxes indicate elements included in AirToxScreen; clear boxes indicate elements that could be included in future assessments. In the “Sources” included here, “Major stationary” includes both major and area sources as defined for regulatory purposes in the CAA. “Nonpoint” refers to smaller (and sometimes less discrete) sources that are typically estimated on a top-down basis (e.g., by county). Additional explanation of source types included in AirToxScreen is presented in Section 2. DPM refers to diesel particulate matter. HQ and HI refer to hazard quotient and hazard index, respectively. AirToxScreen only considers outdoor sources of air toxics. It does not address indoor sources of toxics, for example, those emitted from household chemicals. In addition, AirToxScreen background estimates do not consider background air toxics from other media, such as water. AirToxScreen presents results by both these broad source categories and by more detailed AirToxScreen source groups. Details on this and other aspects of emission sources are presented in Section 2; details on air quality modeling and characterization are presented in Section 3. AirToxScreen 2018 Documentation 10 1.8.2. Stressors that AirToxScreen Evaluates The stressors evaluated through AirToxScreen can include any of the 188 current HAPs defined in the CAA and diesel PM. The set of air toxics included in AirToxScreen is determined by the emission and toxicity data available at the time of the assessment. Diesel PM, an indicator of diesel exhaust, is included in the set of stressors for AirToxScreen. The spreadsheet file “AirToxScreen_Pollutants.xlsx” within the Supplemental Data folder accompanying this TSD lists the pollutants that AirToxScreen assesses and provides more detailed information on the NEI and AirToxScreen pollutants. In Appendix B of this document, Table B-1 lists the CAA pollutants that are not included in AirToxScreen and the reason for their omission. 1.8.3. Exposure Pathways, Routes and Time Frames for AirToxScreen Exposure to air toxics from all sources is determined by multiple interactions among complex factors, including the locations and nature of the emissions, the emission-release conditions, local meteorology, locations of receptor populations, and the specific behaviors and physiology of individuals in those populations. The combination of air toxics that people inhale, and the chemical interactions among those air toxics, influence the risks associated with these exposures. This high level of complexity makes aggregating risk across both substances and sources useful for depicting the magnitude of risks associated with inhalation of air toxics. The air quality modeling step of AirToxScreen includes evaluating the transport of emitted particles and gases through the air to receptors. AirToxScreen modeling accounts for transformation of substances in the atmosphere (also referred to as secondary formation) and losses of substances from the air by deposition, where data are available and the modeling approach supports it. For air toxics with sufficient ambient monitoring data, or with emissions data primarily due to point sources, we estimate background concentrations. With fate and transport of emissions considered, and the presence of some background concentrations, AirToxScreen estimates outdoor ambient concentrations across the nation. AirToxScreen focuses on exposures due to inhalation of ambient air. Human receptors are modeled to account for an individual’s movement among microenvironments, such as residences, offices, schools, exterior work sites and automobiles, where concentration levels can be quite different from general outdoor concentrations. The exposure assessment estimates air concentrations for each substance within each modeled microenvironment. The exposure assessment also accounts for human activities that can affect the magnitude of exposure (e.g., exercising, sleeping). This component of AirToxScreen accounts for the difference between ambient outdoor concentrations and the exposure concentrations (ECs; i.e., long-term-average concentrations to which people are exposed after accounting for human activities). To date, AirToxScreen has not estimated air toxic concentrations in water, soil or food associated with deposition from air, or the bioaccumulation of air toxics in tissues. Similarly, AirToxScreen has not estimated human exposures to chemicals via ingestion or dermal contact. EPA considers these pathways important, but refined tools and data required to model multipathway concentrations and human exposures on the national scale are not yet readily available for use for many air toxics. AirToxScreen estimates average annual outdoor concentrations, which are used to develop long-term inhalation exposures for each of the air toxics. For cancer and chronic (long-term) health effects, the exposure is assumed to be continuous over a lifetime (i.e., 70 years for the purposes of this analysis). AirToxScreen 2018 Documentation 11 Subchronic and acute (lasting less than 24 hours) exposures are not estimated in AirToxScreen because the emissions database contains only annual-total emissions. If the emission inventories are later expanded to cover short-term (e.g., hourly, daily) emission rates, we would consider incorporating shorter exposure times into AirToxScreen. 1.8.4. Receptors that AirToxScreen Characterizes AirToxScreen characterizes average risks to people belonging to distinct human subpopulations. The overall population is divided into cohorts based on residential location, life stage (age) and daily activity pattern. A cohort is generally defined as a group of people within a population assumed to have identical exposures during a specified exposure period. Residential locations are specified according to U.S. Census tracts, which are geographic subdivisions of counties that vary in size but typically contain about 4,000 residents each. Life stages are stratified into six age groups: 0–1, 2–4, 5–15, 16–17, 18–64, and 65 and older. Daily-activity patterns specify time spent in various microenvironments (e.g., indoors at home, in vehicles, outdoors) at various times of day. For each combination of residential census tract and age, 30 sets of age-appropriate daily activity patterns are selected to represent the range of exposure conditions for residents of the tract. A population-weighted typical exposure estimate is calculated for each cohort, and this value is used to estimate representative risks, as well as the range of risks, for a “typical” individual residing in that tract. Risk results for individual cohorts are not included in AirToxScreen results. AirToxScreen does not inclue non-human receptors (e.g., wildlife and native plants). The complexity of the varied ecosystems across the vast area AirToxScreen covers precludes considering potential adverse ecological impacts at this time. Local- and urban-scale assessments can be developed to include non- human receptors, contingent on the availability of necessary resources, data and methodologies. We currently, however, have no plans to include non-human receptors in AirToxScreen. 1.8.5. Endpoints and Measures: Results of AirToxScreen AirToxScreen reports estimated cancer risks and noncancer hazards attributed to modeled sources. Key measures of cancer risk developed for AirToxScreen include:  upper-bound estimated lifetime individual cancer risk; and  estimated numbers of people within specified risk ranges (e.g., number of individuals with estimated long-term cancer risk of 1-in-1 million or greater or less than 100-in-1 million). For noncancer effects, the key measures presented in AirToxScreen are hazard indexes summed across all air toxics modeled for the respiratory system. Other target organs and systems are also shown. AirToxScreen characterizes cancer risk and potential noncancer effects based on estimates of inhalation exposure concentrations determined at the census tract level. This approach is used only to determine geographic patterns of risks within counties, and not to pinpoint specific risk values for each census tract. We are reasonably confident that the patterns (i.e., relatively higher levels of risk within a county) represent actual differences in overall average population risks within the county. We are less confident that the assessment pinpoints the exact locations where higher risks exist, or that the assessment captures the highest risks in a county. EPA provides the risk information at the census tract level rather than just the county level, however, because the county results are less informative (in that they show a AirToxScreen 2018 Documentation 12 single risk number to represent each county). Information on variability of risk within each county would be lost if tract-level estimates were not provided. This approach is consistent with the purpose of AirToxScreen, which is to provide a means to inform both national and more localized efforts to collect air toxics information and to characterize emissions (e.g., to help prioritize air toxics and areas of interest for more refined data collection such as emissions testing or monitoring). Nevertheless, the assumptions made in allocating mobile and nonpoint source emissions within counties can result in significant uncertainty in estimating risk levels, even though general spatial patterns are reasonably accurate. 1.9. Model Design Consistent with the general approach for air toxics risk assessment described in Section 1.7 and illustrated in Figure 1-3, the analysis phase of AirToxScreen includes two main components: estimating exposure and estimating toxicity. The outputs of these analyses are used in the third phase, risk characterization, which produces health-risk estimates that can be used to inform research or risk management. These two phases (analysis and risk characterization) represent the “core” of EPA’s assessment activities associated with AirToxScreen. This set of activities is referred to here as the “AirToxScreen risk assessment process.” The AirToxScreen risk assessment process can be characterized by four main steps: compiling the nationwide inventory of emissions from outdoor sources; estimating nationwide ambient outdoor concentrations of the emitted air toxics; estimating population exposures to these air toxics via inhalation; and characterizing potential health risks associated with these inhalation exposures. The fourth step (risk characterization) also requires that quantitative dose-response or other toxicity values be identified for each air toxic included in the assessment. These values are taken from those developed by other EPA and non-EPA programs. Although this step does not require a “new” quantitative dose-response assessment to be conducted as part of AirToxScreen, it does require that we make important scientific and policy decisions regarding the appropriate values to use in AirToxScreen. Because these decisions are critical to the risk results, the identification of appropriate dose-response values is also described in this TSD in Section 5. The AirToxScreen risk assessment process is illustrated in Figure 1-5. The development of the emission inventory, air quality modeling, inhalation exposure modeling and risk characterization must be conducted sequentially – completing each step requires outputs from the previous step, and toxicity values are required to carry out the risk-characterization calculations. Cancer risks and the potential for noncancer health effects are estimated using available information on health effects of air toxics, risk-assessment and risk-characterization guidelines, and estimated population exposures. Each of these five components is described briefly here and explained in detail in the remainder of this document:  Section 2 explains the source types and air toxics included in the AirToxScreen emissions inventory. It also describes the processes we carried out to prepare the emissions for the air quality models. AirToxScreen 2018 Documentation 13  Section 3 discusses the models and procedures used to estimate ambient concentrations of air toxics, with links and references to technical manuals and other detailed documentation for the models used for AirToxScreen. Compile Nationwide Identify Toxicity Emission Inventory Values (Section 2) (Section 5) Conduct Air National Air Quality Modeling (Section 3) Inventory Ambient Model Inhalation Exposures (Section 4) Concentrations Conduct Risk Exposures Characterization (Section 6) Cancer Risks, Chronic Noncancer Figure 1-5. The AirToxScreen risk assessment process and corresponding sections of this TSD  Section 4 explains the processes used to estimate population-level exposure to outdoor ambient levels of air toxics, accounting for information on activities and other characteristics that can affect inhalation exposures.  Section 5 discusses the dose-response values used for AirToxScreen, the sources from which these values are obtained and assumptions made specific to AirToxScreen.  Section 6 provides the calculations used to estimate cancer risk and potential noncancer hazard.  Section 7 describes the uncertainties and limitations associated with the AirToxScreen process that must be considered when interpreting AirToxScreen results. As noted at the beginning of this section, this document is intended to serve as a resource accompanying the most recent national-scale assessment. Accordingly, although the following sections contain information on the AirToxScreen process that are generally applicable to all previous AirToxScreen 2018 Documentation 14 assessments, references to specific technical processes and supporting details typically emphasize what we did for the current version of AirToxScreen. 1.9.1. The Strengths and Limitations of the Model Design EPA developed AirToxScreen to inform both national and localized efforts to characterize air toxics emissions and health risks (e.g., prioritize air toxics or areas of interest for monitoring and community assessments). Because of this targeted objective, tools other than AirToxScreen may be more appropriate for assessing health risks outside the specific purpose of AirToxScreen (e.g., for evaluating risks from either a broader or more specific perspective). To further define and clarify what AirToxScreen should not be used for, this section contains descriptions of some of the important data and results that are not included in AirToxScreen:  AirToxScreen does not include information that applies to specific locations. The assessment focuses on variations in air concentration, exposure and risk among geographic areas such as census tracts, counties and states. All questions asked, therefore, must focus on the variations among these geographic areas (census tracts, counties, etc.). Moreover, as previously mentioned, results are far more uncertain at the census tract level than for larger geographic areas such as states or regions. (Section 7 contains discussions on the higher uncertainty at small geographic scales such as census tracts.)  AirToxScreen does not include data appropriate for addressing epidemiological questions such as the relationship between cancer risks or noncancer health effects and proximity of residences to point sources, roadways and other sources of air toxics emissions.  The results do not include impacts from sources in Canada or Mexico other than as general background sources. Thus, the results for states bordering these countries do not comprehensively reflect sources of transported emissions that could be significant.  AirToxScreen does not include results for individuals. Within a census tract, all individuals are assigned the same ambient air concentration, chosen to represent a typical ambient air concentration. Similarly, the exposure assessment uses activity patterns that do not fully reflect the actual variations among individuals.  The results do not include exposures and risk from all compounds. For example, of the 181 air toxics included in AirToxScreen, only 127 air toxics have been assigned dose-response values. In EPA’s judgment, the remaining air toxics do not have adequate data to quantitatively assess their impacts on health. Therefore, they do not contribute to the aggregate cancer risk or target-organ-specific hazard indexes estimated in AirToxScreen. Of note, the assessment does not quantify cancer risk from diesel PM, although EPA has concluded that the general population is exposed to levels close to or overlapping with levels that have been linked to increased cancer risk in epidemiology studies. AirToxScreen, however, does quantify noncancer effects of diesel PM.  Other than lead, which is both a CAP and a HAP, the results do not include the air pollutants, known as CAPs (particulate matter, ground-level ozone, carbon monoxide, sulfur oxides, nitrogen oxides), for which the CAA requires EPA to set National Ambient Air Quality Standards (other than CAP impacts on secondary formation of formaldehyde, acetaldehyde and acrolein). AirToxScreen 2018 Documentation 15  The results do not reflect all pathways of potential exposure. The assessment includes risks only from direct inhalation of the emitted air toxics compounds. It does not consider air toxics compounds that may deposit onto soil, water and food and subsequently enter the body through ingestion or skin contact.  The assessment results reflect exposure at outdoor, indoor and in-vehicle locations, but only to compounds released into the outdoor air. The assessment does not include exposure to air toxics emitted indoors, such as those from stoves, those that out-gas from building materials or those from evaporative benzene emissions from cars in attached garages. The assessment also does not consider toxics released directly to water and soil.  The assessment does not fully reflect variation in background ambient air concentrations. Background ambient air concentrations are average values over broad geographic regions.  The assessment may not capture all sources that have episodic emissions (e.g., facilities with short- term deviations in emissions resulting from startups, shutdowns, malfunctions and upsets). Where available, episodic emission information is used (e.g., for electricity generating units). In the absence of additional data, we assume emission rates are uniform throughout the year.  Short-term (acute) exposures and risks are not included in AirToxScreen.  Atmospheric transformation and losses from the air by deposition are not accounted for in AirToxScreen air toxics that are not modeled in CMAQ.  The evaluations to date have not assessed ecological effects, given the complexity of the varied ecosystems across the vast area covered by AirToxScreen. AirToxScreen 2018 Documentation 16 2. Emissions The systematic compilation of a detailed, nationwide inventory of air toxics emissions is the first major step in the AirToxScreen risk assessment process. This section contains descriptions of the emissions used for AirToxScreen. Section 2.1 describes the emissions data sources and preparation of the emissions used in AirToxScreen. Section 2.2 discusses the processing of emissions for input into the CMAQ model, and Section 2.3 discusses the processing for input into the AERMOD model. 2.1. Sources of Emissions AirToxScreen is intended to address outdoor emissions of all hazardous air Sometimes “air toxics” and “HAPs” are used interchangeably. pollutants (HAPs) and diesel particulate In this document, however, “air toxics” refers to the HAPs that EPA is required to control under Section 112 of the 1990 Clean matter (PM), together called “air toxics” Air Act (EPA 2015h) plus diesel PM. The 1990 Clean Air Act in this document. To model air toxics, Amendments required EPA to control HAPs (EPA 2018) and emissions of both air toxics and criteria provided for revisions to be made to that list. Currently, the list air pollutants (CAPs, including CAP includes 188 HAPs. Diesel PM is not a HAP; however, it has precursors such as ammonia and volatile been included in AirToxScreen. Some evidence indicates that organic compounds) are used to address localized high lifetime cancer risks are associated with the chemical interactions that occur exposure to diesel PM. However, EPA currently does not have across all pollutants. sufficient evidence to develop a unit risk estimate for it. Therefore, the potential adverse noncancer effects associated AirToxScreen combines modeling from with diesel PM are estimated in AirToxScreen (using an CMAQ and AERMOD for the contiguous Integrated Risk Information System reference concentration), United States. CMAQ multipollutant but its cancer risks are not. modeling addresses all sources in the National Emissions Inventory (NEI) for CAPs and about 52 air toxics including diesel particulate matter. Emissions from outside the United States are represented by CMAQ boundary conditions as discussed in Section 3.3.2. For the remaining “non-CMAQ” HAPs and non-CMAQ parts of the modeling domain (i.e., Alaska, Hawaii, Puerto Rico and the U.S. Virgin Islands), only AERMOD is used. For these pollutants and geographic regions, spatially uniform background concentrations based on remote concentrations are added to the AERMOD-modeled data to represent influences from transport and emissions outside the modeling domain (Section 3.6.1). AERMOD modeling addresses all pollutants covered by AirToxScreen and all anthropogenic sources except prescribed and agricultural burning. The main source of the emissions data for the CAPs and HAPs modeled for AirToxScreen is the National Emissions Inventory, or NEI. The NEI is a comprehensive and detailed estimate of air emissions of CAPs and HAPs from all air emissions sources in the United States, including the territories of Puerto Rico and the U.S. Virgin Islands, and offshore sources and commercial marine vessels (CMVs) in federal waters. A complete NEI, consisting of point stationary sources, nonpoint sources, mobile sources and fires, is prepared every 3 years by EPA. It is based primarily upon emission estimates and emission model inputs provided by S/L/T air agencies for sources in their jurisdictions, supplemented by data developed by EPA. These data are submitted electronically to the Emissions Inventory System (EIS). CAPs are required under EPA’s Air Emissions Reporting Requirements (AERR). HAPs are submitted voluntarily. Lead is both a HAP and a CAP, so it must be submitted under the AERR. Currently, states are required to report facilities with lead emissions greater than or equal to 0.5 tons per year (TPY). In addition to CAPs and HAPs, the NEI includes speciated particulate matter (PM) and diesel PM. The NEI diesel PM is computed AirToxScreen 2018 Documentation 17 as the PM10 emissions for on-road and nonroad engines burning diesel or residual oil fuels. Although stationary engines also can burn diesel fuel, only mobile source sectors are used for estimating diesel PM emissions. To build as complete an NEI as possible, EPA augments the S/L/T-submitted data using various sources of information, including the Toxics Release Inventory (TRI), and applies HAP-to-CAP emission-factor ratios to CAP emissions reported by S/L/T. Table 2-1 contains a summary of the sources of emissions data in the NEI. More detailed information on all data sources can be found in 2017 NEI documentation and the 2018 Emissions Modeling Platform documentation. Table 2-1. Summary of emissions sources in the 2018 NEI Source Description Most stationary point-source HAP data were submitted voluntarily by S/L/T. For some point sources, EPA gap-filled HAPs. Sources of data included: TRI data for 2018, Stationary augmentation of HAPs using emission-factor ratios (of HAP to CAP) applied to S/L/T-reported CAP point emissions, and the 2017 NEI. 2018 emissions based on estimates used in the 2017 NEI adjusted using Terminal Area Forecast (TAF) data. No adjustments were made to military aircraft from 2017. EPA estimates used the Federal Aviation Administration (FAA) Emission Dispersion Modeling System using landing and takeoff (LTO) information from FAA databases and updated where S/L/T-provided LTO. For Point airports smaller airports (general aviation) without detailed aircraft-specific activity data, straight emission factors were used. Lead emissions were estimated based on per-LTO emissions factors, assumptions about lead content in the fuel and lead-retention rates in the piston engines and oil. For some airports, estimates were provided by S/L/T. NEI has approximately 20,000 airports (including heliports and seaplanes); all are inventoried as point sources. In addition to LTOs, EPA’s emissions estimates for airports included emissions of ground support equipment. 2018 emissions based on estimates used in the 2017 NEI adjusted using annual energy outlook (AEO) and additional factors supplied by ERTAC. The 2017 NEI includes non-zero emissions estimates for nearly 1000 rail yards. EPA emission estimates are associated with the operation of switcher engines at each Class I rail yard. EPA estimates were developed by the Eastern Regional Point rail Technical Advisory Committee’s (ERTAC) rail group using a “top-down” approach that apportions yards 2017 national fuel use data to rail yards and used national fleet-wide information to create weighted average emission factors. S/L/T submitted point rail yard emissions were given priority over the ERTAC estimates when present. HAP emissions were estimated by applying fractions to the VOC or PM estimates. There are also railyard emissions in the nonpoint inventory (see locomotives). Includes many different source stationary source categories that are generally too ubiquitous to be inventoried as point sources and are therefore estimated at the county level. Examples of these sources include residential heating, consumer and commercial product usage, commercial Stationary cooking, oil and gas production, and industrial, commercial and institutional fuel combustion nonpoint (where not in the point inventory). Emission estimates for these are developed by EPA and/or submitted by S/L/T. Where S/L/T submitted CAPs but not HAPs, missing HAP emissions were augmented. 2018 oil and gas production emissions based on 2017 NEI adjusted using U.S. Energy Information Administration and Railroad Commission of Texas historical production data. The remaining 2018 nonpoint source emissions are based on 2017 NEI without adjustment. AirToxScreen 2018 Documentation 18 Source Description Based on Biogenic Emission Inventory System (BEIS3.7) using 2018 meteorology from the Biogenics Weather Research Forecasting Model (WRF). Gridded emissions were used in AirToxScreen and summed to annual county-level estimates for the NEI. Includes VOC, NOX and three HAPs: formaldehyde, acetaldehyde and methanol. 2018 emisions were projected from 2017 using annual energy outlook (AEO) and additional factors supplied by ERTAC. Emissions at county-level resolution for Class I line haul, Class II/III line haul, passenger, commuter and rail yards. ERTAC’s rail group developed EPA estimates for Class I Locomotives line haul and Class II/III line haul. S/L/T also submitted data for locomotives. All passenger, commuter and county-level rail yards estimates were from S/L/T. EPA rail yard emissions were included as point-source rail yards as described above. HAPs were estimated by applying toxic fractions to the VOC or PM estimates. 2018 emissions estimates based on 2017 NEI adjusted using satellite-based Automatic Identification System (AIS) activity data from the US Coast Guard. Emissions from category 1 and category 2 (C1/C2) and category 3 (C3) marine vessels at ports or underway. C1/C2 includes CMVs fishing boats, ferries and tugboats and is assumed to use diesel fuel; C3 includes oceangoing vessels and large ships and are assumed to use residual fuel. Emission estimates were developed using 1) activity data (kilowatt hours or kW), 2) engine operating load factors and 3) emission factors and HAP speciation profiles. This “bottom up” approach was used for the first time for the 2014 NEI. EPA’s CMV estimates use AIS activity data. Except for California, on-road emissions were generated using the SMOKE-MOVES emissions modeling framework, which leverages emission factors generated by the latest released version of the Motor Vehicle Emissions Simulator (MOVES) (MOVES2014b, code version: 20180726, database version: movesdb 20181022); county and SCC-specific activity data and hourly meteorological data. These models used state- or EPA-provided input details specific to each On-road county. Activity data were projected from 2017 to 2018 using factors developed using data from Federal Highway Administration and state departments of transportation. California’s emissions were developed via their EMFAC on-road model, but VOC HAPs were speciated from California- reported VOCs consistent with the MOVES2014b speciation. EPA added DIESEL-PM10 and DIESEL- PM25 for all diesel fuel SCCs, and they were set equal to the PM10 and PM2.5 emissions from these diesel SCCs. 2018 was interpolated between 2017 and 2023 emissions from EMFAC2017. Except for California, EPA estimated these emissions using the MOVES2014b model. Several S/L Nonroad, provided inputs to the model. MOVES2014b also replaces toxic emission estimates for nonroad excluding previously generated using the National Mobile Inventory Model (NMIM), which was used for airports, 2014 and earlier NEIs. MOVES2014b was used for all states other than California, which uses their locomotives own model. EPA added DIESEL-PM10 and DIESEL-PM25 for all diesel fuel SCCs, and they were set and CMVs equal to the PM10 and PM2.5 emissions from exhaust emissions in these diesel SCCs. . California emissions were interpolated to 2018 between California-provided 2017 and 2023 emissions. For purposes of AirToxScreen, fires include agricultural burning, prescribed burning and wildfires. EPA estimated agricultural burning (included in stationary nonpoint) using remote-sensing data, crop-usage maps and emission factors as daily point estimates. Only EPA-developed agricultural Fires fire data are used in this study, thus 2017 NEI state submissions are not included. EPA estimates were modeled as daily point estimatesEPA developed day- and location-specific prescribed burning and wildfire emissions via the SMARTFIRE2 system (which includes the BlueSky modeling framework) with inputs from state agencies where available. Georgia and Washington state submitted emission estimates (day and location specific). AirToxScreen 2018 Documentation 19 2.1.1. Pollutants and Pollutant Groups AirToxScreen air quality modeling requires emissions of criteria air pollutants (CAPs) and their precursors in addition to HAP and diesel PM. We also need to aggregate the emissions of certain NEI pollutants to match the AirToxScreen pollutants. This section discusses the pollutants and pollutant groups in the NEI used in AirToxScreen and the aggregation needed to generate the AirToxScreen pollutants. In CMAQ, we model CAPs and precursors and about 52 air toxics, including diesel particulate matter. Table 2-2 shows the specific air toxics used in CMAQ. For AERMOD, we model nearly all HAPs covered by the NEI; those that are not modeled are due to lack of emissions or risk considerations. Table B-1 in Appendix B provides more detail about each HAP excluded. For AirToxScreen, we aggregated NEI pollutants that can be reported as either a group or as specific individual pollutants belonging to the group into pollutant groups for the AirToxScreen modeling and results. This aggregation was done in the emissions modeling process, prior to the air quality modeling. Table 2-2 lists the groups. For example, individual glycol ethers are grouped into the single AirToxScreen HAP “glycol ethers.” The spreadsheet file “AirToxScreen_Pollutants.xlsx” within the Supplemental Data folder accompanying this TSD shows the individual HAPs and groups used in AERMOD and CMAQ. The following subsections give more details about some specific AirToxScreen pollutant groups. Table 2-2. Pollutant groups Group Chromium Vi (Hexavalent) Cresol cresylic acid (mixed isomers) Cyanide compounds Glycol ethers Nickel compounds PAHPOM Polychlorinated biphenyls (aroclors) Xylenes (mixed isomers) 2.1.1.1. Assignment of PAHs into PAH modeling groups In AirToxScreen, Polycylic aromatic hydrocarbons and polycylic organic matter are represented by a lumped group, “PAHPOM”. The individual compounds in the PAHPOM group have widely varying risks. As a result, we modeled the PAHPOM in separate risk-based groups based on their unit risk estimate (URE). For AirToxScreen, we summed the concentrations and risks across all PAHPOM risk groups. For the tabular emission summaries on the AirToxScreen website (https://www.epa.gov/AirToxScreen), the individual PAHPOM compounds (i.e., in Table 2-3) are provided. The PAH groups are based on the groups established for AirToxScreen and are listed below along with the individual PAHs assigned to each. Note that two pollutants representing polycylic organic matter in the NEI are unspeciated: pollutant code 250 (PAH/POM–Unspecified) and pollutant code 130498292 (PAH, total). These are assigned to PAH_880E5. AirToxScreen 2018 Documentation 20 Table 2-3. PAH/POM pollutants group PAH Group NEI Pollutant Code NEI Pollutant Description URE 1/(μg/m3) PAH_000E0 120127 Anthracene PAH_000E0 85018 Phenanthrene PAH_000E0 129000 Pyrene PAH_101E2 56495 3-Methylcholanthrene 0.01 PAH_114E1 57976 7,12- Dimethylbenz[a]Anthracene 0.114 PAH_176E2 189640 Dibenzo[a,h]Pyrene 9.6E-03 PAH_176E2 189559 Dibenzo[a,i]Pyrene 9.6E-03 PAH_176E2 191300 Dibenzo[a,l]Pyrene 9.6E-03 BENZOAPYRNE 50328 (see Note 1) Benzo[a]Pyrene 9.6E-04 PAH_176E3 192654 Dibenzo[a,e]Pyrene 9.6E-04 PAH_176E3 53703 Dibenzo[a,h]Anthracene 9.6E-04 PAH_176E3 194592 7H-Dibenzo[c,g]carbazole 9.6E-04 PAH_176E3 3697243 5-Methylchrysene 9.6E-04 PAH_176E3 41637905 Methylchrysene 9.6E-04 PAH_176E4 56553 Benz[a]Anthracene 9.6E-05 PAH_176E4 205992 Benzo[b]Fluoranthene 9.6E-05 PAH_176E4 205823 Benzo[j]fluoranthene 9.6E-05 PAH_176E4 226368 Dibenz[a,h]acridine 9.6E-05 PAH_176E4 224420 Dibenzo[a,j]Acridine 9.6E-05 PAH_176E4 193395 Indeno[1,2,3-c,d]Pyrene 9.6E-05 PAH_176E4 5522430 1-Nitropyrene 9.6E-05 PAH_176E5 207089 Benzo[k]Fluoranthene 9.6E-06 PAH_176E5 86748 Carbazole 9.6E-06 PAH_176E5 218019 (see Note 2) Chrysene 9.6E-06 PAH_192E3 8007452 Coal Tar 9.9E-04 PAH_880E5 83329 Acenaphthene 4.8E-05 PAH_880E5 208968 Acenaphthylene 4.8E-05 PAH_880E5 203338 Benzo(a)Fluoranthene 4.8E-05 PAH_880E5 195197 Benzo(c)phenanthrene 4.8E-05 PAH_880E5 192972 Benzo[e]Pyrene 4.8E-05 PAH_880E5 203123 Benzo(g,h,i)Fluoranthene 4.8E-05 PAH_880E5 191242 Benzo[g,h,i,]Perylene 4.8E-05 PAH_880E5 56832736 Benzofluoranthenes 4.8E-05 PAH_880E5 91587 2-Chloronaphthalene 4.8E-05 PAH_880E5 284 (see Note 3) Extractable Organic Matter (EOM) 4.8E-05 PAH_880E5 206440 Fluoranthene 4.8E-05 PAH_880E5 86737 Fluorene 4.8E-05 AirToxScreen 2018 Documentation 21 PAH Group NEI Pollutant Code NEI Pollutant Description URE 1/(μg/m3) PAH_880E5 779022 9-Methyl Anthracene 4.8E-05 PAH_880E5 26914181 Methylanthracene 4.8E-05 PAH_880E5 2422799 12- Methylbenz(a)Anthracene 4.8E-05 PAH_880E5 65357699 Methylbenzopyrene 4.8E-05 PAH_880E5 90120 1-Methylnaphthalene 4.8E-05 PAH_880E5 91576 2-Methylnaphthalene 4.8E-05 PAH_880E5 832699 1-Methylphenanthrene 4.8E-05 PAH_880E5 2531842 2-Methylphenanthrene 4.8E-05 PAH_880E5 2381217 1-Methylpyrene 4.8E-05 PAH_880E5 130498292 PAH, total 4.8E-05 PAH_880E5 198550 Perylene 4.8E-05 PAH_880E5 250 PAH/POM – Unspecified 4.8E-05 Polycyclic aromatic PAH_880E5 N590 compounds (includes 25 4.8E-05 specific compounds) Note 1: Benzo[a]pyrene is the only PAHPOM with an RfC (2E-6 µg/m3). Therefore, chronic noncacer HI values for the PAHPOM group are based solely on benzo[a]pyrene concentrations. Note 2: Even though chrysene URE is 9.6E-7, put into 9.6E-6 group (there is no lower risk group other than 0 so this is conservative). Note 3: pollutant code retired in 2016 2.1.1.2. Metal groups AirToxScreen includes metal compound groups consistent with metal emissions Example: Adjusting Emissions for Chromium VI Compounds in the NEI. Metal emissions in the NEI Chromic acid (VI) (H2CrO4) has a molecular weight of about represent only the mass of the metal 118.01. Chromium, with an atomic mass of 52, is the toxic with a few exceptions for specific element of interest in this metal compound. Emissions reported compounds of hexavalent chromium in NEI are therefore multiplied by 0.4406 (i.e., 52 / 118.01), and (chromium VI) and nickel of known the resulting emission rate is used in AirToxScreen modeling. composition. Prior to modeling, we applied factors that convert the emissions of specific metal compounds to the portion of the compound that is metal. Table 2-4 shows the HAPs that have metal speciation factors other than 1. The three nickel compounds and three chromium VI compounds in the NEI are shown in the table below with the corresponding adjustment factors to compute the emissions that account for just the metal portion of the compound. Note that after applying the adjustments, the chromium VI compounds are grouped into chromium VI and the nickel compounds are grouped into nickel. These are generally small in mass compared to the metal-only pollutants (nickel and chromium VI) and are only present for stationary sources. AirToxScreen 2018 Documentation 22 Table 2-4. Metal speciation factors for NEI metal compounds Description AirToxScreen Pollutant Group pollutant_cd (CAS) Metal Speciation Factor * Nickel oxide NICKEL COMPOUNDS 1313991 0.7858 Chromium trioxide CHROMIUM VI (HEXAVALENT) 1333820 0.52 Chromic acid (VI) CHROMIUM VI (HEXAVALENT) 7738945 0.4406 * Metal speciation factor is the ratio of the molecular weight of the metallic element to the molecular weight of the compound. 2.1.1.3. Diesel PM Diesel PM is neither a CAP nor HAP as defined by Section 112 of the CAA, however it was identified as a mobile source air toxic in EPA’s 2007 rule, “Control of Hazardous Air Pollutants From Mobile Sources final rule” (EPA 2007a). Prior to the 2014 NEI, it was generated separately for AirToxScreen modeling from the NEI and was not included as a separate NEI pollutant. However, starting with the 2014 NEI, diesel PM emissions are included in the NEI, as discussed above. The NEI-generated diesel PM emissions from the mobile-source, engine-exhaust PM10 emissions were used for engines burning diesel or residual-oil fuels. These sources include on-road, nonroad, point-airport-ground support equipment, point-locomotives, nonpoint locomotives, and all PM from diesel or residual-oil-fueled nonpoint CMVs. Diesel PM emissions were set equal to PM10 emissions for these engines. Although stationary engines also can burn diesel fuel, only mobile-related diesel engine SCCs were used. 2.1.1.4. Pollutant information file The spreadsheet file “AirToxScreen_Pollutants.xlsx” within the Supplemental Data folder accompanying this TSD includes a crosswalk that contains NEI pollutant codes/descriptions, AirToxScreen group information, CMAQ names, metal adjustment factors, URE, RfC and target organ information. 2.1.2. Emissions Categorization: NEI and AirToxScreen As explained on the NEI website, the NEI includes five data categories: point, nonpoint (formerly called “stationary area”), nonroad mobile, on-road mobile, and events consisting of wild and prescribed fires. NEI summaries are generally provided by sectors and tiers, which describe the type of emission source (e.g., industrial processes – oil and gas production). Some sectors and tiers cut across data categories since stationary sources are inventoried as both point and nonpoint. For example, the NEI sector “Fuel Combustion – Commercial/Institutional – Oil” results from large institutions inventoried as point sources (e.g., large universities with onsite steam plants) as well as commercial/institutional entities that are small and ubiquitous in nature, so their emissions are inventoried as county sums. AirToxScreen summaries are provided by AirToxScreen broad summary categories and by more detailed source groups. The broad AirToxScreen summary categories are point, nonpoint, on-road, nonroad, fires, biogenics and secondary. Some of these categories are named the same as the NEI data categories, but they are not identical. For example, the NEI nonpoint category includes CMVs and locomotives, while the AirToxScreen category does not. Similarly, the AirToxScreen nonroad category includes airports, CMVs and locomotives, while the NEI category does not. Table 2-5 contains comparisons between the NEI data categories and the AirToxScreen categories. “Secondary” is not included in AirToxScreen 2018 Documentation 23 Table 2-5 since it not a primary emissions category covered in the NEI, but rather a result of atmospheric chemistry from the modeled emissions of CAPS and HAPs. Table 2-5. Map of NEI data categories to AirToxScreen categories NEI Data Category AirToxScreen Category (Reflecting AirToxScreen Summary Results) Point Point1 Emissions estimates for sources that are Same as NEI point except: individually inventoried and usually located at a Excludes portable sources, which are not modeled in either fixed, stationary location (although portable CMAQ or AERMOD because no geographic information sources such as some asphalt- or rock-crushing other than the state code is included. operations are also included). Point sources include Excludes airports and railyards, which are nonroad mobile. large industrial facilities and electric power plants but also increasingly include many smaller industrial and commercial facilities, such as dry cleaners and gas stations, that had traditionally been included as nonpoint sources. The choice of whether these smaller sources are estimated individually and included as point sources or inventoried as a nonpoint source aggregated to county or tribal areas is determined by the separate S/L/T air agency. Nonpoint Nonpoint Sources that individually are too small in Same as NEI nonpoint except excludes locomotive, CMV, magnitude or too numerous to inventory as biogenic emissions and agricultural fires. individual point sources and that can often be estimated more accurately as a single aggregate source for a county or tribal area. Examples are residential heating and consumer solvent use. Agricultural, CMVs and locomotive emissions are included. Biogenic emissions that come from vegetation are also included. On-road On-road Emissions estimates for mobile sources, such as Same as NEI on-road. cars, trucks and buses. EPA’s MOVES2014b model currently generates these estimates (except in California, which uses different models). Nonroad Nonroad Emissions estimates for nonroad equipment such Same as NEI nonroad, but the AirToxScreen nonroad also as lawn and garden equipment, agricultural, includes CMVs, locomotives, aircraft engine emissions construction, industrial and commercial equipment occurring during LTOs, and the ground support equipment and recreational equipment. EPA’s MOVES2014b and auxiliary power units associated with the aircraft. model generates these estimates (except in California, which uses different models). AirToxScreen 2018 Documentation 24 NEI Data Category AirToxScreen Category (Reflecting AirToxScreen Summary Results) Event Fires Prescribed and wildfire emissions computed as Wildfires, prescribed burning and agricultural burning. day- and location-specific events These are modeled in CMAQ but not AERMOD. Wildfires and prescribed burning are generated via the SMARTFIRE2 model at specific geographic coordinates for each day, and are assigned to 12-km grid cells for input into CMAQ. Agricultural burning estimates are in the NEI as county-level emissions, but EPA-derived data are developed as day- and location-specific emissions, and S/L/T-submitted data are county level but prepared for CMAQ as point sources with day-specific emissions. Biogenic Emissions Emissions of formaldehyde, acetaldehyde and methanol from vegetation (trees, plants and soils) computed from the Biogenic Emission Inventory System within CMAQ. These are gridded to 12-km cells and modeled in CMAQ but are not modeled in AERMOD. 1In results presented online for assessments for the 2002 and early NATA inventories, point sources were divided into major sources and area sources; these were sometimes referred to as stationary sources. Major sources are defined in the CAA as stationary sources that have the potential to emit either at least 10 TPY of a HAP or at least 25 TPY of any combination of HAPs. Area sources are stationary sources for which the locations are known but that emit at levels below the major source emissions thresholds. This terminology is not used in AirToxScreen, and stationary-source emissions are referred to only as point-source or nonpoint-source emissions. Point sources in the AirToxScreen results refer to those sources, including smaller sources, for which a specific location for their emissions is identified by latitude and longitude descriptions, and nonpoint sources are those stationary sources that are not point sources. 2.1.3. Differences Between the current NEI and Emissions Used for AirToxScreen Although 2018 NEI is the main basis of the emissions fed into the air quality models for AirToxScreen, there were several differences between the 2018 NEI and emissions data used for the AirToxScreen modeling. State, local, and tribal agencies reviewed emissions information along with scaled preliminary risk information, which were estimated by taking pollutant-specific preliminary draft AERMOD 2017 results and multiplying by the ratio of the 2018 emissions over the 2017 emissions. EPA reviewed the comments and changes and incorprated the accepted changes into the modeling platform. Information about the S/L/T changes that were applied is included in the supplemental data. Ethylene oxide emissions from commercial sterilizers in the NEI were replaced with EPA-estimated values using the same methodology developed to model emissions for a forthcoming proposal to amend the National Emissions Standard for Hazardous Air Pollutants (NESHAP) for Commercial Sterilizers. We adjusted to year 2018 estimates based on control equipment operating in year 2018 and ethylene oxide throughput in year 2017. Much of the information used in this methodology was sourced from the responses to the December 2019 information and September 2021 collection requests for this industry. AirToxScreen 2018 Documentation 25 A table of facilities for which ethylene oxide emissions were based on this methodology is listed in the supplemental data. Differences that result from differences in emissions processing (which reflect the specific role and function of the resulting inventory within the context of the AirToxScreen risk assessment process) are more accurately described as post-processing procedures rather than substantive changes. Release parameter changes were made to the SMOKE flat file to fill in missing data or change out-of- range stack parameters. Data reporters of point sources must provide height, diameter, temperature and either velocity or flowrate, but they do not have to provide fugitive parameters. Prior to emissions processing, we default missing or out-of-range stack or fugitive parameters and compute the velocity from the flowrate (if velocity is not provided). We chose to do the defaulting prior to the emissions processing and include the defaulted parameters directly into the FF10 prior to input into SMOKE. This is done for two reasons: 1) to provide better transparency in the FF10 files with respect to the data used in the model, and 2) to ensure that emission inputs are consistent across CMAQ and AERMOD models, since both use the FF10 as the starting point. The out-of-range parameters were chosen to be consistent with the range checks used in the Emissions Inventory System (EIS). The fugitive defaults were consistent with what has been used in previous NATAs. Table 2-6 shows the changes made and why. Even though SMOKE does not use the fugitive release point parameters, they are included in the table to make it complete. Table 2-6. Release parameter defaults/changes to the FF10 inventory files for point sources Field Existing Value New Value Conditions/Notes1 For point sources with stack releases (ERPtype NOT equal to “1”) stkhgt missing use pstk2 or global defaults2 None stkdiam missing use pstk2 or global defaults3 None stkvel missing calculate from stkflow and vel = 4*stkflow/(pi*stkdiam^2) stkdiam if not missing; If the flow and diam are missing such otherwise reference by SCC that you cannot compute, use new from pstk2 or global defaults3 value based pstk or global defaults stktemp missing use pstk2 or global defaults3 None stkhgt Outside use minimum value or Less than 1 ft (0.3048 m) or greater than SMOKE range maximum value in feet 1300 ft (396 m) stkdiam Outside use minimum value or Less than 0.001 ft (0.0003048 m) or SMOKE range maximum value in ft greater than 300 ft (91.4 m) stkvel Outside use minimum value or Less than 0.001 ft/s (0.0003048 m/s) or SMOKE range maximum value in ft/s greater than 1000 ft/s (304.8 m/s) stktemp Outside use minimum value or Less than -30 F (-34.4 C or 248.15 K) or SMOKE range maximum value in F greater than 4000 F (2204.4 C or 2477.6 K) For fugitive release points (not used in CMAQ) fug_width_ydim missing 32.808 ft None fug_length_xdim missing 32.808 ft None fug_angle missing 0 None fug_height missing 10 ft fug_width_ydim and /or fug_length_xdim are missing AirToxScreen 2018 Documentation 26 Field Existing Value New Value Conditions/Notes1 fug_height missing 0 WHEN fug_width_ydim and fug_length_xdim are not missing and > 0 For coke ovens: any release point that emits coke oven emissions (pollutant code 140)–all pollutants at that release point are changed to the below stkhgt <126 ft 126 ft erptype NOT = “1” fug_height <126 ft 126 ft erptype = “1” fug_length_xdim <50 ft 50 ft erptype = “1” fug_width_ydim <50 ft 50 ft erptype = “1” Comments were put into the modeling file to indicate why a record was changed: ERPVelCompute – velocity computed from the flowrate provided in the inventory ERPHtRange – height in the inventory was out of range ERPDiamRange – diameter in the inventory was out of range ERPVelRange – velocity in the inventory or velocity calculated from the flowrate in the inventory was out of range ERPTempRange – temperature in the inventory was out of range ERPFugHeight0 – fugitive height in the inventory was set to 0 because the width and length were not missing ERPFugMissing – fugitive height, length and width were missing or fugitive length and/or width were missing. ERPCokeoven126 – fugitive or stack height of release point emitting coke oven emissions was less than 126 ft ERPCokeovenFug50 – fugitive length or width was less than 50 ft. Pstk provides default stack parameters and is provided with other SMOKE ancillary files (ge_dat directory) on the website. The pstk file is formatted: region_cd, scc, stkhgt (m), stkdiam (m), stktemp (K) and stkvel (m/s) Global defaults (converted to English): stkvel = 13.1234 ft; stktemp=72.05 F, stkdiam=0.6562 ft, stkhgt=9.8425 ft Out-of-range values exist because the flow range checks in EIS allow some velocities to be above or below the range, and we run the velocity check after computing the missing flowrates. Other emissions processing changes include:  For on-road emissions, there were differences in the HAP VOCs in California. These were due to changes in the post-processing approach to adjust California-submitted pollutants consistent with the MOVES2014ba speciation. Emissions for other HAPs, including metals and PAHs, were used as provided by California. The AirToxScreen inventory also includes a more refined set of SCCs that includes road type to support spatial allocation of county-level emissions to finer scales.  Sources with FIPS state-county codes ending in 777 (in-flight lead and asphalt plants that have no geographic coordinates) were removed from the inventory.  Nonpoint tribal data (FIPS beginning in 88) were not used in the modeling because spatial surrogates were not available and possible double counting would introduce uncertainty.  Other miscellaneous changes included air toxic name conversions, placing individual air toxics into groups, and similar transcription and phraseology conversions (e.g., to crosswalk the identity of an emitted air toxic to a substance with a quantitative dose-response value). 2.1.4. Overview of Differences in Emissions for CMAQ and AERMOD By design, there are differences in the sources of emissions used by CMAQ and AERMOD. Differences in the emissions inputs to the two models are due to design differences in how the models are run. The AirToxScreen 2018 Documentation 27 emissions input into AERMOD exclude AirToxScreen categories more appropriately addressed by CMAQ, namely biogenics and three types of fires: wildfires, prescribed burning and agricultural-field burning. Biogenic emissions are generated within CMAQ using the Biogenic Emission Inventory System (BEIS) model with hourly meteorological inputs to generate hourly gridded (12 km x 12 km) emissions of several photochemical-model species, including three HAPs: formaldehyde, acetaldehyde and methanol. This category of emissions is routinely part of CMAQ runs and is more appropriately modeled in CMAQ due to its broad spatial and refined temporal resolution and meteorological dependence. Wildfires and prescribed and agricultural burning are also included in the CMAQ run but not in AERMOD because CMAQ provides for in-line plume rise of fires to higher vertical layers based on the acres burned. These algorithms are also used for agricultural burning, which is grouped with the other fires to allow us to retain source attribution from the fire and biogenic CMAQ zero-out runs (although not between the different fire types) discussed in Section 3.3.1. Also, SLT Review changes for point sources included for AERMOD were not included in CMAQ modeling because the CMAQ modeling for the 2018 AirToxScreen preceded the SLT Review. In addition to differences in the sources of emissions used for the two models, there are also differences in the way we process or model the emissions. Emissions modeling transforms the emissions inventory into the format needed by the air quality model and provides the source characterization. The emission inventory primarily contains annual emissions represented as either point sources with locations specified by latitude and longitude or county-level sources specified by FIPS code (county code). In the inventory, large facilities are inventoried as point sources; while more ubiquitous sources, such as wood stoves, solvent use, and cars and trucks, are inventoried at the county level. Emissions modeling is performed in three main steps: spatial allocation, temporal allocation and speciation. Spatial allocation provides the models with the horizontal characterization of the emissions source. For example, emissions modeling provides AERMOD with locations for point sources and locations and spatial extent of sources emitted over large or small areas, and it provides CMAQ with gridded emissions. Vertical allocation of emissions is performed within the air quality models using stack parameter information and other model inputs. Temporal allocation produces hourly variation in emissions based on the monthly, day-of-week and/or diurnal variation associated with the specific type of source. Speciation takes the inventory pollutant and converts it to the pollutant used by the air quality model (or in subsequent processes). For example, the compound nickel oxide is converted to nickel because the risk information is only for the nickel portion of the compound. For AERMOD, sources also must be characterized using parameters that may not be in the inventory, such as release heights (which are not inventoried for mobile and nonpoint sources) and initial vertical dispersion (which is not in the inventory for any source). Emissions processing for CMAQ and AERMOD is described in Sections 2.2 and 2.3, respectively. However, Section 2.2 contains only a summary since a separate technical support document is available that describes the CMAQ emissions modeling in detail. Table 2-7 summarizes the spatial allocation differences between the models. As seen in Table 2-7, CMAQ uses 12-km horizontal resolution along with vertical resolution for point sources and fires, whereas AERMOD spatial resolution depends on the source category. Both models use hourly emissions, AirToxScreen 2018 Documentation 28 however CMAQ uses pollutant-specific hourly emissions whereas AERMOD uses the same hourly variation for all sources in a “run group,” as described in Section 2.3. Table 2-7. Differences in spatial characterization of sources between CMAQ and AERMOD Category NEI Resolution Spatial Approach for AERMOD Spatial Approach for CMAQ Point (excluding airports) Point Point – vertical stack and fugitive based on 12-km grid cells NEI information on emission-release point Airports Point Point – runways & 10 m2 areas consistent 12-km grid cells with NEI geographic coordinates Locomotives Point (railyards) Nonpoint – 12-km grid cells in the CONUS 12-km grid cells and domain, tract (non-CONUS) County/Shape Point – point fugitives CMVs, ports and County/Shape Shapes from the NEI; separate shapes 12-km grid cells underway used for CMV at ports versus underway On-road, nonroad County 12-km or 4-km grid cells, depending on 12-km grid cells equipment and other the category, in the CONUS domain; 9-km nonpoint grid cells for AK; 3-km grid cells for HI, PR and VI Agricultural burning and County Not modeled 12-km grid cells biogenic emissions Fires (prescribed and Point Not modeled 12-km grid cells wild) 2.2. Preparation of Emissions Inputs for CMAQ EPA routinely prepares emissions for photochemical grid models such as CMAQ by developing an emissions modeling platform, and the SMOKE modeling system is used as the primary emissions modeling tool. An emissions modeling platform includes the emission inventories, the ancillary data files (e.g., for speciation, temporal allocation and spatial allocation) and the approaches used to transform inventories for use in air quality modeling. The platform used for this study is described in depth in the technical support document found on the emissions modeling platforms website. For 2018 information, please see in particular: https://www.epa.gov/air-emissions-modeling/2018-emissions-modeling-platform-technical-support- document. Emissions inputs and ancillary data for speciation and for temporal and spatial allocation are available at the emissions modeling platform ftp site. A summary of the platform’s key features is below. 2.2.1. Sectors in the CMAQ AirToxScreen Platform For the purposes of preparing the CMAQ model-ready emissions, the NEI is split into finer-grained sectors used for emissions modeling. The significance of an emissions modeling or “platform sector” is that the data are run through all the SMOKE programs independently from the other sectors except the final merge program (Mrggrid). The sectors used for the AirToxScreen CMAQ platform are listed in Table 2-8. AirToxScreen 2018 Documentation 29 Table 2-8. Platform sectors for the 2018 emissions modeling platform Platform Sector: NEI Data abbreviation Category Description and Resolution of the Data Input to SMOKE EGU units: Point 2018 NEI point source EGUs, replaced with hourly Continuous Emissions ptegu Monitoring System (CEMS) values for NOX and SO2, and the remaining pollutants temporally allocated according to CEMS heat input where the units are matched to the NEI. Emissions for all sources not matched to CEMS data come from 2018 NEI point inventory. Annual resolution for sources not matched to CEMS data, hourly for CEMS sources. EGUs closed in 2018 are not part of the inventory. Point source oil and Point 2018 point sources that include oil and gas production emissions processes for gas: facilities with North American Industry Classification System (NAICS) codes pt_oilgas related to Oil and Gas Extraction, Natural Gas Distribution, Drilling Oil and Gas Wells, Support Activities for Oil and Gas Operations, Pipeline Transportation of Crude Oil, and Pipeline Transportation of Natural Gas. Includes U.S. offshore oil production. Production-related sources that did not have 2018 data were pulled forward from the 2017 NEI and adjusted to 2018. Annual resolution. Aircraft and ground Point 2017 NEI point source emissions from airports, including aircraft and airport support equipment: ground support emissions, adjusted to 2018 using Terminal Area Forecast airports (TAF) data. Airport-specific factors were used where available, state average factors were used for regional airports, and no change was made to military aircraft from 2017. Annual resolution. Remaining non-EGU Point All 2018 NEI point source records not matched to the airports, ptegu, or point: pt_oilgas sectors. Closures were reviewed and implemented based on the ptnonipm most recent submissions to the Emissions Inventory System (EIS). Added a unit from the Kennecott copper mine near Salt Lake City. Includes 2017 NEI rail yard emissions, adjusted to 2018 using same projection factors as the rail sector. Annual resolution. Agricultural: Nonpoint 2017 NEI nonpoint livestock emissions adjusted to 2018 using USDA survey livestock data. Livestock includes ammonia and other pollutants (except PM2.5). County and annual resolution. 2018 agricultural fertilizer ammonia emissions computed inline within CMAQ. Agricultural Fertilizer Nonpoint 2018 agricultural fertilizer ammonia emissions computed inline within CMAQ. Agricultural fires with Nonpoint Agricultural fire sources for year 2018 were developed by EPA as point and point resolution: day-specific emissions.1 Agricultural fires are in the nonpoint data category of ptagfire the NEI, but in the modeling platform, they are treated as day-specific point sources. Updated HAP-augmentation factors were applied. Area fugitive dust: Nonpoint PM10 and PM2.5 fugitive dust sources from the 2017 NEI nonpoint afdust inventory; including building construction, road construction, agricultural dust, and paved and unpaved road dust; with paved road dust adjusted to 2018 based on vehicle miles traveled (VMT). The emissions modeling system applies a transport fraction reduction and a zero-out based on 2018 gridded hourly meteorology (precipitation and snow/ice cover). Emissions are county and annual resolution. 1 Only EPA-developed agricultural fire data were included in this study; data submitted by states to the NEI were excluded. AirToxScreen 2018 Documentation 30 Platform Sector: NEI Data abbreviation Category Description and Resolution of the Data Input to SMOKE Biogenic: Nonpoint Year 2018 emissions from biogenic sources. These were left out of the beis CMAQ-ready merged emissions, in favor of inline biogenic emissions produced during the CMAQ model run itself. Version 3.7 of the Biogenic Emissions Inventory System (BEIS) was used with Version 5 of the Biogenic Emissions Landuse Database (BELD5). Category 1, 2 CMV: Nonpoint 2017 NEI Category 1 (C1) and Category 2 (C2), commercial marine cmv_c1c2 vessel (CMV) emissions based on Automatic Identification System (AIS) data, adjusted to 2018. “Emission Control Area (ECA)” inventory and non-AIS Canada emissions are unchanged from 2017. Point and hourly resolution. Category 3 CMV: Nonpoint Within state and federal waters, 2017 NEI Category 3 commercial cmv_c3 marine vessel (CMV) emissions based on AIS data, adjusted to 2018. Point and hourly resolution. ECA inventory and non-AIS Canada emissions are unchanged from 2017. locomotives: Nonpoint Line haul rail locomotives emissions for year 20177, projected to 2018 rail using annual energy outlook (AEO) and additional factors supplied by ERTAC. County and annual resolution. Nonpoint source oil Nonpoint Nonpoint 2017 NEI sources from oil and gas-related processes, and gas: projected to 2018 using based on U.S. Energy Information Administration (EIA) np_oilgas and Railroad Commission of Texas (TXRRC) historical production data. County and annual resolution. Residential Wood Nonpoint 2017 NEI nonpoint sources with residential wood combustion (RWC) Combustion: processes,were used as is, with no projection to represent 2018. County and Rwc annual resolution. Volatile Chemical Nonpoint Volatile chemical product (i.e., solvent) emissions for the year 2018 Products: vcp (Seltzer, 2021). Includes household cleaners, personal care products, adhesives, architectural and aerosol coatings, printing inks, and pesticides. Annual and county resolution. Remaining nonpoint: Nonpoint 2017 NEI nonpoint sources not included in other platform sectors, Nonpt including solvents. No adjustments were made to represent 2018. County and annual resolution. Nonroad: Nonroad 2018 nonroad equipment emissions developed with MOVES3. MOVES Nonroad was used for all states except California, which submitted their own emissions for the 2017 NEI that were adjusted to 2018. California emissions were interpolated to 2018 between California-provided 2017 and 2023 emissions. County and monthly resolution. On-road: On-road 2018 onroad mobile source gasoline and diesel vehicles from parking Onroad lots and moving vehicles. Includes the following emission processes: exhaust, extended idle, auxiliary power units, evaporative, permeation, refueling, and brake and tire wear. For all states except California, activity data were projected from 2017 to 2018 using factors developed using data from Federal Highway Administration and state departments of transportation. MOVES3 was run for 2018 to generate emission factors.. AirToxScreen 2018 Documentation 31 Platform Sector: NEI Data abbreviation Category Description and Resolution of the Data Input to SMOKE On-road California: On-road California-provided 2017 CAP and metal HAP onroad mobile source onroad_ca_adj gasoline and diesel vehicles from parking lots and moving vehicles based on Emission Factor (EMFAC) 2017, gridded and temporalized based on outputs from MOVES2014b. Volatile organic compound (VOC) HAP emissions derived from California-provided VOC emissions and MOVES-based speciation. 2018 was interpolated between 2017 and 2023 emissions from EMFAC2017. Point source Events Point source day-specific prescribed fires (including Flint Hills / grasslands) prescribed fires: for 2018 computed using SMARTFIRE 2 and BlueSky. ptfire-rx Point source wildfires: Point source day-specific wildfires for 2018 computed using ptfire-wild SMARTFIRE 2 and BlueSky. Non-US. fires: N/A Point source day-specific wildfires and agricultural fires outside of the ptfire_othna U.S. for 2018 from v1.5 of the Fire INventory (FINN) from National Center for Atmospheric Research (NCAR, 2017 and Wiedinmyer, C., 2011) for Canada, Mexico, Caribbean, Central American, and other international fires. Other Area Fugitive N/A Area fugitive dust sources from Canada from EQUATES 2016 with dust sources not from transport fraction and snow/ice adjustments based on 2018 the NEI: othafdust meteorological data. Annual and province resolution. Other Point Fugitive N/A 2017 point source fugitive dust sources from Canada EQUATES 2016/2017 dust sources not with transport fraction and snow/ice adjustments based on 2018 from the NEI: meteorological data. Annual and province resolution. othptdust Other point sources N/A Canada and Mexico point source emissions from EQUATES 2016. Canada point not from the NEI: sources were provided by ECCCC and and Mexico point source emissions othpt for 2016 (provided by SEMARNAT). Annual and monthly resolution. Other non-NEI N/A EQUATES 2016 for Canada and Mexico, except Canada nonroad. Projected nonpoint and Canada nonroad to 2018 based on US MOVES3 2018/2016 ratios. 2016 nonroad: othar Mexico (municipio resolution, provided by SEMARNAT) nonpoint and nonroad mobile inventories, annual resolution. Other non-NEI on- N/A Monthly onroad mobile inventory for Canada from EQUATES 2016 projected road sources: to 2018 using US onroad trends. Separate trends applied to refueling onroad_can (gas/diesel) and non-refueling (gas/diesel and LD/HD). Province resolution. Other non-NEI on- N/A Monthly onroad mobile inventory from MOVES-Mexico (municipio road sources: resolution) for 2017 7, adjusted to 2018 using interpolation between 2017 onroad_mex and 2020.. 2.2.2. Fires and Biogenics To approximate how much fires and biogenic primary emissions contribute to overall air toxics concentrations, we use a concept known as “zero-out” runs in CMAQ. This technique lets us split out the individual impacts from these source categories in the AirToxScreen results. AirToxScreen 2018 Documentation 32 CMAQ can compute biogenic emissions during a run, and it has options to take multiple sets of point- source fire emission files as input. These features were used in this study to quantify the impacts of biogenic and fire emissions by running CMAQ three times: 1. The base-case run used all fire and anthropogenic emissions with the option to generate biogenic emissions turned on. The biogenic zero-out run used all fire and anthropogenic emissions but with the option to generate biogenic emissions turned off. The fire zero-out run used all anthropogenic emissions had the option to generate biogenic emissions turned on, but it excluded the input files for wild, prescribed, and anthropogenic fires. 2.2.3. Speciation The emissions modeling step for chemical speciation creates the “model species” needed by the air quality model for a specific chemical mechanism. These model species are either individual chemical compounds (i.e., “explicit species”) or groups of species (i.e., “lumped species”). Model species are created in the emissions modeling process by directly mapping emissions from the emission inventory to the appropriate model species or by speciation of inventory species where a one-to-one match does not exist. For example, VOCs are speciated into numerous VOC-related model species defined by the chemical mechanism. In the AirToxScreen modeling platform, all CMAQ species that are explicit AirToxScreen HAPs were generated by directly mapping the NEI emissions for these HAPs; no HAPs in the United States were generated through speciation of VOC or PM2.5 in the emissions modeling step. More information on speciation approach and detailed tables of modeling species are available in section 3.2 of the technical support document for the 2018 emissions modeling platform. 2.2.4. Temporalization While the total emissions are important, the timing of the occurrence of emissions is also essential for accurately simulating ozone, PM and other pollutant concentrations in the atmosphere. Temporal allocation (i.e., temporalization) is the process of distributing aggregated emissions to a finer temporal resolution, thereby converting annual emissions to hourly emissions as is required by CMAQ. Temporalization takes these aggregated emissions and distributes them to the month, and then distributes the monthly emissions to the day and the daily emissions to the hours of each day. This process is typically done by applying temporal profiles to the inventories in this order: monthly, day of the week, and diurnal, with monthly and day-of-week profiles applied only if the inventory is not already at that level of detail. Sector-specific information, including meteorological-based temporal allocation and detailed charts of the temporal settings used for each sector in the emissions modeling platform can be found in section 3.3 of the technical support document. Ancillary data are available at the emissions modeling platform ftp site. AirToxScreen 2018 Documentation 33 2.2.5. Spatial Allocation For CMAQ, all emissions were allocated to 12-km grid cells. Sources with geographic coordinates such as point sources and fires are mapped to the appropriate grid cell based on those coordinates. Sectors with county-level resolution were allocated to 12-km grid cells using spatial surrogates, which are developed based on shapefiles of data with spatial patterns expected for the emissions category. AirToxScreen 2018 Documentation 34 Spatial surrogates were assigned to emissions sources based on SCC. A detailed description of these assignments can be found in section 3.4 of the emissions modeling platform technical support document. Ancillary files providing the surrogate assignments and describe the data used for the surrogates are available in the Supplemental Data provided with this TSD. 2.3. Preparation of Emissions Inputs for AERMOD For AirToxScreen, we use the capability in SMOKE to produce helper files that can be further post- processed to generate the source (SO) Pathway for AERMOD. The SO Pathway contains the source location and parameter information used by AERMOD. Helper files provide information about source location, release characteristics and temporal variability. They also provide source emissions. For sources in the CONUS, helper files also provide the column and row of the 12-km meteorological grid cell associated with the appropriate meteorological data (see also Section 3) to use for the emission source. Helper file formats and information vary by the type of source. For example, some emissions sources are point sources, some are gridded and some (like ports) are defined by polygon shapes. Gridded sources use different grid resolutions depending on their location (e.g., all Alaska gridded sources are 9 km2) and/or the type of source. The SMOKE interface for AERMOD models each source using a unit emissions rate rather than actual pollutant-specific emissions. The location, source parameter and temporal helper files join to produce the SO Pathway for each source’s AERMOD run. AERMOD provides source-specific concentrations per unit emissions rate (also known as χ/Q, abbreviated here as X/Q) that are not pollutant specific (but are source specific). These concentrations are then combined with pollutant-specific emissions for each source, using the emissions helper file from the emissions modeling process, to create the pollutant- and source-specific concentrations as shown below. X Cpollutant i= × Emissions Q pollutant i Because we are using a unit emission rate in AERMOD, we do not account for chemistry or deposition that would be specific to any individual pollutant. The format and content of the helper files vary by type of source, and we have chosen to organize the creation of separate sets of helper files by “run group.” A run group is a set of sources that use similar source characterization methods or parameters. Run groups allow us to combine sources with similar characteristics and run them in AERMOD together, even though they may have very different emissions. The emissions from the specific sources in the run group can then be applied to the source group X/Qs to create source-specific concentrations. We also combine the sources within the run group into source groups, which is the source-level resolution of concentration and risk results for AirToxScreen. In AirToxScreen, we allocate county-level emissions to grid cells for both CONUS and non-CONUS (i.e., Alaska, Hawaii, Puerto Rico and the Virgin Islands) sources; however, we use different grid cell resolutions, as discussed below. 2.3.1. Source and Run Groups – Overview Source and run groups are similar, but serve different functions in AirToxScreen. Source groups provide source attribution for the AirToxScreen census tract risk results. AirToxScreen 2018 Documentation 35 Run groups organize sources for modeling in AERMOD to minimize the number of separate model runs. It would be resource prohibitive to run the thousands of different source classification codes separately in AERMOD, so we grouped them into a manageable number of run groups that all have the same spatial resolution, approach for temporal variation and release characteristics (height and sigma-z). Run groups typically have multiple source groups, though two of them (OILGAS and RWC) have the same source group as run group. The inventory run groups are shown in Table 2-9. Those modeled in AERMOD as AREA sources using grid cells (i.e., “AREA (gridcell)”) were gridded from county level to grid cells using the same underlying surrogate data (shape files) used in CMAQ. Table 2-9. Run groups for AERMOD f Run Group NEI Data AirToxScreen Resolution o deled in Category Category Inventory Prior to Mo Emissions Modeling AERMOD as: Point Point Point, Point POINT, AREA Nonroad1 (fugitive) Airport Point Nonroad Point POINT, AREA (runway) Onroad On-network LD (LDON) Onroad Onroad County AREA (gridcell) Onroad Off-network LD (LDOFF) Onroad Onroad County AREA (gridcell) Onroad On-network HD (HDON) Onroad Onroad County AREA (gridcell) Onroad Off-network HD (HDOFF) Onroad Onroad County AREA (gridcell) Onroad Off-network-hoteling (extended Onroad Onroad County AREA (gridcell) idling and auxiliary power units) (HOTEL) Onroad On-network LD (LDON) Onroad Onroad County AREA (gridcell) Nonroad (NONRD) Nonroad Onroad County AREA (gridcell) Nonpoint 10-meter release height (NPHI) Nonpoint Nonpoint County AREA (gridcell) Nonpoint low-level release height (NPLO) Nonpoint Nonpoint, County AREA (gridcell) Nonroad2 Nonpoint Oil and Gas (OILGAS) Nonpoint Nonpoint County AREA (gridcell) Nonpoint Residential Wood Combustion Nonpoint Nonpoint County AREA (gridcell) (RWC) Nonpoint Agricultural Livestock (AG) Nonpoint Nonpoint County AREA (gridcell) Nonpoint Oil and Gas (OILGAS) Nonpoint Nonpoint County AREA (gridcell) Commercial Marine Vessels (CMV) Ports Nonpoint Nonroad Shapes POLYGON Commercial Marine Vessels (CMV) Nonpoint Nonroad Gridcell AREA (gridcell) Underway 1Rail yards. 2Locomotives. The run groups modeled as grid cells have different resolutions: 3 kilometers for all grid cells in Hawaii, Puerto Rico and the Virgin Islands; 9 kilometers for all grid cells in Alaska, and either 12 or 4 kilometers for grid cells in the CONUS, depending on the run group. These are summarized in Table 2-10. AirToxScreen 2018 Documentation 36 Table 2-10. Resolution of the run groups modeled as gridded sources Run Group (resolution is the value of the CONUS GRID: Alaska Hawaii Puerto Rico & number in the run group abbreviation) 12US1 GRID: 9AK1 GRID = 3HI1 Virgin Islands GRID = 3PR1 Onroad On-network LD (LDON) LDON4 LDON9AK LDON3HI LDON3PR Onroad Off-network LD (LDOFF) LDOFF12 LDOFF9AK LDOFF3HI LDOFF3PR Onroad On-network HD (HDON) HDON4 HDON9AK HDON3HI HDON3PR Onroad Off-network HD (HDOFF) HDOFF12 HDOFF9AK HDOFF3HI HDOFF3PR Onroad Off-network-hoteling (extended idling HOTEL4 HOTEL9AK N/A N/A and auxiliary power units) (HOTEL) Nonroad (NONRD) NONRD12 NONRD9AK NONRD3HI NONRD3PR Nonpoint 10-meter release height (NPHI) NPHI12 NPHI9AK NPHI3HI NHI3PR Nonpoint low-level release height (NPLO) NPLO12 NPLO9AK NPLO3HI NPLO3PR Nonpoint Oil and Gas (OILGAS) OILGAS4 OILGAS9AK OILGAS3HI N/A Nonpoint Residential Wood Combustion (RWC) RWC12 RWC9AK RWC3HI N/A Nonpoint Agricultural Livestock (AG) AG12 N/A N/A N/A N/A means there are no emissions for this run group in this grid. Table 2-11 and Table 2-12 describe the source characteristics of these run groups and how they relate to the platform sectors used for CMAQ emissions modeling. Note that the run groups developed for the county-level emissions were treated differently for CONUS versus non-CONUS domains. For the CONUS, we allocated to grid cells using the spatial resolution in the table. For the non-CONUS areas (AK, HI, PR, VI), we used these same run groups but allocated to tracts rather than grid cells. Design documents for the generation of the helper files for on-road, nonroad and nonpoint sources in CONUS and non-CONUS areas are provided in the Supplemental Data files. Table 2-11. Non-gridded AERMOD run groups AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach POINT NEI: point Release parameters for individual release All NEI point sources containing Platform: ptegu, points at each facility are taken from the 2017 AirToxScreen HAP emissions pt_oilgas, NEI, with defaulting done for missing or out- except for facility source type 100 ptnonipm of-range parameters. (airports). Spatial: Use specific geographic coordinates for stacks (“POINT”) and geographic coordinates along with fugitive length and width for fugitives (“AREA”). Temporal: Temporal profiles based on SCC codes for ptnonipm; based on continuous emissions monitoring data for ptegu. AirToxScreen 2018 Documentation 37 AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach AIRPORTS NEI: point RH=3 σz =0 Airport Facilities in the point Platform: Spatial: For runway-area (line) sources: inventory – all emissions where ptnonipm length based on a database containing facility source type code = 100 runway endpoint coordinates; 50-m width (airports). Includes seaplane ports for the major airports, 25-m width for the and heliports, Emissions used at OTAQ-provided (smaller) airports. All facility the facility unit’s process release emissions (NEI) spread equally over the point. runway(s). For these facilities all sub-facility emissions (airplane landing and For nonrunway sources: 10-m-square area takeoffs, ground support centered on NEI coordinates. equipment) are summed to the Temporal: monthly/day-of-week/hourly facility level, and are treated the profiles. Different profiles for Alaska same way with regards to spatial seaplanes. and temporal allocation. CMVP NEI: Nonpoint C1/C2 uses RH = 8.4 and σz = RH/2.15. C3 C1/C2 and C3 commercial marine Platform: uses RH = 20 and σz = 40.7 based on CMAQ vessels at ports. cmv_c12 and vertical emissions. cmv_c3 Spatial– polygon shapes updated from the NEI inventory shapes (there could be multiple polygons per port). Temporal: hourly data based on AIS activity data Table 2-12. AERMOD gridded run groups AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach LDON4 NEI: on-road RH = 1.3, σz = 1.2, Resolution is 4 km for On-network light duty mobile LDON9AK Platform: on- CONUS, 9 km for AK and 3 km for HI, PR, VI. emissions such as passenger car LDON3HI road Temporal: monthly temporal variation is exhaust and light duty passenger LDON3PR pollutant-specific and county-specific. truck brake and tire wear. County-specific hourly profiles are the same Emissions derived from SMOKE- for all pollutants based on benzene hourly MOVES. Includes refueling since emissions from SMOKE-MOVES (aggregate temporal profile and spatial only SCCs in this run group). surrogate for on-network is a better match for refueling than off-network. LDOFF12 NEI: on-road RH = 0.5, σz = 0.5, Resolution is 12 km for Off-network light duty mobile LDOFF9AK Platform: on- CONUS, 9 km for AK and 3 km for HI, PR, VI. emissions such as passenger car LDOFF3HI road Temporal: same approach as LDON4, but and passenger light truck start LDOFF3PR aggregate only SCCs in the LDOFF run group emissions. Derived from SMOKE- to compute hourly profiles. MOVES. Tailpipe height (no turbulence). AirToxScreen 2018 Documentation 38 AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach HDON12 NEI: on-road RH = 3.4, σz = 3.2, Resolution is 4 km for On-network heavy duty mobile HDON9AK Platform: on- CONUS, 9 km for AK and 3 km for HI, PR, VI. emissions such as running exhaust HDON3HI road Temporal: monthly temporal variation is from combination long haul and HDON3PR pollutant-specific and county-specific. short haul trucks and buses. County-specific hourly profiles based on Includes brake and tire wear. PM2.5 hourly emissions from SMOKE-MOVES Derived from SMOKE-MOVES. (aggregate only SCCs in this run group). HDOFF12 NEI: on-road RH = 3.4, σz = 0.5, Resolution is 12 km for Off-network heavy duty mobile HDOFF9AK Platform: on- CONUS, 9 km for AK and 3 km for HI, PR, VI. emissions such as start emissions HDOFF3HI road Temporal: same approach as HDON run from combination long haul and HDOFF3PR groups, but use only SCCs in the HDOFF run short haul trucks and buses: groups to compute hourly profiles. Derived from SMOKE-MOVES. Minimal dispersion for start emissions. Tailpipe height (no turbulence. HOTEL4 NEI: on-road RH = 3.4, σz = 0.5, Resolution is 4 km for Extended idling and auxiliary HOTEL9AK Platform: on- CONUS, 9 km for AK and 3 km for HI, PR, VI. power units (APU) that occur at road Temporal: same approach as HDON run truck stops. Minimal dispersion for groups, but use only SCCs in the HOTEL run hoteling (e.g., extended idling) groups to compute hourly profiles. emissions. Tailpipe height (no turbulence). NONRD12 NEI: nonroad RH = 2, σz = 1, Resolution is 12 km for Nonroad equipment such as lawn NONRD9AK Platform: CONUS, 9 km for AK and 3 km for HI, PR, VI. mowers, turf equipment, NONRD3HI nonroad Temporal: monthly temporal variation is agriculture and construction NONRD3PR pollutant-specific and county-specific. equipment, commercial County-specific diurnal scalars computed generators, power-washing based on benzene hourly emissions for the equipment, pleasure craft, aggregate of all SCCs in this run group recreational off-road. summed across all days of the year. NPHI12 NEI: nonpoint RH=10, σz = 4.7, Resolution is 12 km for Industrial processes (e.g., chemical NPHI9AK Platform: some CONUS, 9 km for AK and 3 km for HI, PR, VI plants, refineries, mines, metals); NPHI3HI of nonpt Temporal: uniform monthly/day-of-week. fuel combustion – industrial, NPHI3PR Diurnal: Use SMOKE hourly profile 26 – commercial, institutional and mostly daytime emissions-(use qflag residential (except wood); waste HROFDY). disposal. AirToxScreen 2018 Documentation 39 AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach NPLO12 NEI: nonpoint RH = 3.9, σz = 3.6, Resolution is 12 km for Solvents (consumer, commercial); NPLO9AK Platform: some CONUS, 9 km for AK and 3 km for HI, PR, VI. surface coating; commercial NPLO3LO of nonpt Temporal: same as NPHI run groups. cooking; locomotives, bulk NPLO3PR terminals, gas stations (stage 1); miscellaneous non-industrial (portable gas cans, auto repair shops, structure fires, and nonpoint mercury categories such as human cremation, dental amalgam). OILGAS4 NEI: nonpoint RH=10, σz =4.7, Resolution is 4 km for Oil and gas sources reported in the OILGAS9AK Platform: CONUS and 9 km for AK. There are no nonpoint NEI data category (i.e., OILGAS3HI np_oilgas emissions in HI, PR or VI. county-level emissions). Temporal: monthly profiles that align with the county/scc temporal profiles used in CMAQ v7.1 platform. Use qflag= MONTH and generate county-specific run group monthly profiles based on benzene emissions aggregated over all SCCs within each county. RWC12 NEI: nonpoint RH = 6.4, σz = 3.2, Resolution is 12 km for Fireplaces, woodstoves, hydronic RWC9AK Platform: rwc CONUS, 9 km for AK and 3 km for HI. There heaters used for residential RWC3HI are no emissions in PR or VI. heating. Temporal: hourly by grid cell based on county-specific hourly emissions (created by SMOKE using year-to-day factors derived from meteorological data that are used for many SCCs in this sector)-for all 8760 hours in the year-based on benzene emissions summed by hour and county across all SCCs in the run group. AG12 NEI: nonpoint RH = 1, σz = RH/2.15 = 0.465 m. Miscellaneous area sources; Platform: ag Resolution is 12 km. There are no emissions agriculture production – livestock, in AK, HI, PR or VI. beef cattle – finishing operations, Temporal: hourly profile based on ammonia dairy cattle – drylot/pasture dairy- hourly emissions for the Ag emissions sector confinement; swine production – (which includes both livestock and fertilizer). operations; poultry production. AirToxScreen 2018 Documentation 40 AERMOD Modeling Features: Release NEI Category and Height (RH; meters), Initial Vertical Run Group AirToxScreen Description of Sources CMAQ Sector Dispersion (σz; meters) and Temporal Approach CMVU12 NEI: nonpoint C1/C2 uses RH = 8.4 and σz = RH/2.15. C3 C1/C2 and C3 commercial marine CMVU9AK Platform: uses RH = 20 and σz = 40.7 based on CMAQ vessels underway CMVU3HI cmv_c12 and vertical emissions. Resolution is 12 km for CONUS, 9 km for AK CMVU3PR cmv_c3 and 3 km for HI and PR/VI Temporal: hourly by grid cell based on AIS activity data Each run group contains one or more source groups for purposes of presenting the AirToxScreen results. There are no situations where a source group fits into multiple run groups. Table 2-13 lists these source groups. Table 2-13. AirToxScreen source groups Source Group Name AirToxScreen Category Source Group Description Run Group All facilities in the POINT data category that have geographic coordinates other than those PT-StationaryPoint POINT with facility type = 100 (which are in the POINT "Airport" source group) or facility type = 151 (which are in the "Railyard" source group) OR-LightDuty- On-road, light duty, nondiesel (i.e., gasoline & OffNetwork-Gas Onroad ethanol blends) vehicles – off network LDOFF processes (e.g., “starts”) OR-LightDuty- OffNetwork-Diesel Onroad On-road, light duty, diesel vehicles – off network processes (e.g., “starts”) LDOFF OR-HeavyDuty- OffNetwork-Gas Onroad On-road, heavy duty, nondiesel vehicles – off network processes (e.g., “starts”) HDOFF OR-HeavyDuty- OffNetwork-Diesel Onroad On-road, heavy duty, diesel vehicles – off network processes (e.g., “starts”) HDOFF OR-LightDuty- OnNetwork-Gas Onroad On-road, light duty, nondiesel vehicles – on network processes (e.g., running emissions) LDON OR-LightDuty- OnNetwork-Diesel Onroad On-road, light duty, diesel vehicles – on network processes (e.g., running emissions) LDON OR-HeavyDuty- OnNetwork-Gas Onroad On-road, heavy duty, nondiesel vehicles – on- network processes (e.g., “running”) HDON OR-HeavyDuty- OnNetwork-Diesel Onroad On-road, heavy duty, diesel vehicles – on- network processes (e.g., “running”) HDON OR-Refueling Onroad On-road refueling LDON OR-HeavyDuty-Hoteling Onroad On-road, heavy duty diesel vehicle extended idling and auxiliary power units HOTEL AirToxScreen 2018 Documentation 41 Source Group Name AirToxScreen Category Source Group Description Run Group Off-road motorcycles, snow mobiles, golf NR-Recreational-inc- carts, outboard pleasure craft, personal PleasureCraft Nonroad watercraft, inboard/sterndrive pleasure craft, NONRD etc. Paving equipment, plate compactors, NR-Construction Nonroad trenchers, tampers/rammers, tractors/loaders/backhoes, cranes, signal NONRD boards/light plants, etc. NR- Mowers, leaf blowers, turf equipment, CommercialLawnGarden Nonroad chippers/ stump grinders, tillers, chainsaws, NONRD snow blowers, etc. NR- ResidentialLawnGarden Nonroad Mowers, leaf blowers, tillers, chainsaws, snow blowers, shredders, etc. NONRD NR-Agriculture Nonroad Agricultural tractors, combines, sprayers, balers, tillers, irrigation sets, swathers, etc. NONRD NR- CommercialEquipment Nonroad Generator sets, pressure washers, pumps, hydropower units, etc. NONRD Industrial Equipment (Forklifts, sweepers/scrubbers, other oil field NR-AllOther Nonroad equipment), logging equipment, railroad NONRD maintenance, underground mining equipment NR-CMV_C1C2_ports Nonroad Commercial marine vessel (CMV) emissions – C1/C2 while at ports NR-CMV Ports NR-CMV_C3_ports Nonroad CMV emissions – C3 while at ports NR-CMV Ports NR- CMV_C1C2C3_underway Nonroad CMV emissions – both C1/C2 and C3 marine – NR-CMV while underway Underway NR-Locomotives Nonroad Locomotive emissions NPLO NR-Point-Airports Nonroad Point source inventory where facility type = 100 "Airport" AIRPORT NR- Point-Railyards Nonroad Point source inventory where facility type = 151 "Rail Yard" POINT Nonpoint mining and quarrying, paved and unpaved roads, food and kindred products, NP-industrial Nonpoint mineral products, chemical manufacturing, NPHI non-ferrous metals and nonpoint industrial processes not elsewhere classified NP-CommercialCooking Nonpoint Commercial cooking (charbroiling, frying) NPLO NP-OilGas Nonpoint Oil and gas operations (pumps, dehydrators, tanks, engines OILGAS Degreasing, dry cleaning, consumer and NP-SolventsCoatings Nonpoint commercial solvents, industrial surface NPLO coating, non-industrial surface coating AirToxScreen 2018 Documentation 42 Source Group Name AirToxScreen Category Source Group Description Run Group NP-StorageTransfer_ Bulk terminals, petroleum, organic and BulkTerminals_GasStage Nonpoint inorganic chemical storage and transport – 1 has gas station emissions of only stage 1 (tank NPLO refueling) not stage 2 (vehicle refueling) NP- Poultry/livestock, laboratories, dental alloy, MiscellaneousNonindust Nonpoint motor vehicle fires, portable fuel containers, NPLO rial residential charcoal grilling NP- Residential, industrial, commercial and FuelCombustion_not_R Nonpoint institutional (ICI) fuel combustion, but NPHI WC excludes residential wood combustion NP- ResidentialWoodCombu Nonpoint Residential wood combustion – woodstoves, stionRWC fireplaces RWC Open burning, yard and household waste, NP-WasteDisposal Nonpoint managed burning, slash (logging debris), onsite incineration, waste water treatment, NPHI composting, landfills, scrap materials NP- AgriculturalLivestock Nonpoint Emissions from livestock waste (HAPs only) AG CMAQ category only: wildfires, prescribed fires and agricultural burning; while the FIRE FIRE modeled results are not separated by type of fire, the tabular emissions summary breaks out the three fire types separately CMAQ category only: VOCs and particular BIOGENICS BIOGENICS HAPs (formaldehyde, acetaldehyde and methanol) emitted from vegetation (and NOX from soils) Not an emissions group: formed in the SECONDARY SECONDARY atmosphere due to photochemical reactions (CMAQ) Not an emissions group: concentrations due to ubiquitous nature of some HAPs with long BACKGROUND BACKGROUND residence time or coming into the modeling domain from outside the domain such as carbon tetrachloride The next section describes in more detail the development of emissions and characterization of the emissions for the run groups by NEI data category. 2.3.2. Point Sources Excluding Airports Point sources used in AERMOD are all sources in the point data category in the NEI that are in the United States, Puerto Rico or Virgin Islands;2 have geographic coordinates present in the NEI and emit at least one HAP. This section discusses the nonairport point sources; airports are discussed in Section 2.3.3. 2Offshore platforms (i.e., FIPS = 85; meaning federal waters) are excluded. AirToxScreen 2018 Documentation 43 The starting point for the emissions modeling for point sources is the SMOKE modeling file, or FF10 discussed in Section 2.1.3. The format of this point FF10 file is in the SMOKE User’s Manual https://www.cmascenter.org/smoke/documentation/4.0/html/ch08s02s08.html#sect_input_ptinv_ff10 and the file is posted on the emissions modeling platform ftp site. The SMOKE FF10 was split into the CMAQ platform sectors (see Table 2-8) ptnonipm, ptegu and pt_oilgas, and the ptnonipm sector was further split out into airports and nonairports using the FACILITY_SOURCE_TYPE field. The FACILITY_SOURCE_TYPE indicates the type of facility (where available) and a value of “100” is used for airports (including large and small airports, heliports and seaplanes). For purposes of AERMOD, the ptnonipm (after removal of airports) and pt_oilgas were run in the same way and were not distinguished from one another. The ptegu sector was treated differently due to different temporalization. The ptegu sector allowed for hourly variation at the unit level, whereas the ptnonipm (including pt_oilgas) used the monthly, day-of-week and diurnal profiles used by SMOKE. Within the run, each ptegu unit was temporalized using hourly emission values as discussed in Section 2.3. Non-EGU units were modeled in AERMOD using temporal allocation factors derived from the temporal profiles used in SMOKE for CMAQ. Many facilities included a mixture of EGU and non-EGU processes. In such cases, all sources at a given facility were modeled in the same AERMOD run.3 This ensured that ambient impacts were calculated for a consistent set of receptor locations for all sources at the facility. 2.3.2.1. Point source characterization for AERMOD In the SMOKE FF10 modeling file, a source is a unique combination of EIS process id and EIS release point id. In AERMOD, we chose to group together sources with the same release point characteristics (geographic coordinates, release point type and release point parameters) and temporal profile since they will have the same X/Qs, and grouping them reduces the number of AERMOD runs needed. This is because many processes use the same temporal profiles, so there are fewer unique combinations of these parameters than process id/release point id combinations. The FF10 fields that must be unique (when combined) are: FACILITY_ID, MONTHLY temporal profile code, WEEKLY temporal profile code, ALLDAY temporal profile, ERPTYPE (except that ERPTYPEs 3, 4 and 6 are treated as equivalent), STKHT, STKVEL, STKTEMP, STKDIAM, FUG_HEIGHT, FUG_WIDTH_YDIM, FUG_LENGTH_XDIM, FUG_ANGLE, LATITUDE and LONGITUDE. Sources within the same facility were run together in AERMOD. To assign the proper meteorological data, each facility was assigned to a grid cell based on the geographic coordinates of the release points. Helper files were developed for point locations, point source parameters, area source parameters, temporal parameters and emissions. Point locations were based on the latitude and longitudes in the FF10 (the release point coordinates) and were converted to UTM coordinates. Point source parameters used the release point type code (field name is ERPTYPE in the FF10 file) and associated stack or fugitive parameter information from the FF10. Table 2-14 lists the release point types and how each is assigned an AERMOD type. 3AERMOD temporalization is performed at the level of source IDs, so using different temporalization schemes at one facility is possible. AirToxScreen 2018 Documentation 44 Table 2-14. Assignment of AERMOD source type for point sources Emission Release Point Emission Release Point Type AERMOD Source Type Code or Special Type Code (NEI) Description Adjustment 1 Fugitive AREA 2 Vertical POINT 3 Horizontal POINTHOR 4 Goose neck POINTHOR 5 Vertical with rain cap POINTCAP 6 Downward-facing vent POINTHOR For AERMOD area sources, the following fields are used to characterize a fugitive release (units in parentheses):  Latitude and longitude (decimal degrees)  FUG_HEIGHT – fugitive height (ft)  FUG_WIDTH_YDIM4 – fugitive width (east/west), (ft)  FUG_LENGTH_XDIM4 – fugitive length (north/south), (ft)  FUG_ANGLE– fugitive angle (degrees) The NEI allows fugitive release angles of 0 to 89 degrees. While the fugitive release point latitude and longitude are not specified to be any particular location in the area source in the emission inventory system, due to the limits on the angle and the general conventions for specifying the source in AERMOD, we interpret the release point as the follows: 1) the lat/lon is treated as the most western corner; 2) the angle is measured clockwise from true (geodetic) north (not magnetic north); 3) length is the measure along the side that would run in the north-south direction if the angle were 0 degrees and 4) width is the measure along the side that would run in the east-west direction if the angle were 0 degrees. As shown in Figure 2-1, the release point is at the push pin, the width is 680 feet, the length is 1897 feet and the angle is 22 degrees. Geographic coordinates were converted to UTM, with all release points within the same facility assigned to the same UTM zone. Release parameters are converted to meters in SMOKE as required by AERMOD. The initial vertical dispersion (σz) was set to FUG_HEIGHT/4.3 if the FUG_HEIGHT was greater than 10 meters, and 0 m if otherwise. For AERMOD point sources, the following fields are used to characterize a stack release. The SMOKE FF10 units are listed below; they are converted to metric (m, m/s, Kelvin) as required by AERMOD.  Latitude and longitude (decimal degrees)  STKHGT– height of stack (ft)  STKDIAM – diameter of stack (ft)  STKTEMP – temperature of stream exiting stack (°F)  STKVEL – velocity of stream exiting stack (ft/s) 4These SMOKEFF10 variables will be renamed to FUG_WIDTH_XDIM and FUG_LENGTH_YDIM to keep with the convention of X as the east-west direction and Y as the north-south direction. AirToxScreen 2018 Documentation 45 Separate helper files are derived for AREA versus POINT AERMOD source types because the source parameters required for these two types differ. Figure 2-1. Example fugitive source characterization: NEI length = 1897 feet, width = 680 feet and angle = 22 2.3.2.2. Point source temporalization As in CMAQ, point sources in the ptnonipm sector of the AirToxScreen CMAQ platform were modeled differently in AERMOD from those in the ptegu sector with respect to the temporalization of the emissions. The ptegu sectors were temporalized allowing for hourly variation at the unit level, whereas the ptnonipm and pt_oilgas sectors used the monthly, day-of-week and diurnal profiles used by SMOKE for the CMAQ platform. Within the run, each ptegu unit was temporalized using hourly emission values as discussed below. Non-EGU units were modeled in AERMOD using temporal allocation factors derived from the temporal profiles used in SMOKE for CMAQ. Many facilities included a mixture of EGU and non- EGU processes. In such cases, all sources at a given facility were modeled in the same AERMOD run.5 This ensured that ambient impacts were calculated for a consistent set of receptor locations for all sources at the facility. For non-EGUs, temporal helper files were prepared based on the temporal cross reference file for point sources (PTREF) and the temporal profile files (TPRO) used by SMOKE (same data as are used for CMAQ). The AERMOD scalars reflect diurnal, day-of-week and/or monthly variability and are determined from 5AERMOD temporalization is performed at the level of source IDs, so using different temporalization schemes at one facility is possible. AirToxScreen 2018 Documentation 46 the SMOKE diurnal, weekly and monthly temporal profile values. Table 2-15 shows the temporal variability options for non-EGUs. Table 2-15. Options for temporal variation specification in helper files Temporal Variability TPRO Qflag (for Number of Notes and Calculation of Scalar Using AERMOD) Scalars SMOKE TPRO Fractions1 Uniform – every Monthly profile code MONTH 12 AERMOD Scalar value is equal to 1/12 for hour of every day 262, daily profile each of the 12 months (identical to profile emits the same code 7, hourly profile code 262) amount code 24 Monthly Variation Day-of-week profile MONTH 12 AERMOD Scalar value month is monthfracm only code 7, hourly profile Where monthfracm is monthly fraction code 24 from the TPRO monthly profile assigned to the source Diurnal Variation Monthly profile code HROFDAY 24 AERMOD Scalar value month is dayfracd Only 262, day-of-week Where dayfracd is the diurnal fraction from profile code 7 the TPRO diurnal profile assigned to the source Month and hour Diurnal profile code MHRDOW 864 AERMOD Scalar value = of day type same for all (=24hrs*3d monthfracm ∗ dayfracd ∗ hourfrach variation in which weekdays ay M-F is the same, types*12 Where monthfracm is the monthly fraction, but Sat and months) dayfracd is the day-of-week fraction and Sunday can be hourfrach is the diurnal fraction for each different hour for weekdays, Saturday and Sunday from the TPRO 5 × sum of weekday scalars + sum of Saturday scalars + sum of Sunday scalars = 1 Month and hour Diurnal profile code MHRDOW7 2016 AERMOD Scalar value = of day variation varies by day-of- (=24 hrs*7 monthfracm ∗ dayfracd ∗ hourfrac and every day of h week day types * the week could be 12 months) Where monthfracm is the monthly fraction, different dayfracd is the day-of-week fraction and hourfrach is the diurnal fraction for each hour every day of the week from the TPRO Sum of scalars is 1 For EGUs, hourly temporalization was used, based on continuous emission monitoring data for heat input. The approach is the same as used in CMAQ and is described in more detail in the modeling platform TSD. For AERMOD, separate facility-specific helper files were developed with the hourly scalars for each hour of the year. 2.3.3. Airport Point Sources Airports are in the SMOKE point FF10 file and can be identified by the FACILITY_SOURCE_TYPE field “100” (for an airport facility). We modeled airports as one of two types: runway line sources or small AirToxScreen 2018 Documentation 47 (10- by 10-meter) area sources. The runway airports are those for which we have geographic coordinates for the runway endpoints from either the National Transportation Atlas Database or an OTAQ-supplied database. The nonrunway airports have no runway information and were modeled as small area sources. These include smaller airports, seaplanes and heliports. Although there may be multiple processes at an airport (e.g., commercial aircraft emissions and ground support equipment emissions), we did not treat them differently with respect to source characterization. Emissions at an airport were summed and apportioned equally among the runways or to the small area source. Airports have multiple sources in the AERMOD helper files only if the airport had more than one runway. Runway information is not in the NEI. The NEI has a site latitude and longitude, and all release points (fugitives) use the site coordinates. To get runway endpoints, we matched the NTAD/OTAQ databases with the NEI emissions. Non-matches (NEI facilities with no match to NTAD/OTAQ based on airport identifiers, or NEI facilities with coordinates inconsistent with the runway locations) were modeled as area sources based on the NEI coordinates. Matches were done using the airport location id (typically a 3- or 4-character field), which is an alternate facility id in EIS and is the id used for the NTAD/OTAQ airports. The starting-point airport facilities and final runway airports resulting from the matching effort, and detailed set of process steps taken to do the matching, are available in the Supplemental Data folder. This process resulted in 7,384 nonrunway airports and 11,991 runway airports. 2.3.3.1. Airport characterization for AERMOD For runway airports, an airport source is a runway at an airport facility; for nonrunway airports, an airport source is a 10-by-10-meter area source at an airport facility. Runways are modeled as AERMOD “LINE” sources, using the coordinates of the runway endpoints, which are assumed to be in the center of the runway width. All runways are assumed to be 50 meters wide (NTAD-based runways) or 25 meters wide (OTAQ-based runways) and have a release height and initial vertical dispersion of 3 meters. All pollutant emissions are divided equally across all runways at the airport. Airports without runway characterization are modeled as 10m by 10 m area sources with the same release height and vertical dispersion as runway airports. The NEI airport coordinates are used as the southwest corner of the area source and the angle is 0. 2.3.3.2. Airport temporalization Airports used a single set of temporal profiles for all processes within the airport, since the emissions from all processes are combined on the airport runways or as the small area sources. All airports use the same set of monthly, day-of-week and diurnal profiles except for Alaska seaplanes. See the Emissions Modeling Platform documentation for more information on temporalization of airport emissions. 2.3.3.3. Airport emissions Emissions are applied to X/Qs for airports differently than other point sources. Instead of multiplying the source-based emissions to the X/Qs, we multiply the total facility emissions to a facility-aggregated X/Q, which for runway airports is the aggregate across all runways. In creating the emissions inputs to AERMOD, the unit emission rate, 1000 g/s, is the unit emission rate for the entire airport, not each runway. To apportion the 1000 g/s to each runway, the 1000 g/s is divided by the number of runways. AirToxScreen 2018 Documentation 48 AERMOD then outputs the annual average concentration for the entire airport, not each runway. The annual average concentrations reflect the apportionment between the different runways. Because it can be assumed that each runway will get the same share of actual HAP emissions as the unit emission rate based on the fraction variable, the specific HAP emissions for the entire airport can be used instead of runway-specific emissions. This approach reduces the number of records in both the emissions helper file used for X/Q and the AERMOD annual average concentration output file based on the unit emission rate. NEI Lead emissions near airports included lead emitted during the climb-out and approach modes, which occur at altitude and are not included in AirToxScreen. To account for this, we adjusted the NEI- specific lead6 emissions estimates used in AERMOD down by 50 percent, based on previous modeling conducted at the Santa Monica (SMO) airport indicating that nearly 50 percent of emissions occurred in these higher-altitude modes (see Table 2-16). Table 2-16. Lead emissions (kg/yr) at SMO in 2008 by aircraft operation mode Mode Emissions (% of Total) Taxi to runway 20.4 (17.6%) Run-up 13.5 (11.4%) Takeoff roll 10.0 (8.4%) Climb-out 37.9 (32.7%) Approach 17.9 (15.8%) Landing 9.4 (7.9%) Taxi to apron 9.5 (8.4%) The emissions helper file format for runway and nonrunway airports is the same and is shown in Table 2-17. Table 2-17. Airport emissions file format Field Name Description State abbrev 2-character abbrev. Use TB for tribe Facility id Identifer for facility (i.e., EIS ID) Facility name Facility name FAC_SOURCE_TYPE Code indicating facility type (from SMOKEFF) pollutant name Use SMOKE shortname Emissions (tons) Emissions from the SrcID multiplied by the Metal/CN speciation factor (column AI) from the AirToxScreen_Pollutants.xslx) For airports, for metals only, multiply metal pollutant emissions from (2275050011=General Aviation /Piston; 2275060011=Air Taxi /Piston) by 0.5 6We designed the programs to reduce all metals from the following 2 SCCs: 2275050011=General Aviation /Piston and 2275060011=Air Taxi /Piston by 50%; however, the only metal emissions from these 2 SCCs are lead, so only lead was reduced. AirToxScreen 2018 Documentation 49 2.3.4. Nonroad, On-road and Nonpoint – County-level Sources Nonroad, on-road and nonpoint (other than CMVs and locomotives) sources in the NEI are estimated at the county level and use various levels of temporal variation. For AERMOD, we characterize these as “AREAPOLY” sources, which provide an emission flux over a defined polygon shape. We also characterize the temporal variability based on the various options available in AERMOD. The main steps needed to prepare county-level sources for AERMOD are spatial allocation, assignment of temporal scalar factors and assignment of release parameters. County-level sources must undergo spatial allocation prior to being input into AERMOD. All county-level sources are allocated to finer resolution using spatial surrogates. For CONUS domain sources, they are allocated to 4-km or 12-km resolution, and the size is primarily based on our confidence in the surrogate data at fine resolution and its representativeness of the source category for which it is being used. For non-CONUS county-level sources, they are allocated to 9-km resolution (Alaska) or 3-km resolution (Hawaii, Puerto Rico and the Virgin Islands). This method was first used in the 2014 NATA. Mostly, the same underlying surrogate data are used for AERMOD as CMAQ and are discussed in detail in the technical support document of the emissons modeling platform developed for CMAQ emissions. One difference is that the National Land Cover (NLCD) data are not available for Hawaii, Puerto Rico and the Virgin Islands (i.e., grids 3HI and 3PR), so an alternative land cover database, the Coastal Change Analysis Program (C-CAP) database was used for the land cover categories. C-CAP does not have as many categories as the NLCD; low and medium intensity land was not available and other land categories were substituted. For example, if the surrogate called for open+low, then only open was used in the 3HI and 3PR grids. Another difference is that CMAQ uses 12 km for all gridded emissions whereas AERMOD uses 4 km for some of the source groups in the CONUS; also, CMAQ does not cover non-CONUS areas. As in CMAQ, the spatial surrogates are assigned according to SCC codes. The SCC-to-surrogate cross reference and surrogate documentation workbook is provided in the Supplemental Data folder. In addition to spatial allocation, the emissions are temporally allocated to capture the temporal variability of emissions throughout the year, day and/or hour of day. Temporal allocation is done because not all emissions are emitted uniformly throughout the year. Ideally each specific SCC (and in some cases SCC and county) would be given a different temporal profile as is done when preparing emissions for CMAQ. But because all sources in the same run group must have the same scalars, we need temporal scalars that best represent the aggregation of the individual SCCs. In addition, these scalars, apart from onroad and nonroad mobile run groups, do not vary by pollutant because AERMOD is run at a unit emissions rate and the pollutant-specific emissions are applied after the model is run. The scalars can vary by county. For the onroad and nonroad mobile categories, AERMOD preserves the pollutant-specific monthly variation. For each run group, SMOKE is used to produce AERMOD helper files that provide the locations and coordinates of the gridded sources, the area source parameters, the temporal scalars and the emissions by source and pollutant, broken out by run group. The subsections below provide more information on how the emissions are developed and modeled in AERMOD. Detailed information on the development of the emissions at county level is provided in the national emissions inventory technical support document. AirToxScreen 2018 Documentation 50 2.3.4.1. On-road On-road uses 5 run groups: LDON4, LDOFF12, HDON4, HDOFF12 and HOTEL4, as described in Table 2-12. These run groups distinguish on-network (ON) versus off-network (OFF) sources, heavy-duty versus light-duty sources, and hoteling (extended idling and auxiliary power units). For characterization, the release height of LD ON vs OFF is different (1.3 m vs. 0.5 m) to account for added dispersion from the vehicle wake. Sigma-z is also different (1.2 vs. 0.5) to account for more dispersion on the roads. Off-network sources use minimal dispersion associated with start emissions (no turbulence). HD on and off and hoteling (which is pertinent only to HD vehicles) use the same release height based on average tailpipe height (3.4 m) but different dispersion. Sigma-z for on-network is 3.2 m, but for off-network and hoteling it is 0.5 m to account for less dispersion during idling. On-network sources and hoteling use finer resolution (4 km vs. 12 km) in the CONUS domain than for off-network. This is because we have more confidence in the spatial surrogates used to allocate the county-level emissions. The on-network sources use AADT data, whereas the off-network emissions are allocated to broad classes of land use data. Off-network exhaust and evaporative emissions from passenger cars are allocated to the “all development” land category, which consists of the sum of the following land categories: developed, open space; developed, low intensity; developed, medium intensity; and developed, high intensity. This is land with greater than zero percent impervious surface. For HI, PR and VI, which use C-CAP, there is no low or medium intensity, so it is the sum of open space and high intensity. Hoteling uses a set of truck stops that was updated for the 2017 Modeling Platform. The temporal resolution for onroad run groups is hourly across the year. Different profiles were developed for off-network than on-network due to the types of mobile processes (running vs. starts, for example) that occur. They are also different for heavy duty versus light duty because of the different vehicle mixes (trucks vs. cars). For the modeling platform, more refined temporal profiles were developed for the diurnal variation in vehicle miles travelled based on telematics data analyzed under the CRC Project A-100 (ERG 2017). In developing the hourly scalars for CONUS areas, we used the SMOKE-MOVES hourly emissions reports by county and SCC for a key pollutant for each run group. We chose PM2.5 for HD run groups and benzene for LD run groups. We then aggregated emissions across all SCCs in that run group. Hourly emissions were converted to local time (considering daylight saving time where appropriate). SMOKE- MOVES was run only for the CONUS. National averages were used for non-CONUS temporal profiles. Because we run grid cells (and not counties) in AERMOD, we assigned each grid cell to a county for purposes of assigning county-specific hourly emissions. Due to the unit emissions rate, we could not account for differences across pollutants in temporal variation, which can occur due to different temperature impacts on EFs that vary by pollutant, except at the monthly level. We imparted monthly variation to on-road run groups by pollutant by running unit emissions rates by SEASON to get season- specific X/Q values, and then by post-processing the X/Q by at the seasonal level (i.e., applying seasonal emissions). This is further discussed in Section 3.4. The temporal variability of on-road mobile sources varies by emissions process (e.g., running vs. idling), vehicle type, road type and pollutant. Different pollutants can have different temporal variation because some pollutants’ emission factors depend on meteorology, which has diurnal variation. AirToxScreen 2018 Documentation 51 For non-CONUS areas, we used hourly scalars by run group that were developed by taking a national average of the hourly scalars in the CONUS area. 2.3.4.2. Nonroad Nonroad equipment, i.e., emissions other than planes, trains and ships that are generated by MOVES, are put into a single run group. While there are differences in types of equipment and therefore emissions characterization, for this national study we chose to use a more simplified approach by characterizing equipment with the same spatial resolution, temporal profile and release characteristics. The release characteristics are a release height of 2 meters and an initial vertical dispersion of 1 meter, the same as used for the 2014 NATA. We chose a spatial resolution of 12 km for the CONUS domain; however, we used a variety of spatial surrogates for different SCCs. For example, for agricultural nonroad SCCs such as combines, tractors and balers, agricultural land from NLCD was used, whereas for pleasure craft, water (also from NLCD) was used. For non-CONUS, we also used land use surrogates and allocated to grid cells. We used the same temporal allocation approach as used in the 2014 NATA. For more information, please see the Emission Modeling Platform documentation. AirToxScreen 2018 Documentation 52 2.3.4.3. Nonpoint – NPHI12 and NPLO12 Nonpoint stationary sources are broken out into several run groups: NPHI, NPLO, RWC and OILGAS. NPHI and NPLO are very similar; they use the same temporal profile (diurnal profile 26 and uniform weekday and monthly profiles. They are allocated to a 12-km spatial resolution (12 km for CONUS sources and a 9- or 3-km resolution for non-CONUS). NPHI uses a release height of 10 meters and an initial vertical dispersion of 4.7 meters, which is based on release height/2.15. NPLO uses a release height of 3.9 meters and an initial vertical dispersion of 3.6 meters. Residential nonwood combustion is in the NPHI12, which uses the same release height as industrial, commercial and institutional (ICI) fuel combustion. Other than locomotive emissions, the two nonpoint run groups are the same as the CMAQ “nonpt” sector. The CMAQ “nonpt” sector excludes locomotive emissions, which are in the NPLO run group. 2.3.5. Residential Wood Combustion This is a source group that has its own run group; i.e., the run group and source group are identical. It is separate from other run groups because its temporalization, which depends on daily temperature, is unique. This run group comprises wood stoves (indoor and outdoor) and fireplaces. The NEI SCC codes are 2104008* (includes wood stoves and fireplaces, fire pits/chimneas and hydronic heaters) and 2104009000 (fire logs). We used a 6.4 m release height and 3.2 m initial vertical dispersion. All SCCs in the source group used the same spatial surrogate: NLCD Low Intensity Development. The temporal approach uses temperature data to assign emissions to days of the year. This was done using the “Gentpro” feature of SMOKE that was developed specifically for this sector, as the level of residential wood combustion activity depends on the daily minimum temperature. We took the same approach using 2018 meteorological data. Diurnal profiles based on device type were also applied. For AERMOD, we took the SMOKE-generated hourly emissions by SCC and created the temporal scalars using benzene emissions for all SCCs by hour of the year. Hourly scalars were computed by county as: Scalarhour-i = 8760 * (Benzene)county,hour-i ∕ Σall hours,county(Benzene) To assign grid cells to counties, we used the county having the most RWC HAP emissions. Puerto Rico and the Virgin Islands do not have any emissions for this source group. For Alaska and Hawaii, the hourly scalars were computed in the same way as the CONUS scalars. However, the underlying data SMOKE used for generating the hourly benzene emissions were not based on day- specific meteorological data. Instead, monthly temporal profiles for Alaska and Hawaii were calculated from the national average of the 2018 meteorology-based profiles, and SMOKE SCC-to-profile mappings were used for daily and diurnal variation. 2.3.6. Oil and Gas This run group covers all nonpoint oil and gas sector emissions in the NEI. It contains only one source group – the oil and gas source group. This allows a fine spatial allocation resolution of 4 km. This resolution was chosen because the surrogates are based on well locations from a commercially available AirToxScreen 2018 Documentation 53 database called HPDI that compiles oil and gas data (such as individual well locations, production information, drilling information and well completion data) from state databases – the same underlying data used by the Oil and Gas tool during the creation of the NEI. Because of the different types of activity data underlying the tool, there are numerous spatial surrogates applied to the various SCCs that make up the nonpoint oil and gas category. The same source of data (HPDI) was used to develop temporal profiles by county and SCC. This run group uses only monthly variation. County-specific monthly scalars were computed from benzene monthly emissions across all SCCs from the run group in the county (as shown below). If for a certain county there are no benzene emissions, then the sum of all HAPs was used. Benzene Scalar i 1 i= × # of days Benzene i ∑ i # of daysi Where: Benzenei = the tons of benzene in month i for the county associated with the met grid cell across all SCCs in OILGAS # of daysi = the number of days in month i The release height chosen is the same as NPHI – 10 m height and 4.7 m for the initial vertical dispersion. 2.3.7. Agricultural Livestock Like oil and gas, this run group contains just the one source group – agricultural livestock. We kept this run group separate due to the different temporal nature of livestock emissions from other nonpoint run groups and to be consistent with the temporal allocation used in CMAQ. In CMAQ, these emissions were modeled in the “ag” sector and use a meteorologically based temporalization. For AERMOD, a separate run group was created for consistent temporalization across the two models. For AERMOD, hourly scalars were computed from hourly ammonia emissions. The release height chosen was 1 m, and 0.465 m for the initial vertical dispersion. 2.3.8. Commercial Marine Vessels – County/Shape-level Sources This run group has the same sources as the cmv_c1c2 and cmv_c3 CMAQ modeling sectors except that for AERMOD, we include only sources in U.S. state or territory waters, whereas in CMAQ, ships in federal waters were also included. The NEI inventory uses different SCCs to distinguish between ships that are at ports (hoteling or maneuvering) versus ships that are moving along waterways, or “underway” (cruising or in low speed zones), and uses different SCCs for diesel versus residual oil ships. The NEI provides emissions by port or underway “shape,” which are drawn to provide sub-county geographic resolution of the cmv emissions. In all, there are four SCCs in this run group. Table 2-18 lists the SCCs in the CMV run group. Table 2-18. NEI SCCs covered in the CMV run group SCC Sector Description: Mobile Sources Prefix for All 2280002100 cmv Marine Vessels; Commercial; Diesel; Port 2280002200 cmv Marine Vessels; Commercial; Diesel; Underway AirToxScreen 2018 Documentation 54 SCC Sector Description: Mobile Sources Prefix for All 2280003100 cmv Marine Vessels, Commercial; Residual; Port emissions 2280003200 cmv Marine Vessels, Commercial; Residual; Underway emissions For AERMOD, each port source is characterized by a polygon shape with multiple vertices and SCC. The polygon shape is based on the NEI shape. It is assumed that the emissions are homogeneous in each polygon. Note that in the NEI, a single port may be characterized by multiple NEI polygon port shapes. The polygon emissions for all polygons within a single port are proportional to the total port emissions based on the shape’s area compared to the total area of all polygons for that port. The underway shapes were the 12 km CMAQ grid cells for CONUS and 9 km grid cells for Alaska and 3 km grid cells for Hawaii and Puerto Rico. The AirToxScreen underway polygons were simplified from the NEI underway shapes such that multiple simplified polygons were constructed to represent complex NEI underway shapes. For underway shapes, the emissions of the simplified shape are assumed proportional to the NEI shape based on the simplified shape area. An example of a port shape is shown in Figure 2-2. Figure 2-2. Port shapes for Los Angeles and Long Beach, California For AirToxScreen, release characteristics are different for C1/C2 vessels than for C3. For C1/C2, we used a release height of 8.4 meters and an initial vertical dispersion of 3.907 meters. Category 3 vessels are larger and different release parameters were used for consistency with CMAQ. In CMAQ, C3 vessels were characterized as point sources so the model would compute plume rise. We developed the C3 AirToxScreen 2018 Documentation 55 AERMOD release height and initial vertical dispersion to be consistent with the characterization in CMAQ. We used the same release height, 20 m. For the initial vertical dispersion, we examined summaries of plume rise of C3 from CMAQ; in particular, we looked at a SMOKE report of the annual NOX vertical distribution by county for the CMAQ cmv_c3 sector. For every county, we found that 50 percent of the emissions were above layer 3 and 50 percent were below. The midpoint of layer 3 is 60.7 meters. Therefore, we computed sigma-z as 60.7-20 = 40.7 m. 2.3.9. Urban/Rural Determination for All Emission Sources The urban/rural determination for all emissions sources was based on the urban/rural classification of nearby census blocks. We classified each populated census block in the United States as urban if its centroid falls within an “urbanized area.” For the 2010 Census, an urbanized area comprises a densely settled core of census tracts or census blocks, or both, that meet minimum population density requirements (50,000 or more people), along with adjacent territory. The adjacent territory generally is also densely settled, but it may also contain some lower-density areas as well as nonresidential urban land uses. About 500 such areas are included in the 2010 Census. For point sources, each facility was designated as urban or rural based on the urban/rural classification of the nearest census-block centroid receptor to the average facility latitude/longitude (averaged across all release points emitting HAPs). Each gridded source was designated based on the urban/rural classification of the nearest block receptor to the center of the 12-km grid cell (so all 4-km sources within the same 12-km grid cell are treated the same, based on the center of the 12-km grid cell). There were 353 grid cells on the borders of the United States that were inadvertently excluded from the approach; these were assigned a classification of rural, which would be expected based on their locations. For CMV ports, each individual port shape was designated based on the urban/rural classification of the census block nearest to the center of the port shape. All CMV underway sources were designated as rural. AirToxScreen 2018 Documentation 56 3. Air Quality Modeling and Characterization The AirToxScreen emission estimates described in Section 2 are used as inputs to EPA air quality models to estimate ambient concentrations of emitted air toxics. An air quality model is a set of mathematical equations that uses emissions, meteorological and other information to simulate the behavior and movement of air toxics in the atmosphere. Air quality models estimate outdoor concentrations of air toxics at specified locations. The modeling approach for the HAPs in AirToxScreen includes development and application of a hybrid approach blending a chemical transport model (CMAQ) with a dispersion model (AERMOD) to estimate ambient concentrations of about 50 of the more prevalent and higher risk HAPs as described in Section 3.5. The HAPs modeled in the hybrid approach capture a vast majority of the total risk nationally. Treatment for the remaining “non-hybrid” HAPs are described in Section 3.1.2. 3.1. Modeling Overview 3.1.1. Photochemical Model Selection For AirToxScreen photochemical modeling, we used CMAQv5.3.2 (https://zenodo.org/record/4081737#.YwO1KXbMI2w; https://www.epa.gov/cmaq/cmaq-models-0; http://www.cmascenter.org.) with the Carbon-Bond 6r3 (CB6r3-CMAQ) chemical mechanism, AERO7 aerosol module with non-volatile Primary Organic Aerosol (POA). CMAQ is a comprehensive, three- dimensional grid-based Eulerian air quality model designed to simulate the formation and fate of gaseous and particulate species, including ozone, oxidant precursors, primary and secondary PM concentrations, and sulfur and nitrogen deposition over urban and regional spatial scales. CMAQ includes numerous science modules that simulate the emission, production, decay, deposition and transport of organic and inorganic gas-phase and pollutants in the atmosphere (Appel et al., 2018). While most compounds are grouped when model chemistry is applied, the CB6r3-CMAQ chemical mechanism treats formaldehyde, acetaldehyde, benzene, methanol, and naphthalene explicitly within the chemistry. In this version, xylene concentrations are aggregated across the xylene isomers rather than separately generating concentrations for each isomer. For more information on CMAQ, see https://www.epa.gov/air-research/community-multi-scale-air-quality-cmaq-modeling-system-air-quality- management or http://www.cmascenter.org. Table 3-1 lists HAPs included in the multipollutant version of CMAQv5.3.1 used for AirToxScreen. Table 3-1. CMAQ HAPs Air Toxic CMAQ Species Name(s) 1,1,2,2-Tetrachloroethane CL4_ETHANE 1,3-Butadiene BUTADIENE13 1,3-Dichloropropene DICL_PROPENE 1,4-Dichlorobenzene(p) DICL_BENZENE 2,4-Toluene diisocyanate TOL_DIIS Acetaldehyde ALD2, ALD2_PRIMARY Acetonitrile ACET_NITRILE Acrolein ACROLEIN, ACROLEIN_PRIMARY Acrylic acid ACRYACID AirToxScreen 2018 Documentation 57 Air Toxic CMAQ Species Name(s) Acrylonitrile ACRY_NITRILE Arsenic AASI, AASJ, ASSK Benzene (including benzene from gasoline) BENZENE Benzo-A-Pyrene ABENAPYI, ABENAPYJ, BENAPY Beryllium ABEK, ABEI, ABEJ Cadmium ACDI, ACDJ, ACDK Carbon tetrachloride CARBONTET Carbon tetrachloride (without background) CARB_TET_NBC Carbonyl sulfide CARBSULFIDE Chlorine CL2 Chloroform CHCL3 Chloroprene CHLOROPRENE Hexavalent Chromium Compounds ACR_VIK, ACR_VIJ, ACR_VII Trivalent Chromium Compounds ACR_IIIK, ACR_IIII, ACR_IIIJ Diesel PM ADE_ECI, ADE_ECJ, ADE_OCI, ADE_OCJ, ADE_SO4J, ADE_NO3J, ADE_OTHRI, ADE_OTHRK, ADE_K Ethyl benzene ETHYLBENZENE Ethylene dibromide (Dibromoethane) BR2_C2_12 Ethylene dichloride (1,2-Dichloroethane) CL2_C2_12 Ethylene oxide ETOX Formaldehyde FORM, FORM_PRIMARY Hexamethylene-1,6-diisocyanate HEXMETH_DIS Hexane HEXANE Hydrazine HYDRAZINE Hydrochloric acid HCL Lead Compounds APBK, APBJ, APBI Maleic anhydride MAL_ANHYDRID Manganese Compounds AMN_HAPSK, AMN_HAPSJ, AMN_HAPSI Mercury Compounds HG, HGIIGAS, APHGI, APHGJ, APHGK Methanol MEOH Methyl chloride (Chloromethane) METHCHLORIDE Methylene chloride (Dichloromethane) CL2_ME m-xylene, p-xylene, o-xylene and xylenes XYLENE (isomers and mixture) Naphthalene NAPHTHALENE Nickel Compounds ANIK, ANII, ANIJ Polycyclic Organic Matter PAH_000E0 Polycyclic Organic Matter PAH_176E5 Polycyclic Organic Matter PAH_880E5 Polycyclic Organic Matter PAH_176E4 AirToxScreen 2018 Documentation 58 Air Toxic CMAQ Species Name(s) Polycyclic Organic Matter PAH_176E3 Polycyclic Organic Matter PAH_192E3 Polycyclic Organic Matter PAH_101E2 Polycyclic Organic Matter PAH_176E2 Polycyclic Organic Matter PAH_114E1 Propylene dichloride (1,2-Dichloropropane) PROPYL_DICL Quinoline QUINOLINE Styrene STYRENE Tetrachloroethylene (Perchloroethylene) CL4_ETHE Toluene TOLU Trichloroethylene CL3_ETHE Triethylamine TRIETHYLAMIN Vinyl chloride CL_ETHE 3.1.2. Dispersion Model Selection For AirToxScreen air dispersion modeling, we used AERMOD (Cimorelli et al. 2005), a steady-state plume model that incorporates air dispersion based on planetary boundary layer turbulence structure and scaling concepts. AERMOD is EPA’s preferred near-field modeling system of emissions for distances up to 50 km (EPA 2017). AERMOD version 21112 was used for all AirToxScreen rungroups. 3.2. Meteorological Data 3.2.1. Meteorological Data Inside the Contiguous States 3.2.1.1. Input WRF Data For use in all AirToxScreen modeling, we derived gridded meteorological data for the contiguous United States (CONUS) from version 3.8 of the Weather Research and Forecasting Model (WRF), Advanced Research WRF (ARW) core (Skamarock et al. 2008). The WRF Model is a state-of-the-science mesoscale numerical weather prediction system developed for both operational forecasting and atmospheric research applications (http://wrf-model.org/). The CONUS WRF simulation used the same 12-km CMAQ map projection, a Lambert Conformal projection centered at coordinates (-97, 40) with true latitudes at 33 and 45 degrees north. The 12-km WRF domain consisted of 459 by 299 grid cells and 35 vertical layers up to 50 millibars. The 12-km CONUS WRF model was initialized using the 12-km North American Model (12NAM; Only, http://nomads.ncdc.noaa.gov/data.php; download from ftp://nomads.ncdc.noaa.gov/NAM/analysis_only/) analysis product provided by National Climatic Data Center (NCDC). Where 12NAM data was unavailable, the 40-km Eta Data Assimilation System (EDAS) analysis (ds609.2) from the National Center for Atmospheric Research (NCAR) was used. Analysis nudging for temperature, wind, and moisture was applied above the boundary layer only. The model simulations were conducted continuously. The ‘ipxwrf’ program was used to initialize deep soil moisture at the start of the run using a 10-day spin-up period. The 2018 WRF meteorology simulated was based on 2011 National Land Cover Database (NLCD; National Land Cover Database 2011, http://www.mrlc.gov/nlcd2011.php). The WRF simulation included the physics options of the AirToxScreen 2018 Documentation 59 Pleim-Xiu land surface model (LSM), Asymmetric Convective Model version 2 planetary boundary layer (PBL) scheme, Morrison double moment microphysics, Kain- Fritsch cumulus parameterization scheme utilizing the moisture-advection trigger (Ma and Tan 2009) and the RRTMG long-wave and shortwave radiation (LWR/SWR) scheme (Gilliam and Pleim, 2010). These simulations also incorporated observed lightning data from the National Lightning Detection Network (NLDN) to suppress (force) deep convection where lightning is absent (present) (Heath et al. 2016). In addition, the Group for High Resolution Sea Surface Temperatures (GHRSST; Stammer et al. 2003, analysis, https://www.ghrsst.org/) 1-km SST data was used for SST information to provide more resolved information compared to the more coarse data in the NAM analysis. 3.2.1.2. MCIP Processing for CMAQ The 2018 WRF meteorological outputs were processed using the Meteorology-Chemistry Interface Processor (MCIP) package (Otte and Pleim 2010), version 4.4, to derive the specific inputs to CMAQ: horizontal wind components (i.e., speed and direction), temperature, moisture, vertical diffusion rates and rainfall rates for each grid cell in each vertical layer. Table 3-2 shows the vertical layer structure used in WRF and the CMAQ meteorological inputs. Table 3-2. Vertical layer structure for WRF and CMAQ (heights are layer top) WRF & CMAQ Layers Sigma P Approximate Height (m) 35 0.0000 17,556 34 0.0500 14,780 33 0.1000 12,822 32 0.1500 11,282 31 0.2000 10,002 30 0.2500 8,901 29 0.3000 7,932 28 0.3500 7,064 27 0.4000 6,275 26 0.4500 5,553 25 0.5000 4,885 24 0.5500 4,264 23 0.6000 3,683 22 0.6500 3,136 21 0.7000 2,619 20 0.7400 2,226 19 0.7700 1,941 18 0.8000 1,665 17 0.8200 1,485 16 0.8400 1,308 15 0.8600 1,134 14 0.8800 964 13 0.9000 797 AirToxScreen 2018 Documentation 60 WRF & CMAQ Layers Sigma P Approximate Height (m) 12 0.9100 714 11 0.9200 632 10 0.9300 551 9 0.9400 470 8 0.9500 390 7 0.9600 311 6 0.9700 232 5 0.9800 154 4 0.9850 115 3 0.9900 77 2 0.9950 38 1 0.9975 19 0 1.0000 0 3.2.1.3. MMIF Processing for AERMOD WRF output was processed through the Mesoscale Model Interface (MMIF) program to create AERMET- ready meteorological input data and processed in AERMET (version 19191) MMIF was processed for AERMOD in accordance with EPA guidance, Guidance on the Use of the Mesoscale Model Interface Program (MMIF) for AERMOD Applications (U.S. EPA 2022). As options, we used the default FLM layers (see MMIF User’s Guide for details), TOP for vertical interpolation and MMIF-calculated mixing heights. AERMET (version 21112) was run with the adjusted u* option to better represent concentrations in AERMOD under low-wind stable conditions. 3.2.2. Meteorological Data Outside the Contiguous States 3.2.2.1. Input WRF Data The 2018 gridded meteorological data covering areas outside the contiguous states, the 3-km Hawaii and Puerto Rico/Virgin Islands and the 9-km Alaska domains, was derived from the publicly available WRF version 4.1.1, ARW core (https://www2.mmm.ucar.edu/wrf/users/). Grid Domain Specifications: • Alaska with 9 km horizontal grid spacing (312 cells in the x-direction, 252 cells in the y-direction) with 35 layers centered on 63° N and 155° W. • Hawai’i with 3 km horizontal grid spacing (225 cells in the x-direction, 201 cells in the y- direction) with 35 layers centered on 21° N and 157° W • Puerto Rico and Virgin Islands with 3 km horizontal grid spacing (150 cells in the x-direction, 150 cells in the y-direction) with 35 layers centered on 18 ° N and 66° W. The WRF model was initialized using the 0.25-degree Global Forecast System (GFS) model analysis (NCEP/NWS/NOAA/USDC, 2015, https://doi.org/10.5065/D65D8PWK. Analysis nudging for AirToxScreen 2018 Documentation 61 temperature, wind, and moisture was applied above the boundary layer only. The model simulations were conducted continuously. The ‘ipxwrf’ program was used to initialize deep soil moisture at the start of the run using a 10-day spin-up period. The 2018 WRF meteorology simulated was based on IGBP-Modified MODIS 20-category Land Use data. The WRF simulation included the physics options indicated in Table 3-3. In addition, the Group for High Resolution Sea Surface Temperatures (GHRSST; Stammer et al. 2003, analysis, https://www.ghrsst.org/) 1-km SST data was used for SST information to provide more resolved information compared to the more coarse data in the NAM analysis. Table 3-3. WRF Options Used for the Alaska, Hawaii, and Puerto Rico/Virgin Islands domains. Option Settings for ALASKA Settings for HI & PR/VI Cumulus Kain-Fritsch (w/ Trigger 1) w/ Kain-Fritsch (w/ Trigger 1) w/ cu_rad_feedback = .true. cu_rad_feedback = .true. Precipitation Hourly Incremental Hourly Incremental PBL MYNN2 YSU LSM MYNN Noah Microphysics New Eta (Ferrier, mp_physics = 5) New Eta (Ferrier, mp_physics = 5) Longwave and RRTMG RRTMG Shortwave radiation Model Top 50 mb 50 mb 3.2.2.2. MCIP Processing for CMAQ The 2018 WRF meteorological outputs for the the 3-km Hawaii and Puerto Rico/Virgin Islands and the 9-km Alaska domains were processed using the Meteorology-Chemistry Interface Processor (MCIP) package (Otte and Pleim 2010), version 4.4, to derive the specific inputs to CMAQ: horizontal wind components (i.e., speed and direction), temperature, moisture, vertical diffusion rates and rainfall rates for each grid cell in each vertical layer. Table 3-2 shows the vertical layer structure used in WRF and the CMAQ meteorological inputs. 3.2.2.3. MMIF Processing for AERMOD For meteorological data covering areas outside the contiguous states (Alaska, Hawaii, Puerto Rico and the U.S. Virgin Islands), we used WRF data processed through the AERMET program (version 21112) specific to each of the areas. For MMIF processing to AERMET, all three domains passed through MMIF calculated mixing heights using the MMIF guidance heights (25, 50, 75, 100, 125, 150, 175, 200, 250, 300, 350, 400, 450, 500, 600, 700, 800, 900, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 4500, and 5000) using the ‘TOP’ interpolation technique. The minimum wind speed threshold was set to 0.0 m/s (AER_MIN_SPEED = ‘0.0’) and the AERMET adjust u* option was used. AirToxScreen 2018 Documentation 62 3.3. CMAQ Setup 3.3.1. Sources Modeled in CMAQ AirToxScreen CMAQ modeling included a base year run (primary and secondary annual average concentrations) and “zero-out” runs for biogenics and fires (primary annual average concentrations). The annual simulations included a “ramp-up” period of 10 days to mitigate the effects of initial concentrations. All 365 model days were used in the annual average levels of air toxics modeled. The model extends vertically from the surface to 50 millibars (approximately 17,600 meters) using a sigma- pressure coordinate system. The CMAQ model runs were performed for four domains covering: (1) Contiguous United States (CONUS), as shown in Figure 3-1(a). This single domain covers the entire CONUS and large portions of Canada and Mexico using 12 km by 12 km horizontal grid spacing. (2) Alaska (Figure 3-1(b)) with 9 km by 9 km horizontal grid spacing. (3) Hawai’i (Figure 3-1(c)) with 3 km by 3 km horizontal grid spacing. (4) Puerto Rico and Virgin Islands (Figure 3-1(d)) with 3 km by 3 km horizontal grid spacing. Table 3-4 provides some basic geographic information regarding the four CMAQ domains. Air quality conditions at the outer boundary of the four CMAQ domains were taken from a global model (described in Section 3.3.2). AirToxScreen 2018 Documentation 63 Figure 3-1(a). Map of the CMAQ modeling domain; the blue box denotes the 12 km CONUS modeling domain AirToxScreen 2018 Documentation 64 Figure 3-2(b). Map of the CMAQ modeling domain; the green box denotes the 9 km Alaska modeling domain AirToxScreen 2018 Documentation 65 Figure 3-3(c). Map of the CMAQ modeling domain; the red box denotes the 3 km Hawai’i modeling domain AirToxScreen 2018 Documentation 66 Figure 3-4(d). Map of the CMAQ modeling domain; the red box denotes the 3 km Puerto Rico/Virgin Islands modeling domain Table 3-4. Geographic information for the CMAQ modeling domainsNSS CMAQ Modeling Domain Configurations Domain Map Projection Grid Coordinate True Latitudes Dimensions & Resolution Center Vertical Extent Continental US Lambert Conformal 12 km 97 W, 40 N 33 and 45 N 459 x 299 x 35 Projection Alaska Lambert Conformal 9 km 63 N, 155 W 60 and 70 N 312 x 252 x 35 Projection Hawai’i Lambert Conformal 3 km 21 N, 157 W 19 and 22 N 225 x 201 x 35 Projection Puerto Rico/Virgin Lambert Conformal 3 km 18 N, 66 W 17 and 19 N 150 x 150 x 35 Islands Projection AirToxScreen 2018 Documentation 67 3.3.2. Boundary and Initial Conditions The lateral boundary and initial species concentrations for the four CMAQ modeling domains (i.e., CONUS, Alaska, Hawaii and Puerto Rico/Virgin Islands) were provided by a global application of GEOS-Chem (Bey et al. 2001) to the year 2018. The global-scale platform uses v12.0.1 (doi: 10.5281/zenodo.1403144) at 2 x 2.5-degree resolution (half cells at the north and south pole) with year-specific meteorology for 2018. Meteorology is provided by the Goddard Earth Observation System Forward Product (GEOS_FP) using 72 vertical layers from the surface to 0.01 hPa. The GEOS- Chem standard chemistry option was applied at all levels. Emissions were provided using the standard configuration of the Harmonized Emissions Component (HEMCO; Keller et al. 2014) distributed with v12.0.1 of GEOS-Chem. The biogenic, soil NOx, dust, and sea salt emissions were all calculated online using the GEOS-FP 2018 emissions. At the time this simulation was run, GEOS- Chem did not have updated fire emissions for 2018. For this run, 2016 monthly fire emissions were used as a climatology. Note this would have larger implications for applying the model over the continental US, but less so for lateral boundary conditions. The anthropogenic configuration uses a combination of the Community Emissions Data System (CEDS; Hoesly et al. 2018) from the 2014 year with regional inventories superseding CEDS. For example, regional inventories are used over the United States, Mexico, Canada, and Europe. These inventories have differing base years and are projected to a nominal present day using either International Energy Agency fuel statistics or using regional emission trends. In addition to the standard chemistry, we also included Persistent Organic Pollutants (POPs) and mercury from specialized applications of GEOS-Chem. The POPs model configuration (Friedman et al. 2014) in v12.0.1 for 2018 applied to BaP, PHE, and PYR using 4x5 degree resolution. For mercury, we applied the mercury model configuration (Selin et al., 2007) from a 2013 v9 simulation with 47 layers and 2x2.5 degree resolution. The mercury concentrations were averaged to months and used as a climatology. Where CMAQ species were not available from either the standard chemistry, the POPs or mercury simulation, then they were treated as climatology using values from previous multipollutant boundary condition applications. Because GEOS-Chem does not include all modeled HAPs, we also used remote concentration estimates as nonvarying background (in space and time). These were computed based on data from the five NOAA GMD sites: Cape Kumukahi, Hawaii (KUM); Mauna Loa, Hawaii (MLO); Niwot Ridge, Colo. (NWR); Barrow, Alaska (BRW); and Alert, Canada (ALT) and the Trinidad Head Site (AGAGE). More information on how these were derived is in Appendix C. For the remaining CMAQ HAP BCs not provided by H-CMAQ nor estimated using remote concentrations listed in Table 3-4, a value of zero was applied (shown in Table 3-5) due to a lack of data. Table 3-4. Boundary conditions from 2018 remote concentration estimates RCE at 298K and 1 Pollutant atm (μg/m3) RCE (pptv) Remote 2018 2018 Network Location(s) CMAQ HAP Chloroform 0.082 16.9 AGAGE Trinidad Head x AirToxScreen 2018 Documentation 68 Methyl chloride 1.151 55 8 NOAA GMD KUM, MLO, NWR, BRW, x (chloromethane) ALT Carbon tetrachloride 0.509 81.0 NOAA GMD KUM, MLO, NWR, BRW, x ALT Methyl bromide 0.027 7.0 NOAA GMD KUM, MLO, NWR, BRW, (bromomethane) ALT Methyl chloroform (1,1,1- 0.012 2.2 NOAA GMD KUM, MLO, NWR, BRW, trichloroethane) ALT Dichloromethane 0.220 63.4 NOAA GMD KUM, MLO, NWR, BRW, x (methylene chloride) ALT Tetrachloroethene 0.013 1.8 NOAA GMD KUM, MLO, NWR, BRW, x (perchloroethylene, ALT tetrachloroethylene) Table 3-5. CMAQ HAP boundary conditions applied as zero value Air Toxic 2018 CMAQ Species Name(s) 1,1,2,2-Tetrachloroethane CL4_ETHANE 1,3-Dichloropropene DICL_PROPENE 1,4-Dichlorobenzene(p) DICL_BENZENE 2,4-Toluene diisocyanate TOL_DIIS Acetonitrile ACET_NITRILE Acrylic acid ACRYACID Acrylonitrile ACRY_NITRILE Arsenic AASI, AASJ, ASSK Beryllium ABEK, ABEI, ABEJ Benzo-A-Pyrene ABENAPYI, ABENAPYJ, BENAPY Cadmium ACDI, ACDJ, ACDK Carbon tetrachloride CARB_TET_NBC Carbonyl sulfide CARBSULFIDE Chlorine CL2 Chloroprene CHLOROPRENE Chromium Compounds ACR_VIK, ACR_VIJ, ACR_VII Chromium Compounds ACR_IIIK, ACR_IIII, ACR_IIIJ Diesel PM* ADE_ECI, ADE_ECJ, ADE_OCI, ADE_OCJ, ADE_SO4J, ADE_NO3J, ADE_OTHRI, ADE_OTHRK, ADE_K Ethyl benzene ETHYLBENZENE Ethylene dibromide (Dibromoethane) BR2_C2_12 Ethylene dichloride (1,2-Dichloroethane) CL2_C2_12 Ethylene oxide ETOX Hexamethylene-1,6-diisocyanate HEXMETH_DIS Hexane HEXANE Hydrazine HYDRAZINE AirToxScreen 2018 Documentation 69 Air Toxic 2018 CMAQ Species Name(s) Lead Compounds APBK, APBJ, APBI Maleic anhydride MAL_ANHYDRID Manganese Compounds AMN_HAPSK, AMN_HAPSJ, AMN_HAPSI Mercury Compounds APHGI, APHGK m-xylene, p-xylene, o-xylene and xylenes XYLENE (isomers and mixture) Naphthalene NAPHTHALENE Nickel Compounds ANIK, ANII, ANIJ Polycyclic Organic Matter PAH_000E0 Polycyclic Organic Matter PAH_176E5 Polycyclic Organic Matter PAH_880E5 Polycyclic Organic Matter PAH_176E4 Polycyclic Organic Matter PAH_176E3 Polycyclic Organic Matter PAH_192E3 Polycyclic Organic Matter PAH_101E2 Polycyclic Organic Matter PAH_176E2 Polycyclic Organic Matter PAH_114E1 Propylene dichloride (1,2-Dichloropropane) PROPYL_DICL Quinoline QUINOLINE Styrene STYRENE Trichloroethylene CL3_ETHE Triethylamine TRIETHYLAMIN Vinyl chloride CL_ETHE 3.4. AERMOD Setup 3.4.1. Sources Modeled in AERMOD AERMOD modeling comprised point, nonpoint, on-road and nonroad sources. We excluded fires (agricultural burning, wildfires and prescribed fires) and biogenic emissions. 3.4.2. Receptor Placement For the CONUS domain, we used the following receptors: 1. Equally spaced “gridded” receptors (1 km in highly populated areas, 4 km otherwise) 2. Populated census-block centroid receptors (discussed in Section 3.4.2.2) 3. Monitoring site receptors (discussed in Section 3.4.2.3) AirToxScreen 2018 Documentation 70 For non-CONUS areas (AK, HI, PR, VI), we used: 1. Equally spaced “gridded” receptors (1 km in Hawaii and Puerto Rico/Virgin Islands, 3 km in Alaska) 2. Census-block centroid receptors (populated in all domains and non-populated in Alaska domain) 3. Monitoring site receptors To facilitate the CMAQ/AERMOD hybrid modeling, we used these receptors to compute an AERMOD average concentration corresponding to each CMAQ grid cell. These concentrations could then be used in the hybrid equation. Gridded receptors were also used for interpolating to block receptors and monitor receptors. Receptors at the monitoring locations were used in the model evaluation. 3.4.2.1. Gridded receptors We used gridded receptors throughout the CONUS area and the three non-CONUS areas. The purposes of the gridded receptors were to provide a uniform grid in the CMAQ grid cells to adequately capture near-field concentration gradients of sources in and surrounding the grid cells and to provide a grid used to interpolate census block and monitor receptors post-modeling. For the CONUS domain, we based spacing of the gridded receptors on the 2013 populations of Core Base Statistical Areas (CBSA). For CMAQ grid cells that intersected a CBSA with a population of 1 million people or more (Figure 3-2), the receptors were placed 1 km apart, resulting in 144 receptors per 12-km CMAQ grid cell. Otherwise, the receptors were placed 4 km apart, resulting in nine receptors per 12-km CMAQ grid cell. This resulted in 1.4 million receptors nationwide, with 1-km spacing used in 6,935 12-km cells and 4-km spacing used in 49,490 12-km cells. Correct receptor placement was verified by overlaying receptors with CBSA’s in ArcGIS. AirToxScreen 2018 Documentation 71 Figure 3-5. CBSAs exceeding 1 million people Each gridded receptor, with either 1- or 4-km spacing, represented the center of a subgrid cell within the 12-km CMAQ grid cells (Figure 3-3). For the Hawaii and Puerto Rico domains, grid spacing was uniform throughout the domains at 1 km within each grid cell. The Alaskan domain used a uniform grid spacing of 3 km within each grid cell. These gridded receptors, plus populated block and monitor receptors when available within a subgrid cell, were averaged (Figure 3-4). These subgrid-cell averages were then used to calculate the overall AERMOD average of the 12-km grid cell, which was then used in the hybrid equation that combines CMAQ and AERMOD results. Receptor elevations and hill heights were determined using AERMAP (version 18081). AirToxScreen 2018 Documentation 72 0 0 0 0 0 0 0 0 0 Figure 3-6. Dense (left) and coarse (right) receptor grid layout in CMAQ Lambert Projection Figure 3-7. Example grid cell with subgrid cells and census blocks When performing the dispersion modeling for point sources, gridded receptors within 50 km of any emission point at the facility were explicitly modeled in AERMOD. For airports, any gridded receptor within 50 km of any point along a runway or within 50 km of any part of the 100-by-100 m area sources AirToxScreen 2018 Documentation 73 was explicitly modeled. For gridded sources (all domains), any gridded receptor within 50 km of the center of the gridded source was explicitly modeled. For ports and underway sources, any gridded receptor that was within 50 km of a side of the polygon or within 50 km of the center of the source was explicitly modeled. 3.4.2.2. Census-block centroid receptors The locations of census-block centroids were based on the 2010 U.S. Census (2010). When performing the dispersion modeling for point sources in all domains, populated block receptors within 10 km of any emission point at the facility were explicitly modeled in AERMOD. For airports in all domains, any populated block receptor within 10 km of any point along a runway or 10 km from any point of the 100- by-100 m area sources was modeled. For gridded sources in all domains, census blocks were not modeled; these were later interpolated from gridded receptors during model post-processing. For ports and underway sources in all domains, any block receptor within 10 km of a side of the polygon or within 10 km of the center of the source was explicitly modeled. For the Alaska, point, airports, etc., non- populated blocks outside the CMAQ domain but within 50 km of a souce were modeled as receptors as well. Non-populated blocks were not modeled in the CONUS area. 3.4.2.3. AERMOD receptors at monitoring sites The Ambient Monitoring Archive for HAPs monitoring data sites was used for the model evaluation (found in the Supplemental Data folder on the AirToxScreen website). Therefore, we obtained a unique set of geographic coordinates for all monitors to include as receptors in the AERMOD modeling. AERMAP (version 18081) was used to generate receptor elevations and hill heights for input into AERMOD. The modeling distance criteria used for census blocks was used for the monitor receptors. 3.4.3. Model Options For all AERMOD runs, the FASTALL option was used to decrease model runtimes, especially for gridded, airport, and CMV shapes. For all AERMOD runs excluding point and airport sources, the FLAT option was used (terrain ignored). For sources determined to be urban, the AERMOD urban option was used. 3.4.4. AERMOD Simulations We ran AERMOD for each of the run groups described in Table 2-11. For the run groups that included the gridded source types in the CONUS area, gridded sources were also used as the sources in the non- CONUS areas. Receptor placement was as described in Section 3.4.2, and each source was assigned the meteorological files corresponding to the WRF grid cells in which they were located for the gridded sources. For the point sources and airports, each facility or airport was run in its own AERMOD run. For the point sources, each AERMOD source in the facility was its own AERMOD source group with the source group name corresponding to the AERMOD source ID. For each airport, all emissions sources were assigned to a total group (group ALL). Similarly, for the CMV runs and gridded sources, each CMV shape or grid cell was run in its own AERMOD run. The CMV emissions sources in each AERMOD run were assigned to groups based on the source. For underway emissions, a single group representing all three ship types was used. For the port emissions, groups were assigned: 1) a group for C1 and C2 and 2) a C3 group. For AirToxScreen 2018 Documentation 74 the 12-km gridded sources and non-CONUS gridded sources, the source was assigned to source group with the same name as the AERMOD source ID. For the 4-km gridded sources, all sources that shared the same parent 12-km grid cell and thus the same meteorological data were run together in one AERMOD run. Each 4-km source was given its own source group in the AERMOD output. As discussed in Section 2.3 for the various run groups, different temporalization resolutions were used (hourly, monthly, etc.). The scalars from the temporal helper files were used in conjunction with the unit emission rate of 10,000 tons/year to develop hourly unit emissions for the modeled sources. For hourly emissions, this was simply the product of the hourly scalar and 10,000 tons/year, then applying a conversion factor of 251.99577 to yield an emission rate in g/s. If a source was an hourly source, the resulting emission rate was divided by the area of the source to yield the correct units for AERMOD. For sources using monthly emission factors, a weighting factor was calculated by multiplying each monthly scalar by the number of days for the month. These products were summed across all 12 months. The hourly emission rate E for each month i was then calculated as: 10000 × �EFi� E Weight� i= × 251.9957 24 Where EFi is the monthly scalar factor from the temporal helper file and 10,000 is the unit emission rate in tons/year. If the source was an area source, Ei was divided by the area of the source. Since each hour for a particular month has the same emission rate, the AERMOD EMISFACT keyword was used to represent the hourly emissions. The emission rate calculated above was used for the factor, and the base emission rate on the SRCPARAM line was set to 1.0 g/s or 1.0 g/s/m2 for area sources. For emissions that varied by hour of day only, the AERMOD EMISFACT HROFDY was used and the hourly emission rate Ei for hour i (1–24) was calculated as: 10000 × EF Ei= i × 251.9957 365 Where 365 represents the number of days per year. Area sources’ emission rates were divided by the area of the source. For emissions that varied by month, day of the week, MHRDOW or MHRDOW7, and hour of day, the month weighting factor was applied in the temporal helper file. The hourly emission rate applied in AERMOD was calculated as the product of 10,000 tons/year, the emission factor, and the conversion factor of 215.9957. As with the other emission factors, area sources’ emission rates were divided by the area of the source. 3.4.5. Post-processing of AERMOD Results Post-processing of the AERMOD runs consisted of two steps: 1) interpolation of populated census blocks and monitors from gridded receptors and 2) calculation of HAP-specific concentrations. For all AirToxScreen run groups except the gridded sources, interpolation of concentrations to census blocks and monitors occurred at blocks and monitors from 10 to 50 km from the source. For the gridded source 7215.9957 is the conversion factor from tons/hr to g/s. AirToxScreen 2018 Documentation 75 types, census blocks and monitors were interpolated 0 to 50 km from the sources. Interpolation was done for each individual facility (point source, airport, CMV shape) or gridded source. For a receptor located between four gridded receptors, the concentration at the receptor was based on linear interpolation between the four gridded receptors. For receptors near the 50-km edge of the modeling domain and thus without four receptors around it, the interpolated receptor’s concentration was the average of the nearest gridded receptors. For nonmobile AirToxScreen run groups (point, airports, CMV, nonpoint, RWC, oil and gas, non-CONUS nonroad) the HAP/source group-specific concentrations at each receptor were calculated by dividing the HAP emissions (tons/year) for each source group at each AERMOD source by 10,000 tons/year. For the mobile run groups, the HAP/source group-specific concentrations were calculated based on monthly emissions. The AERMOD output for these rungroups was average concentrations by month. AERMOD also output the hours that had missing or calm meteorological data. Only hours that were noncalm and not missing in the meteorological data were included in the averaging, consistent with the calculation of long-term concentrations in AERMOD. The following methodology was used to calculate the annual average HAP/source group specific concentrations at each receptor: 1. Multiply the concentration from the monthly output file by the number of valid hours for the particular month based on review of the AERMOD errors file. Add the resulting product to a running total of concentration for each month and modeled source group. 2. Interpolate the results from step 1 to unmodeled census/blocks and monitors within 50 km for each facility or gridded source. 3. For each season, multiply the total calculated in step 1 and 2 by the ratio of the source group’s monthly HAP emissions (tons/month) to the modeled monthly emissions (based on 10,000 tons/year) listed in the AERMOD output file. 4. Add the result from step 3 to a running total across all months. Also, for each month, loop through the months to determine the total number of valid hours for the year. If all meteorological data hours were noncalm and not missing, this results in a total of 8,760 hours. The number may be less if calms and missing data are present in the meteorological data. 5. Divide running total concentration from step 4 by the running total of hours from step 4 to calculate an annual average HAP/source group-specific concentration. This division of concentration by hours is equivalent to how AERMOD calculates annual average concentrations in a simulation, a sum of hourly concentrations divided by the number of noncalm and nonmissing hours. Once HAP/source group-specific concentrations are calculated for each gridded receptor, block receptor and monitor receptor, the resulting concentrations were output for input into the hybrid program. 3.5. Hybrid Modeling The hybrid approach combines the annual concentration results from the AERMOD and CMAQ models to compute ambient concentrations at census block receptors. The subsections below contain AirToxScreen 2018 Documentation 76 discussions on the hybrid air modeling approach used for AirToxScreen for the CONUS and three non- CONUS domains. 3.5.1. Overview For 58 of the most prevalent and highest risk air toxics (see Table 3-1), we used a hybrid air quality modeling method combining the fine spatial scale and source attributions of AERMOD (Cimorelli et al. 2005; EPA 2015e) with the full treatment of chemistry and transport afforded by CMAQ. In this application, AERMOD treated all species as chemically nonreactive. The emissions and meteorological data sets used in CMAQ were processed further to generate AERMOD inputs consistent with CMAQ. AERMOD receptor locations were based on the centroids of populated census blocks, monitoring-site positions, and evenly distributed points within each horizontal CMAQ grid cell in the four domains C (see Figure 3-1 and Figure 3-5 for an example in the CONUS domain), resulting in at least nine, and sometimes more than 10,000, receptors per cell and 6.5 million receptors nationwide in the CONUS domain. Figure 3-8. CMAQ domain with expanded cell showing hybrid receptors; colors indicate modeled concentrations; dots in inset show locations of receptors within a grid cell The equation below was used to calculate the annual average estimates of air concentrations at receptor locations, which were constrained to CMAQ-grid-average values, with AERMOD providing subgrid-scale spatial texture. CMAQ C = AERMOD PNFB REC × � � + CMAQ AERMOD SEC + CMAQPFIRES + CMAQPBIOGENICS + CMAQBACKGROUND GRIDAVG Where: C = concentration at a receptor CMAQPNFB = concentration in CMAQ grid cell, contributed by primary emissions, excluding fires and biogenics AirToxScreen 2018 Documentation 77 AERMODREC = concentration at AERMOD receptor AERMODGRIDAVG = average of all AERMOD results within a CMAQ grid, calculated through averaging the receptors in each of the nine 4-km cells in a CMAQ grid cell, and then taking the average of the nine grid cells CMAQSEC = contribution from atmospheric reactions in CMAQ grid cell CMAQPFIRES = contribution from primary emissions of fires in CMAQ grid cell CMAQPBIOGENICS = contribution from primary emissions of biogenics in CMAQ grid cell CMAQBACKGROUND = contribution from background in CMAQ grid cell for carbon tetrachloride This hybrid approach, which builds on earlier area-specific applications to Philadelphia, Pa. (Isakov et al. 2007) and Detroit, Mich. (Wesson et al. 2010), attempts to characterize nonreactive and reactive species across multiple spatial scales. However, the blending of two different modeling platforms challenges adherence to basic mass-conservation principles. To address this issue, CMAQ tracks primary and secondary contributions by source type, enabling the AERMOD estimate at each receptor location to be normalized to the CMAQ primary contribution. Anchoring concentration averages to CMAQ largely retains mass conservation. The constraint to CMAQ average grid values imposed by the above equation minimized possible redundancies and allows us to combine results from these two very different models. 3.5.2. Treatment of Species As noted above, we applied the hybrid model to 58 of the highest risk air toxics (shown in Table 3-1) among the 181 air toxics included in AirToxScreen. CMAQ bases its treatment of atmospheric chemistry on gas-phase reaction processes optimized to characterize ozone, linked with heterogeneous and thermodynamic processes for PM formation. This structure allows us to model explicit chemical species. For example, formaldehyde and acetaldehyde generate significant amounts of peroxy radicals, which lead to enhanced ozone production and secondary PM formation. This demonstrates the multipollutant linkages driven by atmospheric processes that CMAQ can simulate. Chemical species not incorporated as explicit species in chemical mechanisms are added as nonreactive tracers (e.g., several halogenates) or included in simple reaction schemes, such as 1,3-butadiene decay and subsequent acrolein generation, decoupled from the chemical mechanism. The emissions mass of several less reactive VOCs, such as the prevalent benzene, toluene and xylene species, are tracked as nonreactive tracers. CMAQ treats these as lumped carbon bond species in its reaction calculations, assuming that atmospheric chemistry minimally influences air concentrations. AERMOD, which treats all pollutants as nonreactive, was applied to the remaining air toxics not incorporated within CMAQ. The calculation of the AERMOD grid cell average, AERMODGRIDAVG in the hybrid equation is a three-step process: 1. Calculate a total AERMOD concentration at each gridded receptor, block receptor and monitor receptor by reading the individual post-processed run group concentrations and adding to a running total. AirToxScreen 2018 Documentation 78 2. Calculate the average concentration at each subgrid cell centered on each gridded receptor in each CMAQ grid cell (see Section 3.4.2.1 for details on gridded receptors and example subgrid cells). This results in either 9 or 144 subgrid-cell averages in each CMAQ cell in the CONUS domain or 9 averages in the non-CONUS domains. 3. Average the 9 or 144 averages in each grid cell to calculate an overall average for the grid cell, AERMODGRIDAVG. After calculating AERMODGRIDAVG the following steps are taken: 1. For secondary HAPs (acetaldehyde, formaldehyde and acrolein), calculate the secondary concentration by subtracting the CMAQ primary concentration from the total CMAQ concentration at each grid cell. For nonsecondary HAPs, secondary concentrations are zero. 2. For fire and biogenic HAPs, calculate the primary concentration at each grid cell by subtracting the no-fire/no-biogenic CMAQ concentration from the base CMAQ concentration, where base concentration includes all sources. For nonfire and nonbiogenic HAPs, secondary concentrations are zero. 3. At each grid cell, calculate the primary anthropogenic concentrations by subtracting primary fire and primary secondary concentrations from the total primary concentration from the base CMAQ results. 4. Divide the CMAQ primary concentration from step 3 by the AERMODGRIDAVG and multiply by the AERMOD total concentration at each receptor in the CMAQ grid cell. This is the primary hybrid concentration. 5. To yield the individual source groups’ hybrid concentrations (e.g., nonroad construction equipment), multiply each receptor’s hybrid primary concentration by the ratio of the receptor’s source group concentration to its total AERMOD concentration. 6. For carbon tetrachloride, two CMAQ species were used, one for which boundary conditions and initial conditions were set to zero (CARBONTET_NOBC) and one with non-zero initial and boundary conditions (CARBONTET). Then calculate background concentrations for carbon tetrachloride for each CMAQ grid cell by subtracting CARBONTET_NOBC from CARBONTET. Set total primary concentrations equal to the CARBONTET_NOBC concentrations. 7. Calculate total hybrid concentrations at each receptor by adding the total primary hybrid from step 4 with the secondary, fire and biogenic concentrations from steps 1 and 2. For non-CMAQ HAPs in all areas, relevant background was added to the total AERMOD concentrations for methyl bromide (0.027 µg/m3) and methyl chloroform (0.220 µg/m3). Background for the two HAPs were based on the five NOAA GMD sites: Cape Kumukahi, Hawaii (KUM); Mauna Loa, Hawaii (MLO); Niwot Ridge, Colo. (NWR); Barrow, Alaska (BRW); and Alert, Canada (ALT) and the Trinidad Head Site (AGAGE). The hybrid program then output hybrid concentrations for each run group, including fires, biogenics, secondary and total concentrationfor each HAP. Separate files were created for gridded receptors, block receptors and monitors by CMAQ domain. The program output the AERMOD results for the non-CMAQ HAPs in a similar format. AirToxScreen 2018 Documentation 79 After the initial CMAQ and AERMOD runs for all domains, changes were made to the inventories based on state/local/tribal comments on preliminary review of the inventories. AERMOD was rerun with the updated inventories, but not CMAQ. To account for the changes, CMAQ results were scaled based on the ratio of the updated AERMOD grid averages to the original AERMOD grid averages and the hybrid concentration was calculated. Bounds were put on the ratios for CMAQ scaling. If the AERMOD ratio was between 0.1 and 10, then the CMAQ results were scaled based on the AERMOD ratios, otherwise, the AERMOD results for the receptor were used to represent the hybrid concentration. For the non- CMAQ HAPs, the AERMOD results were used as normal for the hybrid calculations. For AirToxScreen, the hybrid ambient concentrations at the block level were used to estimate exposures. These were then summarized at the tract level. We do this by excluding nonpopulated census blocks and weighted the remaining populated blocks’ hybrid concentrations by their population. This gives us a population-weighted average concentration, which we used to calculate tract-level exposures and risks. An operational model performance evaluation of the HAPs simulated for the 2018 AirToxScreen was conducted using the Ambient Monitoring Archive for HAPs for the year 2018 (https://www.epa.gov/amtic/amtic-air-toxics-data-ambient-monitoring-archive); more details found in the Supplemental Data folder on the AirToxScreen website). The model evaluation included both the hybrid air toxics and non-hybrid air toxics. The hybrid evaluation compared the HAPs for which there are valid ambient data (i.e., completeness criteria protocol) to compare against CMAQ, AERMOD and the hybrid model predictions. Likewise, the HAP non-hybrid evaluation used similar observational completeness criteria constraints to compare against HAPs estimated by adding AERMOD to remote ambient concentrations (where available) that are assumed to reflect background conditions. Note that when pairing observed to model data, there are spatial-scale differences between CMAQ, AERMOD and the hybrid model predictions. A CMAQ concentration represents a 12-km grid-cell volume- averaged value. The AERMOD model concentration represents a specific point within the modeled domain. The hybrid model concentration combines the AERMOD point concentration gradients with the 12-km CMAQ grid-cell volume average. The ambient observed measurements are made at specific spatial locations (latitude/longitude). Several annual graphical presentations and statistics of model performance were calculated and prepared. Graphical presentations include: 1. Box and whisker plots that show the distribution and the bias of the predicted and observed data, and 2. Regional maps that show the mean bias and error calculated at individual monitoring sites and 3. Box and whisker plots which show the model-to-monitor ratios. 3.5.3. Ambient Monitoring Data Preparation EPA has created annual average concentrations for year 2018 using data in the Ambient Monitoring Archive for HAPs. These data primarily come from AQS; however, they also come from special studies and various other networks that may not have been included in AQS. All annual averages are in units of micrograms per cubic meter (µg/m3) using local meteorological data where available or standard conditions otherwise. An annual average is created for each unique pollutant/monitoring site/sampling duration used the following procedures: AirToxScreen 2018 Documentation 80 1. There must be at least 3 valid quarters of monitoring data for the year. A quarter is considered valid if it contains at least 7 daily averages. 2. Hourly monitoring data are averaged to daily (and subhourly data are averaged to hourly) using the following criteria translating to the ceiling of 75% completeness: Sampling Duration Averaging To Minimum Count 5 MINUTES HOURLY 9 10 MINUTES HOURLY 5 15 MINUTES HOURLY 3 30 MINUTES HOURLY 2 150 MINUTES DAILY 8 90 MINUTES DAILY 12 1 HOUR DAILY 18 2 HOUR DAILY 9 3 HOURS DAILY 6 4 HOUR DAILY 5 5 HOUR DAILY 4 6 HOUR DAILY 3 8 HOUR DAILY 3 12 HOUR DAILY 2 24 HOURS DAILY 1 3. The median Regression on Order Statistic (ROS) is the annual average used, requiring at least 80% of daily averages to be above the method detection limit (MDL). 4. Some pollutants were summed to better reflect the AirToxScreen modeled pollutant. This occurred for PAH groups (summed individual PAHs belonging to the PAH groups), xylenes (summed m/p with o-xylene), and 1,3-dichloropropylene (summed cis and trans). The data used in the model evaluation are provided in the Supplemental Data folder. 3.5.4. Model Performance Statistics The Atmospheric Model Evaluation Tool (AMET) was used to conduct AirToxScreen HAP evaluation (Appel et al., 2011). There are various statistical metrics used by the science community for model performance evaluation. For a robust evaluation, the principal evaluation statistics used to evaluate model performance are based on the following metrics: two-bias metrics (mean bias and normalized mean bias); three-error metrics (mean error, normalized mean error and root mean square error) and correlation coefficient. AirToxScreen 2018 Documentation 81 Common Variables: M = predicted concentration O = observed concentration X = predicted or observed concentration σ = standard deviation I. Mean Bias, Mean Error and Root Mean Square Error (µg/m3) 1 n Mean Bias = ∑ ( M − O) n 1 1 n Mean Error = ∑ M − O n 1 n ∑ ( M − O) 2 Root Mean Square Error = 1 n Mean Bias (MB) quantifies the tendency of the model to overestimate or underestimate values, while Mean Error (ME) and Root Mean Square Error (RMSE) measure the magnitude of the difference between modeled and observe values regardless of whether the modeled values are higher or lower than observations. II. Normalized Mean Bias and Error (unitless) n ∑ ( M − O) Normalized Mean Bias = 1 n ∑ (O) 1 Normalized mean bias (NMB) is used as a normalization to facilitate a range of concentration magnitudes. This statistic averages the difference (modeled minus observed) over the sum of observed values. NMB is a useful model performance indicator because it avoids overinflating the observed range of values, especially at low concentrations. n ∑ M − O Normalized Mean Error = 1 n ∑ (O) 1 AirToxScreen 2018 Documentation 82 Normalized mean error (NME) is similar to NMB, where the performance statistic is used as a normalization of the mean error. NME calculates the absolute value of the difference (model minus observed) over the sum of observed values. III. Correlation Coefficient (unitless) 1 n   O − O  Correlation = ( ) ∑  M − M      *    n − 1 1   σo   σ   m Correlation coefficient (r) provides an indication of the strength of linear relationship and is signed positive or negative based on the slope of the linear regression. 3.5.5. Hybrid Evaluation An annual operational model performance evaluation for HAPs used in the hybrid model calculation was conducted to estimate the ability of the hybrid model as well as to compare to the predictions from the CMAQ and AERMOD modeling systems to replicate the 2018 HAP observed ambient concentrations. Inclusion of all three model results is intended to demonstrate the merged attributes of the hybrid model used for AirToxScreen. Statistical assessments of each model versus observed pairs were paired in time and space and aggregated on an annual basis. Results from the hybrid evaluation are presented in Appendix E. Table 3-9 provides a list of HAPs evaluated in the hybrid model performance evaluation and the number of observed monitoring sites (based on completeness criteria of observations, found in the model evaluations in the Supplemental Data folder on the AirToxScreen website). Figure 3-8 shows the 2018 HAP monitoring locations. Table 3-6. List of hybrid HAPs evaluated Model Air Toxic Measured Air Toxic No. of Sites Acetonitrile ACET_NITRILE_24_HOURS 41 Acrolein ACROLEIN_24_HOURS 51 Acrylonitrile ACRY_NITRILE_24_HOURS 5 Acetaldehyde ALD2_24_HOURS 97 Arsenic ARSENIC_PM10_24_HOURS 36 ARSENIC_PM25_24_HOURS 231 ARSENIC_TSP_24_HOURS 48 Benzene BENZENE_1_HOUR 29 BENZENE_24_HOURS 217 BENZENE_5_MINUTES 7 Beryllium BERYLLIUM_PM10_24_HOURS 28 BERYLLIUM_TSP_24_HOURS 35 Ethylene dibromide BR2_C2_12_24_HOURS 18 1,3-Butadiene BUTADIENE13_1_HOUR 26 BUTADIENE13_24_HOURS 121 Benzo-a-pyrene BaP_PM10_24_HOURS 25 AirToxScreen 2018 Documentation 83 Model Air Toxic Measured Air Toxic No. of Sites Cadmium CADMIUM_PM10_24_HOURS 36 CADMIUM_PM25_24_HOURS 136 CADMIUM_TSP_24_HOURS 49 Carbon tetrachloride CARBONTET_24_HOURS 201 CARBONTET_1_HOUR 3 Carbonyl sulfide CARBSULFIDE_5_MINUTES 7 Chloroform CHCL3_24_HOURS 167 Chloroprene CHLOROPRENE_24_HOURS 9 Vinyl chloride CL_ETHE_24_HOURS 35 Chlorine CL2_PM10_24_HOURS 2 CL2_PM2_5_24_HOURS 289 Ethylene dichloride CL2_C2_12_24_HOURS 135 Methylene chloride CL2_ME_24_HOURS 192 CL2_ME_5_MINUTES 7 Trichloroethylene CL3_ETHE_24_HOURS 80 1,1,2,2-Tetrachloroethane CL4_ETHANE_24_HOURS 19 Tetrachloroethylene CL4_ETHE_24_HOURS 137 CL4_ETHE_5_MINUTES 7 Chromium Compounds (only CR_VI_PM10_24_HOURS 9 hexavalent chromium was modeled) 1,4-Dichlorobenzene(p) DICL_BENZENE_24_HOURS 76 1,3-Dichloropropene DICL_PROPENE_24_HOURS 8 Ethyl benzene ETHYLBENZENE_1_HOUR 29 ETHYLBENZENE_24_HOURS 189 Formaldehyde FORM_24_HOURS 97 Hexane HEXANE_1_HOUR 28 HEXANE_24_HOURS 127 Lead Compounds LEAD_PM10_24_HOURS 41 LEAD_PM25_24_HOURS 289 LEAD_TSP_24_HOURS 159 Manganese Compounds MANGANESE_PM10_24_HOURS 39 MANGANESE_PM25_24_HOURS 289 MANGANESE_TSP_24_HOURS 70 Methyl chloride METHCHLORIDE_24_HOURS 174 METHCHLORIDE_5_MINUTES 7 Naphthalene NAPHTHALENE_24_HOURS 29 Nickel Compounds NICKEL_PM10_24_HOURS 36 NICKEL_PM25_24_HOURS 289 NICKEL_TSP_24_HOURS 62 Polycyclic Organic Matter PAH_000E0_24_HOURS 29 Polycyclic Organic Matter PAH_176E3_24_HOURS 20 Polycyclic Organic Matter PAH_176E4_24_HOURS 29 AirToxScreen 2018 Documentation 84 Model Air Toxic Measured Air Toxic No. of Sites Polycyclic Organic Matter PAH_176E5_24_HOURS 28 Polycyclic Organic Matter PAH_880E5_24_HOURS 29 Propylene dichloride PROPYL_DICL_24_HOURS 32 Styrene STYRENE_1_HOUR 29 STYRENE_24_HOURS 133 Toluene TOLUENE_1_HOUR 29 TOLUENE_24_HOURS 209 Xylene XYLENE_24_HOURS 187 XYLENE_1_HOUR 29 Figure 3-8. 2018 monitoring locations for the hybrid HAPs evaluation AirToxScreen 2018 Documentation 85 3.5.6. Non-hybrid Evaluation An annual operational model performance evaluation for HAPs used in the non-hybrid model calculation was conducted to estimate the ability of the AERMOD model to replicate the 2018 HAP observed ambient concentrations. Statistical assessments of modeled results versus observed pairs were paired in time and space and aggregated on an annual basis. Table 3-11 provides a list of HAPs evaluated in the non-hybrid model performance evaluation and the number of pairs (based on completeness criteria of observations, found in the model evaluations in the Supplemental Data folder on the AirToxScreen website) used in the annual median. Figure 3-30 shows the 2018 non-hybrid HAP monitoring locations. Results from the non-hybrid evaluation are presented in Appendix E. Table 3-10. List of non-hybrid HAPs evaluated Model Air Toxic Measured Air Toxic No. of Sites Antimony ANTIMONY_PM25_24_HOURS 144 ANTIMONY_PM10_24_HOURS 23 ANTIMONY_TSP_24_HOURS 22 Cobalt COBALT_PM25_24_HOURS 136 COBALT_PM10_24_HOURS 26 COBALT_TSP_24_HOURS 14 Selenium SELENIUM_PM25_24_HOURS 289 SELENIUM_PM10_24_HOURS 25 SELENIUM_TSP_24_HOURS 9 Methyl Bromide (Bromomethane) METHYLBROM_5_MINUTES 7 METHYLBROM_24_HOURS 102 Methyl Chloroform (1,1,1-Trichloroethane) MTHYLCHLRF_5_MINUTES 7 MTHYLCHLRF_24_HOURS 101 MTHYLCHLRF_1_HOUR 2 Carbon disulfide CARBNDISULF_24_HOURS 58 Methyl Isobutyl Ketone (4-Methyl-2- pentanone) MIBK_24_HOURS 52 Propanal (Propionaldehyde) PROPIONAL_24_HOURS 71 Cumene (Isopropylbenzene) CUMENE_24_HOURS 45 CUMENE_1_HOUR 29 CUMENE_3_HOURS 1 2,2,4-Trimethylpentane TRMEPN224_24_HOURS 79 TRMEPN224_1_HOUR 29 TRMEPN224_3_HOURS 1 Bromoform (Tribromomethane) BROMOFORM_24_HOURS 12 Chlorobenzene CHLROBZNE_24_HOURS 43 1,2,4-Trichlorobenzene TRICBZ124_24_HOURS 43 Benzyl Chloride (alpha-Chlorotoluene) BENZYLCHLO_24_HOURS 1 Hexachloro-1,3-butadiene HEXCHLRBT_24_HOURS 40 AirToxScreen 2018 Documentation 86 Model Air Toxic Measured Air Toxic No. of Sites Methyl tert-butyl ether MTBE_24_HOURS 6 p-Dioxane (1,4-Dioxane) P_DIOXANE_24_HOURS 10 Vinyl Acetate VINYLACET_24_HOURS 15 Ethyl Chloride (Chloroethane) ETHYLCHLRD_24_HOURS 43 Ethylidene Dichloride (1,1-Dichloroethane) ETHIDDICHLD_24_HOURS 18 Methyl Methacrylate MMETACRYLAT_24_HOURS 11 Vinylidene Chloride (1,1-Dichloroethylene) VINYLIDCLOR_24_HOURS 20 1,1,2-Trichloroethane TRICLA112_24_HOURS 24 Figure 3-9. 2018 monitoring locations for the non-hybrid HAPs evaluation AirToxScreen 2018 Documentation 87 4. Estimating Exposures for Populations Estimating inhalation exposure concentrations (ECs) is a critical step in determining potential health risks. Ambient concentrations do not consider that people move through locations and microenvironments where pollutant concentrations can differ. Different people have different daily activities, spend different amounts of time engaged in those activities, and engage in those activities in different locations. Most activities occur indoors (e.g., in homes, workplaces, schools and vehicles) where pollutant concentrations can differ from those outdoors. Therefore, the average concentration of a pollutant that people breathe can differ significantly from the ambient concentration at a fixed outdoor location. This section contains a discussion of how EPA estimated ECs for AirToxScreen. It begins with an overview of the surrogate approach used that included new exposure modeling for some AirToxScreen pollutants and applications of exposure-to-ambient concentration ratios for the remaining pollutants. This is followed by a more detailed description of this approach, a summary of the user inputs and other data required, and an overview of the quality-assurance measures included in estimating exposures. Further details on the exposure calculations for AirToxScreen can be found in Appendix E. 4.1. Estimating Exposure Concentrations For AirToxScreen, EPA used a combination of direct modeling and exposure factors to estimate inhalation ECs for AirToxScreen. This approach used census tract-level ambient concentrations estimated with air quality models, as described in Section 3, and yielded census tract-level exposure concentration estimates that we used to determine potential health risks for AirToxScreen. EPA used version 7 of the EPA Hazardous Air Pollutant Exposure Model (HAPEM7) to conduct direct exposure modeling for AirToxScreen. HAPEM, described in detail in Section 4.2, is a screening-level exposure model that estimates inhalation ECs corresponding to estimated ambient-pollutant concentrations. We used HAPEM7 for a selected group of surrogate pollutants and source categories. For each surrogate pollutant and AirToxScreen category (i.e., point, nonpoint, on-road mobile and nonroad mobile), EPA calculated the ratio of EC to ambient concentration (i.e., an exposure factor) for each census tract. Using each pollutant’s chemical properties, we then mapped each pollutant/category combination to the surrogate pollutants and source categories. Per census tract, we multiplied the ambient concentration of the pollutant by the surrogate’s exposure factor, resulting in estimated ECs. Section 4.4 further describes this exposure-factor approach. 4.2. About HAPEM Nearly two decades ago, EPA developed HAPEM for Mobile Sources (HAPEM-MS) to assess inhalation exposure to air toxics from highway mobile sources. This version of HAPEM used carbon monoxide as a tracer for highway mobile-source air toxic emissions. Today, HAPEM7 predicts inhalation ECs for a wide range of air toxics using either modeled ambient concentrations or measured data (without regard to source category), and the model no longer uses carbon monoxide as a tracer. Following other improvements, HAPEM version 4 and later (including HAPEM7) can estimate annual average human-exposure levels nationwide at a spatial resolution as fine as the census tract level (EPA 2002b, EPA 2005c, EPA 2007b, EPA 2015b). These changes make HAPEM7 suitable for regional and AirToxScreen 2018 Documentation 88 national inhalation risk assessments such as AirToxScreen. The 1996 and 1999 NATAs estimated Inhalation ECs using HAPEM4 and HAPEM5 respectively; the 2011 and 2014 NATAs both used HAPEM7 (EPA did not use HAPEM6 for NATA). Table 4-1 outlines some key differences between these three HAPEM versions. A complete history of the model can be found in the User’s Guide for HAPEM7 (EPA 2015b). Table 4-1. Key differences between recent versions of HAPEM Characteristic HAPEM4 HAPEM5 HAPEM7 Data source for population 1990 U.S. Census 2000 U.S. Census 2010 U.S. Census demographics Characterization of Point estimates Probability distributions Same as HAPEM5 microenvironmental factors Method for creation of Resampling of daily diaries Sampling a limited number Same as HAPEM5, except annual average activity for each of 365 days of daily diaries to now includes commuter- patterns from daily activity- without accounting for represent an individual’s status criterion pattern data autocorrelation range of activities, accounting for Interpretation of exposure- Uncertainty for the Variability of an nual ECs Same as HAPEM5, except concentration range for a average annual EC for the across cohort/tract now includes adjustments given cohort/tract cohort/tract combination members based on proximity to combination roadway HAPEM uses a general approach of tracking representative individuals of specified demographic groups as they move among indoor and outdoor microenvironments and between locations. As described in the following section, personal-activity and commuting data, specific to a hypothetical person’s demographic groups, are used to determine the census tracts containing residential and work locations and the microenvironments within each tract. Using stochastic sampling, the model estimates ECs by selecting empirically based factors reflecting the relationship between ECs within each microenvironment and the outdoor (ambient) air concentrations at that location. To estimate long-term ECs for a hypothetical person, the pollutant concentrations in each microenvironment visited are first combined into a daily-average concentration. The daily averages are then combined with proper weighting for season and day type to calculate a long-term average. Finally, the long-term averages are stratified by demographic group and census tract to create a distribution of ECs for each stratum. The median of each distribution represents the best estimate of exposure for a “typical” person of that demographic group in that census tract. In this case, “typical” does not refer to a specific individual in the population or even the average over a group of individuals. Rather, this is a hypothetical person living at the centroid of a census tract who engages in the usual activities (both indoor and outdoor) for someone in that demographic group and census tract. Additional technical information on HAPEM can be found in the User’s Guide for HAPEM7 (EPA 2015b). 4.3. HAPEM Inputs and Application HAPEM requires four main types of information to estimate ECs: (1) ambient concentrations of air toxics, (2) population data from the U.S. Census Bureau, (3) population-activity data and (4) microenvironmental data. The subsections below discuss these inputs, along with descriptions of the AirToxScreen 2018 Documentation 89 data used for AirToxScreen and related information on how EPA configured the model and applied it to conduct direct exposure modeling. 4.3.1. Data on Ambient Air Concentrations HAPEM typically uses annual average, diurnally distributed ambient air concentrations as input data. These concentrations can be monitoring data or concentrations estimated using a dispersion model or other air quality model. For AirToxScreen, EPA estimated annual average ambient concentrations for each census tract using a hybrid CMAQ-AERMOD approach discussed in Section 3.5. To ensure that exposure concentrations were completed in time for AirToxScreen’s release, the ambient data used in the EC calculations came from a preliminary assessment based on NEI version 1 emissions data; this concession does not significantly affect the ECs calculated. EPA stratified the air quality outputs for a selected group of pollutants by one or more of the four principal AirToxScreen categories (i.e., point, nonpoint, on-road mobile and nonroad mobile), using those results as surrogates for the remaining pollutants not modeled in CMAQ-AERMOD. Thus, exposure-model results generated for AirToxScreen can be summarized for each principal AirToxScreen category or any combination of those categories. 4.3.2. Population Demographic Data HAPEM divides the exposed population into cohorts such that each person in the population is assigned to one and only one cohort, and all the cohorts combined make up the entire population. A cohort is defined as a group of people whose exposure is expected to differ from exposures of other cohorts due to certain characteristics shared by the people within that cohort. For AirToxScreen, we specified cohorts by residential census tract and age, with the population in each census tract divided into six age groups: 0–1, 2–4, 5–15, 16–17, 18–64 and ≥65 years of age. These groups were developed using demographic data derived from the 2010 U.S. Census. EPA aggregated the predicted inhalation ECs across cohorts to estimate ECs for the general population. 4.3.3. Data on Population Activity HAPEM uses two types of data to define activities for the modeled population: activity-pattern data (specifying the frequency, location and duration of daily activities) and commuting-pattern data (specifying the work tracts for people living in each home tract). HAPEM uses these data together to place a hypothetical commuter in either the home or work tract and in a specific microenvironment at each 3-hour time step (the time step used for AirToxScreen). The microenvironment assignments and locations derived from these data are then used to calculate ECs (as explained in the next section). Data on human activity patterns are used to determine the frequency and duration of exposure within various microenvironments (such as indoors at home, in-vehicle and outdoors). Activity-pattern data are taken from demographic surveys of individuals’ daily activities that specify the sequence, duration and locations of those activities. The default source of activity-pattern data used by HAPEM and for AirToxScreen is EPA’s Consolidated Human Activity Database (CHAD; EPA 2015a). To develop the version of CHAD (version June 2014) used in AirToxScreen, data from 21 individual U.S. studies of human activities were combined into one complete data system containing over 45,000 person-days of activity- pattern records. Because of design limitations in the studies from which it is derived, CHAD may not well represent all demographic groups, particularly ethnic minorities and low-income populations. Also, the AirToxScreen 2018 Documentation 90 activity-pattern data in CHAD is mostly limited to one- or two-day periods. Extrapolation of these short- term records to the annual activity patterns needed for assessments of air toxics exposure introduces some uncertainty into the analysis. To address this extrapolation uncertainty, HAPEM uses a stochastic process to create simulated long- term (multi-day) activity patterns from daily activity-pattern data that account for day-to-day autocorrelation. These algorithms create annual average activity patterns from daily activity-pattern data to better represent the variability among individuals within a cohort-tract combination. For each day type and demographic group, daily-activity diaries were divided into three groups based on similarity using a cluster analysis. To simulate the activities of an individual, one diary was selected from each group for each day type, resulting in nine diaries in total. Then, for each day type, the sequence of the selected diaries was determined according to the probability of transition from one cluster group to another, as determined by analyses of the CHAD data. The simulation was repeated 30 times, resulting in a set of 30 estimates of annual ECs for each demographic group in each census tract. Use of a limited number of diaries and the transition probabilities is a way to account for day-to-day autocorrelation of activities for an individual, so each exposure-concentration estimate represents an estimate for an individual rather than an average for the group. Therefore, with this approach, the range represents the variability of ECs across the group. Commuting-pattern data, the second type of population activity data used in HAPEM, are derived for each cohort from a U.S. Census database containing information on tract-to-tract commuting patterns. These data specify the number of residents in each tract that work in that tract and every other census tract (i.e., the population associated with each home-tract/work-tract pair) and the distance between the centroids of the two tracts. An important limitation is that the commuting-pattern data included in HAPEM do not account for the movement of school-age children who travel (or commute) to a school located outside of their home tract. 4.3.4. Microenvironmental Data A microenvironment is a three-dimensional space in which human contact with an environmental pollutant occurs. In HAPEM, this space is treated as a well-characterized, relatively homogenous location with respect to pollutant concentrations for a specified period. The inhalation exposure estimate is determined by the sequence of microenvironments visited by the individual. The concentration in each microenvironment is estimated by using the three microenvironmental factors listed below to adjust the ambient-concentration estimate for the census tract where it is located:  a penetration factor that is an estimate of the ratio of the microenvironmental concentration to the concurrent outdoor concentration in the immediate vicinity of the microenvironment; penetration factors are pollutant-specific estimates that are derived from reported measurement studies;  a proximity factor that is an estimate of the ratio of the outdoor concentration in the immediate vicinity of the microenvironment to the outdoor concentration represented by the ambient air concentration input to the model; and  an additive factor that accounts for emission sources within or near a particular microenvironment, such as indoor emission sources. As noted below, the additive factor is not used for AirToxScreen. AirToxScreen 2018 Documentation 91 The relationship between the estimated ECs, the input ambient concentrations and these three factors is demonstrated by the equation below. C(i,k,t) = CONC(i,t) × PENk × PROXk + ADDk Where: C(i,k,t) = EC predicted within census tract i and microenvironment k for time step t, in units of μg/m3 CONC(i,t) = ambient concentration for census tract i for time step t, in units of μg/m3 PENk = penetration factor for microenvironment k PROXk = proximity factor for microenvironment k ADDk = additive factor accounting for sources within microenvironment k, in units of μg/m3 Stochastic processes can be used to select work tracts, ambient air concentrations and microenvironmental factors. This important feature allows exposures to be characterized with probability distributions rather than point estimates, which more accurately reflect the variability of these components and simulate some of the variability found in measurement studies. In HAPEM, the characteristics of each microenvironment are used to assign each microenvironment to one of three groups: indoors, outdoors and in-vehicle. AirToxScreen uses the 18 microenvironments shown in Table 4-2. The microenvironments in the indoor group were further classified as associated with either residence or other buildings, while those in the outdoor group were categorized as either near-road or away-from-road. Each group consists of microenvironments expected to have similar penetration factors, thus allowing microenvironmental factors developed for one microenvironment to be applied to other microenvironments in the same group. Within each census tract, HAPEM uses estimates of the number of people living within each of three distance-from-road bins to stochastically vary the proximity factor based on distance-from-road (i.e., proximity factors are higher for microenvironments near major roadways, lower for microenvironments relatively far from major roadways). The additive factor (ADDk) in the expression for EC, above, was set to zero for AirToxScreen because indoor-source data are currently incomplete (recall that AirToxScreen covers only pollutants derived from outdoor sources). An important consideration is that data to support quantitative microenvironmental factors are not well developed for many of the air toxic compounds and for most of the microenvironments, which introduces uncertainty into the analysis of exposures. Section 7 contains a discussion on uncertainty and variability regarding this and other issues for AirToxScreen. AirToxScreen 2018 Documentation 92 Table 4-2. Microenvironments used in HAPEM modeling for AirToxScreen Indoors Outdoors In Vehicle Residence Near-road Car/Truck Residential Motorcycle/Bicycle Public Transit Other Building Outdoors, Near Roadway Air Travel Outdoors, Parking Garage Bar/Restaurant Outdoors, Service Station Hospital Residential Garage Office Waiting Outdoors for Public Transit Public Access Away-from-road School Ferryboat Outdoors Waiting Inside for Public Transit Other 4.4. Exposure Factors HAPEM exposure modeling for AirToxScreen requires substantial time and resources for data collection and processing, computing and model processing. Due to these requirements, EPA conducted HAPEM modeling for AirToxScreen only for selected pollutants, which we present below along with how we used them to estimate ECs for the remaining AirToxScreen pollutants.  Coke oven emissions (emitted by point sources and present in ambient air as either particulates or gases) and diesel PM (modeled as particulates from nonpoint and mobile sources) were special cases that EPA modeled as themselves in HAPEM and not used as surrogates for any other pollutants not modeled in HAPEM.  Benzene and 1,3-butadiene are gas-phase pollutants emitted by many processes (and all four principal AirToxScreen categories) in nearly all U.S. locations. EPA selected benzene as the surrogate for all other gas-phase pollutants not modeled in HAPEM (EPA considers benzene modeling in AirToxScreen to be more reliable than 1,3-butadiene modeling).  Unspeciated, generic PAHs (“PAH, total”), which are pollutants that can be present in either gas phase or particulate phase in ambient air, are emitted by all four principal AirToxScreen categories and from a wide variety of processes. EPA selected “PAH, total” as the surrogate for all other mixed- phase pollutants not modeled by HAPEM.  Chromium (VI) is a highly toxic particulate-phase pollutant emitted by all four principal AirToxScreen categories. EPA selected it as the surrogate for all other particulate pollutants not modeled in HAPEM and emitted by point or nonpoint sources.  EPA selected nickel, a particulate-phase pollutant emitted by a variety of processes spread across the United States, as the surrogate for all other particulate pollutants not modeled in HAPEM and emitted by mobile sources. Appendix E to this document contains further details on the application of HAPEM7 in the AirToxScreen analysis. A spreadsheet file within the Supplemental Data folder accompanying this TSD spreadsheet contains the tract-level exposure-to-ambient concentration ratios (i.e., exposure factors) for each surrogate pollutant. (Access the AirToxScreen website at https://www.epa.gov/AirToxScreen.) AirToxScreen 2018 Documentation 93 Overall, the HAPEM exposure predictions are lower than the corresponding predicted air quality values. This reduction likely results from the inability of many pollutants to penetrate efficiently into an indoor environment. (Recall that indoor sources of air toxics are not included in AirToxScreen). 4.5. Evaluating Exposure Modeling A model-performance evaluation can provide valuable information regarding model uncertainty when using computer-simulation models of human exposures to pollutants. Also, a well-conducted evaluation can greatly increase confidence in model results for a given application or use. One type of performance evaluation uses measurements and environmental data as a benchmark to compare modeling estimates. EPA has worked with the Mickey Leland Center (NUATRC 2011) on past assessments to help identify new and independent sources of personal-monitoring data for use in comparison with the AirToxScreen results. Extensive peer review involving independent scientific and technical advice from scientists, engineers and economists can be another valuable component of a model evaluation. In July 2000, HAPEM4 underwent external peer review by technical experts for both the microenvironmental factors used in the model and the overall application of the model for use in air toxics screening assessments. A discussion of several of the issues addressed by these reviews is included in Appendix A of the report for the 1996 NATA presented to EPA’s Science Advisory Board for review (EPA 2001b). In 2001, EPA’s Scientific Advisory Board reviewed the application of HAPEM4 as part of the 1996 NATA review (EPA 2001a). Although several limitations were identified in the current methodology, HAPEM4 was acknowledged as an appropriate tool to help better understand the relationship of human exposures to ambient-concentration levels. Since then, numerous studies have used and cited subsequent HAPEM versions, including HAPEM7. 4.6. Summary  Estimating inhalation ECs is a critical step in determining potential health risks because ambient concentrations do not account for movements of individuals among locations and microenvironments where pollutant concentrations can differ.  We estimated inhalation ECs for each pollutant/source group/census tract for AirToxScreen using the HAPEM7 model.  These ECs can be used to determine census tract-level potential health risks. AirToxScreen 2018 Documentation 94 5. Characterizing Effects of Air Toxics Exposure to air toxics is associated with increased incidence of cancer and a variety of adverse noncancer health effects. The type and severity of effects depends on several factors, including the identity and nature of the chemical to which an individual is exposed, the magnitude and duration of exposure, and the unique behaviors and sensitivities of exposed individuals. EPA uses a toxicity assessment to identify and quantify the adverse health effects associated with The phrase “dose-response” is used generally exposure to a chemical, following EPA risk throughout this document to refer to the relationship between a level of a chemical and a assessment methods. As indicated in Figure 1-3 of physical response. The values EPA uses for this document and described in more detail in inhalation, however, are derived for exposure Volume 1 of EPA’s ATRA Reference Library (EPA concentration, although with consideration of dose. 2004a), two processes constitute toxicity Consideration of the relationship between assessment: hazard identification (during which the exposure concentration, dose and dosimetry (how specific adverse effects are identified that can be the body handles a chemical once it is inhaled) is causally linked with exposure to a given chemical) inherent in the derivation of values. The term and dose-response assessment (which “toxicity values” is used here to refer to the RfCs characterizes the quantitative relationship between and UREs used in inhalation risk assessment. chemical dose or concentration and adverse effects, that is, the hazard(s) identified in the first step). Ultimately, the results of the toxicity assessment, referred to in this document as “toxicity values,” are used along with exposure estimates to characterize health risks for exposed populations (as described in Section 6). Although the toxicity assessment is integral to the overall air toxics risk assessment, it is usually done prior to the risk assessment. EPA has completed this toxicity assessment for many air toxics and has made available the resulting toxicity information and dose-response values, which have undergone extensive peer review. This section explains how toxicity assessments are used in the AirToxScreen risk assessment process. Specifically, the sections that follow provide an overview of the cancer and noncancer toxicity values used in AirToxScreen and the primary sources of these values. They also describe several adjustments and assumptions to toxicity values specific to the AirToxScreen risk assessment process. 5.1. Toxicity Values and Their Use in AirToxScreen The toxicity values used in AirToxScreen are quantitative expressions used to estimate the likelihood of adverse health effects given an estimated level and duration of exposure. These toxicity values are based on the results of dose-response assessments, which estimate the relationship between the dose and the frequency or prevalence of a response in a population or the probability of a response in any individual. Because AirToxScreen is focused on long-term exposures, the toxicity values used in AirToxScreen are based on the results of chronic dose-response studies when such data are available. Chronic dose-response assessments can be used to help evaluate the specific 70-year-average (i.e., “lifetime”) ECs associated with cancer prevalence rates, or, for noncancer effects, the concentrations at which noncancer adverse health effects might occur given exposure over an extended time (possibly a lifetime, but the time frame also can be shorter). AirToxScreen 2018 Documentation 95 The toxicity values that are combined with ECs to conduct the risk characterization in AirToxScreen are based on the results of quantitative dose-response assessments. The actual values used, however, are not strictly considered dose-response or concentration-response values. To estimate cancer risks in AirToxScreen, EPA converted the results of cancer dose-response assessments for a given chemical to a URE that incorporates certain exposure assumptions. This value can be multiplied by the 70-year- average EC to obtain a lifetime cancer risk estimate for each individual. To evaluate the potential for noncancer adverse health effects, we used chronic dose-response data to estimate a threshold that is the EC in air at which adverse health effects are assumed to be unlikely (i.e., the RfC). These two types of values are described in more detail in the following section. The toxicity values used in AirToxScreen are consistent with those OAQPS has compiled for chronic inhalation exposures to air toxics. The full set of toxicity values used for AirToxScreen (and their sources) can be accessed via the AirToxScreen Supplemental Data file; see Appendix B for details. Sources of chronic dose-response assessments used for AirToxScreen were prioritized according to OAQPS risk assessment guidelines and level of peer review, as discussed below. 5.2. Types of Toxicity Values Each toxicity value used in AirToxScreen is best described as an estimate within a range of possible values appropriate for screening-level risk assessments. Note that the uncertainty in the dose-response assessments and toxicity values that AirToxScreen relies on is to some extent one-sided, providing a conservative (health-protective) estimate of risk. The “true” cancer risk and potential for adverse noncancer impacts are believed to be lower than those estimated in this assessment, although the possibility remains that they could be greater. Uncertainty in the derivation of the dose-response values and in other aspects of the AirToxScreen process is discussed in Section 7. 5.2.1. Cancer URE A cancer dose-response curve is used to demonstrate the quantitative relationship between dose and the The URE is the upper-bound excess lifetime cancer risk estimated to result from likelihood of contracting cancer. If the dose-response continuous exposure to an agent at a relationship is linear, the cancer response is assumed concentration of 1 microgram per cubic meter to increase proportionally with the dose (which might (µg/m3) in air. UREs are considered upper- be expressed as an EC, an absorbed internal dose, a bound estimates, meaning they represent a dose to a specific organ or tissue, or other measure). plausible upper limit to the true value. The We have proposed that linear extrapolation of true risk is likely to be less, but could be carcinogenic risk in the low-dose region of the curve greater. is a reasonable approach for estimating risk at relatively low exposures, such as those typically experienced by the general population for air toxics (i.e., the true value of the risk is unknown, and could be as low as zero). An upper-bound lifetime cancer risk represents a plausible upper limit to the true probability that an individual will contract cancer due to exposure over a 70-year lifetime to a given hazard (e.g., exposure to an air toxic). For an inhalation risk assessment (and for AirToxScreen), a URE can be used to calculate the estimated cancer risk from inhalation ECs. A URE is calculated by using dose-response information for a chemical and developing a factor in the appropriate units that can be combined directly with ECs in air to estimate individual cancer risks, given certain assumptions regarding the exposure conditions. AirToxScreen 2018 Documentation 96 Specifically, the URE represents the upper-bound of the excess cancer risk estimated to result from continuous exposure to a concentration of 1 µg of a substance per m3 of air, over a 70-year lifetime and assuming a daily inhalation rate of about 20 m3/day. The risk value is derived from the slope of the dose-response curve as estimated using a linearized multistage statistical model in the low-dose portion of the curve. The interpretation of the URE is as follows: If the URE is 3 × 10-6 µg/m3, no more than three excess tumors would develop per 1 million people if they were exposed daily for a lifetime to a concentration of 1 µg/m3. To the extent that true dose-response relationships for some air toxics compounds are not strictly linear, this assumption could result in overestimates of cancer risk. The upper bound is not a true statistical confidence limit because the URE reflects unquantifiable assumptions about effects at low doses. Thus, although the actual carcinogenic risk is likely to be lower than what is reflected in the URE, it also might be higher. The URE estimates the toxic potency of a chemical. EPA’s weight-of-evidence (WOE) descriptors provide EPA’s Weight of Evidence (WOE) estimates of the level of certainty regarding a Descriptors chemical’s carcinogenic potential. We evaluate three (EPA 2005a) broad categories of toxicological data to make a WOE • Carcinogenic to humans determination: (1) human data (primarily • Likely to be carcinogenic to humans epidemiological); (2) animal data (results of long-term • Suggestive evidence of carcinogenic experimental animal bioassays) and (3) supporting potential data, including a variety of short-term tests for • Inadequate information to assess genotoxicity and other relevant properties, carcinogenic potential pharmacokinetic and metabolic studies and structure- • Not likely to be carcinogenic to humans activity relationships. We evaluate these data in combination to characterize the extent to which they support the hypothesis that an agent or chemical causes cancer in humans. The approach outlined in EPA’s Guidelines for Carcinogen Risk Assessment (EPA 2005a) considers available scientific information regarding carcinogenicity. It provides a narrative approach to characterizing carcinogenicity rather than assigning chemicals to specific categories (as was done previously by EPA according to the 1986 guidelines). To provide some measure of clarity and consistency in an otherwise free-form, narrative characterization, standard descriptors are used as part of the hazard narrative to express the conclusion regarding the WOE for carcinogenic-hazard potential. The five recommended standard hazard descriptors are described below. Carcinogenic to Humans: This descriptor indicates strong evidence of human carcinogenicity. It is appropriate when the epidemiologic evidence of a causal association between human exposure and cancer is convincing. Alternatively, this descriptor might be equally appropriate with a lesser weight of epidemiologic evidence that is strengthened by other lines of evidence. It can be used when all the following conditions are met: (1) evidence of an association between human exposure and either cancer or the key precursor events of the agent’s mode of action is strong but insufficient for a causal association; (2) evidence of carcinogenicity in animals is extensive; (3) the mode(s) of carcinogenic action and associated key precursor events have been identified in animals and (4) evidence is strong that the key precursor events that precede the cancer response in animals are anticipated to occur in humans and progress to tumors, based on available biological information. Likely to Be Carcinogenic to Humans: This descriptor is appropriate when the WOE is adequate to demonstrate carcinogenic potential to humans but does not reach the WOE for the descriptor “Carcinogenic to Humans.” Adequate evidence consistent with this descriptor covers a broad spectrum. AirToxScreen 2018 Documentation 97 At one end of the spectrum is evidence for an association between human exposure to the agent and cancer and strong experimental evidence of carcinogenicity in animals. At the other end, with no human data, the weight of experimental evidence shows animal carcinogenicity by a mode or modes of action that are relevant or assumed to be relevant to humans. The use of the term “likely” as a WOE descriptor does not correspond to a quantifiable probability. Moreover, additional data, such as information on the mode of action, might change the choice of descriptor for the illustrated examples. Suggestive Evidence of Carcinogenic Potential: This descriptor is appropriate when the WOE is suggestive of carcinogenicity; that is, a concern for potential carcinogenic effects in humans is raised, but the data are judged insufficient for a stronger conclusion. This descriptor covers a spectrum of evidence associated with varying levels of concern for carcinogenicity, ranging from a positive cancer result in the only study on an agent to a single positive cancer result in an extensive database that includes negative studies in other species. Depending on the extent of the database, additional studies may or may not provide further insights. Inadequate Information to Assess Carcinogenic Potential: This descriptor is appropriate when available data are judged inadequate for applying one of the other descriptors. Additional studies generally would be expected to provide further insights. Not Likely to Be Carcinogenic to Humans: This descriptor is appropriate when the available data are considered robust for deciding no basis for human hazard concern exists. In some instances, positive results in experimental animals can occur when the evidence is strong and consistent that each mode of action in experimental animals does not operate in humans. In other cases, there can be convincing evidence in both humans and animals that the agent is not carcinogenic. A descriptor of “not likely” applies only to the circumstances supported by the data. For example, an agent might be “Not Likely to Be Carcinogenic” by one route but not necessarily by another. In those cases that have one or more positive animal experiments but the results are judged to be not relevant to humans, the narrative discusses why the results are not relevant. As with the “likely” descriptor, the term “not likely” here does not correspond to a quantifiable probability. Important to note is that these WOE categories express only a relative level of certainty that these substances might cause cancer in humans. The categories do not specifically connote relative levels of hazard or the degree of conservatism applied in developing a dose-response assessment. For example, a substance with suggestive evidence of carcinogenic potential might impart a greater cancer risk to more people than another substance that is carcinogenic to humans. The process of developing UREs includes several important sources of uncertainty. Many of the air toxics in AirToxScreen are classified as “likely” carcinogens. The term likely, as used in this instance, means that data are insufficient to prove these substances definitively cause cancer in humans. That some are not human carcinogens at environmentally relevant ECs is possible, and the true cancer risk associated with these air toxics might be zero. UREs for most air toxics were developed from animal data using health-protective methods to extrapolate to humans. Actual human responses may differ from those predicted. For more information, see EPA’s Guidelines for Carcinogen Risk Assessment (EPA 2005a). AirToxScreen 2018 Documentation 98 5.2.2. Noncancer Chronic RfC The RfC is an estimate of a continuous inhalation exposure that is thought to be without an appreciable The RfC is an estimate (with uncertainty risk of adverse health effects over a lifetime. The spanning perhaps an order of magnitude) of a continuous inhalation exposure to the human population considered when deriving RfCs includes population (including sensitive subgroups) sensitive subgroups (i.e., children, asthmatics and the that is likely to be without an appreciable risk elderly). The RfC is derived by reviewing a health- of deleterious effects during a lifetime. effects database for a chemical and identifying the most sensitive and relevant endpoint, along with the principal study or studies demonstrating that endpoint. The value is calculated by dividing the no- observed-adverse-effect level (or an analogous exposure level obtained with an alternate approach, e.g., a lowest-observed-adverse-effect level or a benchmark dose) by uncertainty factors reflecting the limitations of the data used. As with UREs for cancer risk assessment, the process of developing RfCs includes several important sources of uncertainty, which span perhaps an order of magnitude. Uncertainty factors are intended to account for (1) variation in sensitivity among the individuals in the population, (2) uncertainty in extrapolating laboratory animal data to humans, (3) uncertainty in extrapolating from data obtained in a study involving a less-than-lifetime exposure, (4) uncertainty in using lowest-observed-adverse-effect- level or other data rather than no-observed-adverse-effect-level data and (5) inability of any single study to address all possible adverse outcomes in humans adequately. Additionally, an adjustment factor is sometimes applied to account for scientific uncertainties in the data or study design not explicitly captured in the uncertainty factors (e.g., a statistically inadequate sample size or poor exposure characterization). For more information, refer to EPA’s Methods for Derivation of Inhalation Reference Concentrations and Application of Inhalation Dosimetry (EPA 1994). Unlike linear dose-response assessments for cancer, noncancer risks generally are not expressed as a probability that an individual will experience an adverse effect. Instead, in an air toxics risk assessment, the potential for noncancer effects in humans is typically quantified by calculating the ratio of the inhalation EC to the RfC. This ratio is referred to as the hazard quotient (HQ). For a given air toxic, exposures at or below the RfC (i.e., HQs are 1 or less) are not likely to be associated with adverse health effects. As exposures increase above the RfC (i.e., HQs are greater than 1), the potential for adverse effects also increases. The HQ, however, should not be interpreted as a probability of adverse effects. Additional information is provided in the description of risk characterization for AirToxScreen in Section 6 of this document. 5.3. Data Sources for Toxicity Values Information on dose-response assessments for evaluating chronic exposures for AirToxScreen was obtained from multiple sources and prioritized according to OAQPS risk assessment guidelines and level of peer review. Our approach for selecting appropriate toxicity values generally places greater weight on the EPA-derived toxicity values than those from other agencies (listed below). Additionally, the approach of favoring EPA values (when they exist) has been endorsed by EPA’s Science Advisory Board. This ensures the use of values most consistent with well-established and scientifically based EPA science policy. A spreadsheet file “AirToxScreen_Pollutants.xlsx” within the Supplemental AirToxScreen 2018 Documentation 99 Data folder accompanying this TSD contains the toxicity values for both cancer and noncancer chronic effects used in AirToxScreen. Cancer effects are characterized according to the extent to which available data support the hypothesis that a pollutant causes cancer in humans. 5.3.1. U.S. EPA Integrated Risk Information System We disseminate dose-response assessment information in several forms, depending on the level of internal review. The Integrated Risk Information System (IRIS) is an electronic database prepared and maintained by EPA that contains information on human-health effects that could result from exposure to various substances in the environment. These assessments have undergone external peer review and subsequent revision, compliant with requirements EPA instituted in 1996 for the IRIS review process. Externally peer-reviewed assessments under development for IRIS were given first consideration for AirToxScreen. These assessments, which reflect the most recent available toxicity information and data analysis, were used in some cases to supersede existing values on IRIS. Current IRIS values were used for AirToxScreen when peer-reviewed IRIS values under development were not available. 5.3.2. U.S. Department of Health and Human Services, Agency for Toxic Substances and Disease Registry The Agency for Toxic Substances and Disease Registry (ATSDR) publishes minimal risk levels (MRLs) for many substances based on health effects other than cancer. The MRL is defined as an estimate of human exposure to a substance that is likely to be without an appreciable risk of adverse effects (other than cancer) over a specified duration of exposure. For noncancer values in AirToxScreen, inhalation MRLs were used when IRIS RfC values were not available or when the ATSDR value was based on more recent, peer-reviewed data and analysis methods than the IRIS value (because the ATSDR concept, definition and derivation are analogous to IRIS). ATSDR does not develop assessments based on carcinogenicity. After internal and external review, MRLs are published in pollutant-specific toxicological-profile documents. ATSDR regularly updates these toxicological-profile documents; they are available at Toxic Substances Portal MRLs (ATSDR 2015). 5.3.3. California Environmental Protection Agency Office of Environmental Health Hazard Assessment California’s Office of Environmental Health Hazard Assessment (OEHHA) develops UREs based on carcinogenicity and reference exposure levels (RELs) based on health effects other than cancer. The REL is defined as a concentration level at or below which no adverse health effects are anticipated. For cancer and noncancer values in AirToxScreen, OEHHA UREs and inhalation RELs were used when their derivation was determined to be consistent with the concepts and definitions of IRIS or ATSDR. OEHHA dose-response information is available at Air Toxicology and Epidemiology (OEHHA 2016). Technical support documents for assessing hot spots are available on the OEHHA website at Hot Spots Guidelines (OEHHA 2015). 5.3.4. U.S. EPA Health Effects Assessment Summary Tables The Health Effects Assessment Summary Tables (EPA 1997) are a comprehensive listing consisting almost entirely of provisional UREs, RfCs and other risk assessment information of interest that various EPA offices have developed. The assessments, which have never been submitted for EPA consensus, AirToxScreen 2018 Documentation 100 were last updated in 2001. AirToxScreen uses information from these tables only when no values from the sources discussed in Sections 5.3.1 through 5.3.3 are available. 5.3.5. World Health Organization International Agency for Research on Cancer The International Agency for Research on Cancer of the World Health Organization (WHO) coordinates and conducts research on cancer and provides information on related cancer research and epidemiology. Although the agency does not develop quantitative dose-response values, it has published a series of monographs (WHO 2018) on the carcinogenicity of a wide range of substances. The following “degrees of evidence” published by the International Agency for Research on Cancer were used when EPA WOE determinations were not available for a substance or are out of date (see the AirToxScreen Glossary of Terms in Appendix A for definitions of each):  Group 1: Carcinogenic to humans;  Group 2A: Probably carcinogenic to humans;  Group 2B: Possibly carcinogenic to humans;  Group 3: Not classifiable as to human carcinogenicity; and  Group 4: Probably not carcinogenic to humans. 5.4. Additional Toxicity Decisions for Some Chemicals After the dose-response information was prioritized, we made additional changes to some of the chronic inhalation exposure values to address data gaps, increase accuracy and avoid underestimating risk for AirToxScreen. 5.4.1. Polycyclic Organic Matter A substantial proportion of polycyclic organic matter (POM) reported in the NEI was not speciated into individual compounds. For example, some emissions of POM were reported in NEI as “PAH, total” or “PAH/POM-Unspecified.” In other cases, individual POM compounds were reported for which no quantitative cancer dose-response value has been published in the sources used for AirToxScreen. As a result, we made simplifying assumptions that characterize emissions reported as POM. This allows us to quantitatively evaluate cancer risk for these species without substantially under- or overestimating risk (which can occur if all reported emissions of POM are assigned the same URE). To accomplish this, POM emissions as reported in NEI were grouped into categories. EPA assigns dose-response values based on the known or estimated toxicity for POM within each group. For AirToxScreen, we divided unspeciated POM emissions into ten POM groups. We discuss the POM groups used for the AirToxScreen in Section 2.1.1.1. We concluded that three PAHs – anthracene, phenanthrene and pyrene – are not carcinogenic, and therefore no URE was assigned to these for AirToxScreen. Details of the analysis that led to this conclusion can be found in the document entitled Development of a Relative Potency Factor (RPF) Approach for Polycyclic Aromatic Hydrocarbon (PAH) AirToxScreen 2018 Documentation 101 Mixtures: In Support of Summary Information of the Integrated Risk Information System (IRIS) (EPA 2010a). 5.4.2. Glycol Ethers Much of the emission-inventory information for the glycol ether category reported only the total mass for the entire group without distinguishing among individual glycol ether compounds. In other cases, emissions of individual glycol ether compounds that had not been assigned dose-response values were reported. Individual glycol ether compounds vary substantially in toxicity. To avoid underestimating the health hazard associated with glycol ethers, we protectively applied the RfC for ethylene glycol methyl ether (the most toxic glycol ether for which an assessment exists) to glycol ether emissions of unspecified composition. 5.4.3. Acrolein EPA first derived an IRIS RfC for acrolein in 2003 (EPA 2003a). This value was based on a 1978 subchronic rodent study that identified a lowest-observed-adverse-effect level (LOAEL) for nasal lesions (Feron et al. 1978). In 2008, the OEHHA derived a chronic reference exposure level for acrolein that was based on a more recent subchronic rodent study. The newer study identified a no-observed-adverse-effect level (NOAEL) for nasal lesions (OEHHA 2008; Dorman et al. 2008). Because both studies identified nasal lesions as the critical effect and because the Dorman et al. (2008) study identified a NOAEL, we used the more recent OEHHA REL for acrolein in AirToxScreen. 5.4.4. Metals Several decisions made for AirToxScreen regarding the toxicity values used for metal compounds are discussed in this section. Chromium (VI) compounds. We used the IRIS RfC for particulate chromium (VI) instead of the RfC for chromic acid mists and dissolved aerosols to avoid underestimating the health hazard associated with these compounds. The RfC for particulate chromium (VI) is less than those RfCs for chromic acid mists and dissolved aerosols. Lead. We consider the primary National Ambient Air Quality Standard (NAAQS) for lead, which incorporates an ample margin of safety, to be protective of all potential health effects for the most susceptible populations. The NAAQS, developed using the EPA Integrated Exposure, Uptake, Biokinetic Model, was preferred over the RfC for noncancer adverse effects because the NAAQS for lead was developed using more recent toxicity and dose-response information on the noncancer adverse impacts of lead. The NAAQS for lead was set to protect the health of the most susceptible children and other potentially at-risk populations against an array of adverse health effects, most notably including neurological effects, particularly neurobehavioral and neurocognitive effects (which are the effects to which children are most sensitive). The lead NAAQS, a rolling 3-month average level of lead in total suspended particles, was used as a long-term value in AirToxScreen. Nickel compounds. The cancer inhalation URE for most of the emissions of nickel compounds included in AirToxScreen (including unspecified nickel emissions reported as “nickel compounds”) was derived from the IRIS URE for insoluble nickel compounds in crystalline form. Soluble nickel species, and insoluble species in amorphous form, do not appear to produce genotoxic effects by the same toxic AirToxScreen 2018 Documentation 102 mode of action as insoluble crystalline nickel. Nickel speciation information for some of the largest nickel-emitting sources, including oil and coal combustion, suggests that at least 35 percent of total nickel emissions could be soluble compounds. The remaining insoluble nickel emissions, however, are not well characterized. Consistent with this limited information, we conservatively assumed for AirToxScreen that 65 percent of emitted nickel is insoluble and that all insoluble nickel is crystalline. Because the nickel URE listed in IRIS is based on nickel subsulfide and represents pure insoluble crystalline nickel, it was adjusted to reflect an assumption that 65 percent of the total mass of emitted nickel might be carcinogenic. In cases where a chemical-specific URE was identified for a reported nickel compound, it was used without adjustment. Furthermore, the MRL in Table 2 of the ATSDR is not adjusted because the noncancer effects of nickel are not thought to be limited to the insoluble crystalline form. Manganese. We used the ATSDR MRL for manganese (Mn) as the chronic inhalation reference value in AirToxScreen. There is an existing IRIS RfC for Mn (EPA 1993), and ATSDR published an assessment of Mn toxicity which includes a chronic inhalation reference value (i.e., an ATSDR Minimal Risk Level, MRL). (ATSDR 2012). Both the 1993 IRIS RfC and the 2012 ATSDR MRL were based on the same study (Roels et al. 1992); however, ATSDR used updated dose-response modeling methodology (benchmark dose approach) and considered recent pharmacokinetic findings to support their MRL derivation. Because of the updated methods, we determined that the ATSDR MRL is the appropriate noncancer health reference value for manganese in AirToxScreen. 5.4.5. Adjustment of Mutagen UREs to Account for Exposure During Childhood For carcinogenic chemicals acting via a mutagenic mode of action (i.e., chemicals that cause cancer by damaging genes), we recommend that estimated risks reflect the increased carcinogenicity of such chemicals during childhood. This approach is explained in detail in the Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens (EPA 2005b). Where available data do not support a chemical-specific evaluation of differences between adults and children, the Supplemental Guidance recommends adjusting for early-life exposures by increasing the carcinogenic potency by 10- fold for children up to 2 years old and by 3-fold for children 2 to 15 years old. These adjustments have the aggregate effects of increasing the estimated risk by about 60 percent (a 1.6-fold increase) for a lifetime of constant inhalation exposure. EPA recommends making these default adjustments only for carcinogens known to be mutagenic and for which data to evaluate adult and juvenile differences in toxicity are not available. For AirToxScreen, we adjusted the UREs for acrylamide, benzidine, chloroprene, coke oven emissions, ethyl carbamate, ethylene oxide, methylene chloride, nitrosodimethylamine and PAHs upward by a factor of 1.6 to account for the increased risk during childhood exposures. Although trichloroethylene is carcinogenic by a mutagenic mode of action, the age-dependent adjustment factor for the URE only applies to the portion of the slope factor reflecting risk of kidney cancer. For full lifetime exposure to a constant level of trichloroethylene exposure, we adjusted the URE upward by a factor of 1.12 (rather than 1.6 as discussed above). For more information on applying age-dependent adjustment factors in cases where exposure varies over the lifetime, see Toxicological Review of Trichloroethylene (EPA 2014). These nine air toxics were the only ones that met the adjustment criteria described above at the time of this assessment. The overall lifetime adjustment was applied because a single, lifetime-average EC was estimated for AirToxScreen rather than age-group-specific exposures. Note that the IRIS assessment AirToxScreen 2018 Documentation 103 contains vinyl chloride UREs for both exposure from birth and exposure during adulthood. However, the overall vinyl chloride URE already includes exposure from birth. Thus, we used that value with no additional factor applied. 5.4.6. Diesel Particulate Matter Diesel PM (DPM) mass (expressed as µg DPM/m3) has historically been used as a surrogate measure of exposure for whole diesel exhaust. Although uncertainty exists as to whether DPM is the most appropriate parameter to correlate with human health effects, it is considered a reasonable choice until more definitive information about the mechanisms of toxicity or mode(s) of action of diesel exhaust becomes available. In EPA’s 2002 Diesel Health Assessment Document (Diesel HAD), exposure to diesel exhaust was classified as likely to be carcinogenic to humans by inhalation from environmental exposures, in accordance with the revised draft 1996/1999 EPA cancer guidelines. Several other agencies (National Institute for Occupational Safety and Health, the International Agency for Research on Cancer, the World Health Organization, California EPA and the U.S. Department of Health and Human Services) had made similar hazard classifications prior to 2002. EPA also concluded in the 2002 Diesel HAD that it was impossible to calculate a cancer unit risk for diesel exhaust due to limitations in the exposure data for the occupational groups or the absence of a dose-response relationship. In the absence of a cancer unit risk, the Diesel HAD sought to provide additional insight into the significance of the diesel exhaust cancer hazard by estimating possible ranges of risk that might be present in the population. An exploratory analysis was used to characterize a possible risk range, and found that environmental risks from diesel exhaust exposure could plausibly range from a low of 10-5 to as high of 10-3 for long-term exposures. Because of uncertainties, the analysis acknowledged that the risks could be lower than 10-5, and a zero risk from diesel exhaust exposure was not ruled out. Noncancer health effects of acute and chronic exposure to diesel exhaust emissions are also of concern to EPA. The agency derived a diesel exhaust reference concentration (RfC) after considering four well- conducted chronic rat inhalation studies showing adverse pulmonary effects. The RfC is 5 µg/m3 for diesel exhaust measured as diesel particulate matter. This RfC does not consider allergenic effects such as those associated with asthma, immunologic effects or the potential for cardiac effects. Emerging evidence in 2002, discussed in the Diesel HAD, suggested that exposure to diesel exhaust could exacerbate these effects, but the exposure-response data were lacking at that time to derive an RfC. The Diesel HAD noted that the cancer and noncancer hazard conclusions applied to the general use of diesel engines then on the market, and as cleaner engines replace a substantial number of existing ones, the applicability of the conclusions would need to be reevaluated. Several studies published since 2002 continue to report increased lung cancer with occupational exposure to older engine diesel exhaust. Of note since 2011 are three new epidemiology studies that examined lung cancer in occupational populations (for example, in truck drivers, underground nonmetal miners and other diesel motor related occupations). These studies (Garshick et al. 2012; Silverman et al. 2012; Olsson et al. 2011) reported increasing risks with exposure to diesel exhaust, and positive exposure-response relationships were also evident to varying degrees. These newer studies (along with others that have appeared in the scientific literature) add to the evidence EPA evaluated in the 2002 Health Assessment Document and further reinforce the lung cancer hazard concern. The findings from AirToxScreen 2018 Documentation 104 these newer studies do not necessarily apply to newer technology diesel engines since the newer engines have large reductions in the emission constituents. In June 2012, the World Health Organization’s International Agency for Research on Cancer (IARC) evaluated the full range of cancer-related health-effects data for diesel engine exhaust (IARC 2014). IARC concluded that diesel exhaust should be regarded as “carcinogenic to humans.” This designation was an update from its 1988 evaluation that considered the evidence to be indicative of a “probable human carcinogen.” IARC is a recognized international authority on the carcinogenic potential of chemicals and other agents. Also in 2012, EPA and industry asked the Health Effects Institute to convene a panel to review recently published epidemiology studies of occupational exposures to diesel engine exhaust. The request was in part to determine whether new studies could be used in a quantitative risk assessment (QRA) to calculate a cancer URE. In a final report published at the end of 2015 (HEI 2015), the panel concluded that newer studies made considerable progress toward addressing a number of major limitations in previous epidemiologic studies of diesel engine exhaust. It further stated that, although uncertainties still remain, these newer studies provide a useful basis for potentially conducting a QRA of diesel engine exhaust exposures, but specifically to diesel engine exhaust from older diesel engines. Currently, there are no ongoing activities at EPA related to conducting a QRA for diesel engine exhaust. 5.5. Summary  To evaluate the potential of a given air toxic to cause cancer and other adverse health effects, we identified potential adverse effects that the substance causes and evaluated the specific ECs at which these effects might occur.  The URE represents the upper-bound excess cancer risk estimated to result from continuous exposure to a concentration of 1 µg of a substance per m3 of air over a 70-year lifetime.  The RfC is an estimate of a continuous inhalation EC over a 70-year lifetime that is thought to be without an appreciable risk of adverse effects. The population considered in the derivation of RfCs includes sensitive subgroups (i.e., children, asthmatics and the elderly).  Dose-response-assessment information for chronic exposure was obtained from multiple sources and prioritized according to conceptual consistency with OAQPS risk assessment guidelines and level of peer review.  After considering dose-response information, EPA adjusted some chronic-toxicity values to increase accuracy and to avoid underestimating risk. AirToxScreen 2018 Documentation 105 6. Characterizing Risks and Hazards in AirToxScreen Risk characterization, the final step in our risk assessment process for air toxics, combines the information from modeled exposure estimates with the dose-response assessment. The result is a quantitative estimate of potential cancer risk and noncancer hazard associated with real-world exposure to air toxics. The term “risk” implies a statistical probability of developing cancer over a lifetime. Noncancer “risks,” however, are not expressed as a statistical probability of developing a disease. Rather, noncancer “hazards” are expressed as a ratio of an exposure concentration (EC) to a reference concentration (RfC) associated with observable adverse health effects (i.e., a hazard quotient). This section contains information on the risk characterization conducted for AirToxScreen. After a brief overview of the risk-related questions that AirToxScreen is intended to address, the methods used to characterize cancer risk and noncancer hazards for AirToxScreen are described. A discussion of the quantitative results included in AirToxScreen follows this description. 6.1. The Risk-characterization Questions AirToxScreen Addresses AirToxScreen risk characterization considers both cancer risk and the potential for noncancer effects from inhalation of air toxics nationwide, in both urban and rural areas. The purpose of AirToxScreen is to understand cancer risks and noncancer hazards to help EPA and others identify air toxics and source categories of greatest potential concern and to set priorities for collecting additional data. The assessment represents a “snapshot” in time for characterizing risks from exposure to air pollutants; it is not designed to characterize risks sufficiently for regulatory action. The risk characterization for AirToxScreen, which was limited to inhalation risk from outdoor sources, was designed to answer the following questions:  Which air toxics pose the greatest potential risk of cancer or adverse noncancer effects across the entire United States?  Which air toxics pose the greatest potential risk of cancer or adverse noncancer effects in specific areas of the United States?  Which air toxics pose less, but still significant, potential risk of cancer or adverse noncancer effects across the entire United States?  When risks from inhalation exposures to all outdoor air toxics are considered in combination, how many people could experience a lifetime cancer risk greater than levels of concern (such as 1-in-1 million or 100-in-1 million)?  When potential adverse noncancer effects from long-term exposures to all outdoor air toxics are considered in combination for a given target organ or system, how many people could experience exposures that exceed the reference levels intended to protect against those effects (i.e., a hazard quotient greater than 1)? AirToxScreen 2018 Documentation 106 6.2. How Cancer Risk Is Estimated To estimate cancer risks in AirToxScreen, the results of cancer dose-response assessments for a given chemical were converted to a unit risk estimate (URE). That URE was then multiplied by the estimated inhalation exposure concentration to obtain an estimate of individual lifetime cancer risk. The approach used in AirToxScreen for characterizing cancer risk is consistent with EPA’s 2005 final Guidelines for Carcinogen Risk Assessment (EPA 2005a). When used with the cancer UREs described in Section 5, the approach is also consistent with EPA’s Supplemental Guidance for Assessing Susceptibility from Early- Life Exposure to Carcinogens (EPA 2005b). 6.2.1. Individual Pollutant Risk In AirToxScreen, individual lifetime cancer risk associated with exposure to a single air pollutant was estimated by multiplying an average estimated long-term exposure concentration by the corresponding URE for that pollutant. Thus, the equation below estimates the probability of an individual developing cancer over a lifetime from the exposure being analyzed due to a given inhalation exposure, over and above that due to any other factors. Risk = EC × URE Where: Risk = estimated incremental lifetime cancer risk for an individual due to exposure to a specific air toxic, unitless (expressed as a probability) EC = estimate of long-term inhalation exposure concentration for a specific air toxic, in units of μg/m3 URE = the corresponding inhalation unit risk estimate for that air toxic, in units of 1/(μg/m3) Note that UREs are typically upper-bound estimates, so actual risks may be lower than predicted. Also, the true value of the risk is unknown. 6.2.2. Multiple-pollutant Risks EPA estimates individual lifetime cancer risks resulting from exposure to multiple air toxics by summing the chronic cancer risk for each air toxic that can be quantified. This estimate of risk focuses on the additional lifetime risk of cancer predicted from the exposure being analyzed, over and above that due to any other factors. The following equation estimates the predicted cumulative individual cancer risk from inhalation of multiple substances: Risktot = Risk1 + Risk2 + …+ Riski Where: Risktot = total cumulative individual lifetime cancer risk, across i substances AirToxScreen 2018 Documentation 107 Riski = individual risk estimate for the ith substance In AirToxScreen, the estimated ECs are not considered upper bounds. Rather, they represent central- tendency estimates of ECs at the census tract level. Because cancer slope factors are 95-percent upper- confidence intervals (not “most probable estimates”), summing traditional risk levels can cause the resulting sum to overestimate a 95-percent upper-confidence-level risk for a mixture. In AirToxScreen, we assume that exposures to multiple carcinogens can be added together to estimate risks. This approach has drawbacks: Effects from multiple chemicals may be greater or less than additive, and statistical limitations exist. But this straightforward calculation is widely used to estimate cumulative risks, especially in screening assessments like AirToxScreen. Information on non-additive interactions is not readily available in a form that can be used for AirToxScreen. Without this specific information, cancer risk from various chemicals is conservatively assumed to be additive. Thus, the cancer risks from all air toxic compounds listed as carcinogenic or likely carcinogenic to humans were summed to determine cumulative cancer risks for AirToxScreen. More information on EPA’s methods for conducting risk assessment of mixtures can be found in the Framework for Cumulative Risk Assessment (EPA 2003b). 6.3. How Noncancer Hazard is Estimated To evaluate the potential for noncancer adverse health effects, EPA uses chronic dose-response data to EPA’s Chronic Noncancer estimate a reference concentration (RfC) – the EC at Guidelines which adverse health effects are assumed to be • Mutagenicity (EPA 1986) unlikely. (See Section 5.2.2 for more information on • Developmental Toxicity (EPA 1991) noncancer RfCs.) Due to the wide variety of endpoints, • Neurotoxicity (EPA 1998) hazard-identification procedures for noncancer effects • Reproductive Toxicity (EPA 1996) have not been described as completely in EPA guidance as those for carcinogens. But EPA has published guidelines for assessing several specific types of chronic noncancer effects (mutagenicity, developmental toxicity, neurotoxicity and reproductive toxicity). These can be found at Products and Publications Relating to Risk Assessment Produced by the Office of the Science Advisor (EPA 2016). EPA has also published a framework for using studies of these and other effects in inhalation risk assessment (EPA 1994). 6.3.1. Individual Pollutant Hazard EPA estimated chronic noncancer hazards for AirToxScreen by dividing a chemical’s estimated long-term EC by the RfC for that chemical to yield a hazard quotient (HQ). The following equation estimates the noncancer hazard due to a given inhalation exposure: EC HQ= RfC Where: HQ = the hazard quotient for an individual air toxic, unitless AirToxScreen 2018 Documentation 108 EC = estimate of long-term inhalation exposure concentration for a specific air toxic, in units of mg/m3 RfC = the corresponding reference concentration for that air toxic, in units of mg/m3 An HQ value less than or equal to 1 indicates that the exposure is not likely to result in adverse noncancer effects. An HQ value greater than 1, however, does not necessarily suggest a likelihood of adverse health effects and cannot be interpreted to mean that adverse health effects are statistically likely to occur. The statement is simply whether, and by how much, an EC exceeds the RfC, indicating that a potential exists for adverse health effects. 6.3.2. Multiple-pollutant Hazard We estimated chronic noncancer hazards for multiple air toxics by summing chronic noncancer HQs for individual air toxics that cause similar adverse health effects. The result is a hazard index (HI). Aggregation in this way produces a target-organ-specific HI, defined as a sum of HQs for individual air toxics that affect the same organ or organ system. More information on chemical mixtures risk assessment methods can be found in the EPA supplementary guidance for risk assessment of mixtures (EPA 2000). The following equation estimates the HI from inhalation of multiple substances: HI = HQ1 + HQ2 + … +HQi Where: HI = the hazard index for chronic exposure to air toxics 1 through i, unitless HQi = the hazard quotient for the ith air toxic, where all i air toxics are assumed to affect the same target organ or organ system, unitless As with the HQ, an HI value less than or equal to 1 indicates that the exposure is not likely to result in adverse noncancer effects. An HI value greater than 1, however, does not necessarily suggest a likelihood of adverse health effects and cannot be interpreted as a statistical probability of adverse effects occurring. This equation assumes an additive effect from simultaneous exposures to several chemicals. Summing HQs is inappropriate when effects from multiple chemicals are synergistic (greater than additive) or antagonistic (less than additive). As is the case with cancer risk, quantitative information on non-additive interactions resulting in noncancer hazards is not readily available; consequently, the noncancer HQs are assumed to be additive for chemicals with the same target organ or organ system. For AirToxScreen, we report HIs for 14 target organs or systems. Results indicate that respiratory hazards are the primary noncancer hazards for inhalation exposures to the modeled chemicals. AirToxScreen 2018 Documentation 109 6.4. How Risk Estimates and Hazard Quotients Are Calculated for AirToxScreen at Tract, County and State Levels The cancer risk and HQs for each modeled air toxic are estimated from ECs (not ambient concentrations) by combining them with UREs and inhalation RfCs (or their equivalents). As described previously, the modeling conducted for AirToxScreen resulted in ambient concentrations for each air toxic emitted by modeled sources, with the level of spatial resolution varying by source type and the corresponding modeling approach (see Section 3). 6.4.1. Model Results for Point Sources: Aggregation to Tract-level Results The AirToxScreen modeling step generates ambient concentrations at the block level. For risk and exposure calculations, we aggregated these results from the block level to the tract level by taking a population-weighted average of all block-level concentrations within a given tract, as follows: ∑ Concblockj×Popblock Conctracti= j ∑ Popblockj Where: Conctracti = ambient concentration for census tract i Concblockj = ambient concentration for census block j (contained within tract i), estimated by AERMOD Popblockj = population of blocks contained in tract i 6.4.2. Background Concentrations and Secondary Pollutants: Interpolation to Tract-level Results Background concentrations, as well as estimated concentrations of secondary pollutants generated by the CMAQ model, were estimated for levels other than census tract and thus required interpolation “down” to the tract level. Background concentrations were estimated at the county level. To obtain tract-level concentrations, the county-level estimate was assigned to all census tracts within that county. For secondary pollutants, concentrations were estimated using CMAQ. The results for each grid were then applied evenly to all tracts located within the grid. 6.4.3. Aggregation of Tract-level Results to Larger Spatial Units Tract-level ambient concentrations were aggregated up to the county, state and national levels using a method that weights concentration according to the population within a region. For a county, for example, a population-weighted ambient concentration was estimated by multiplying the tract-level concentrations by the population of each tract, summing these population-weighted concentrations, and dividing by the total county population encompassing all tracts to obtain a final population- weighted, county-level concentration. The process for aggregating from the tract to the county level can be expressed using the following equation: AirToxScreen 2018 Documentation 110 ∑ Conctract ×Pop Conc i county = tracti k Popcountyk Where: Conccountyk = population-weighted concentration for county k Conctracti = ambient concentration in tract i (contained within county k) Poptracti = population in tract i (contained within county k) Popcountyk = population in county k This same method was applied when aggregating up to the state or national level, using the appropriate concentration and population values. AirToxScreen results include ambient concentrations, ECs, cancer risks, and noncancer HQs at the tract, county, state and national levels. The ambient concentrations derived at the block level also were used to estimate ECs using either direct exposure modeling with HAPEM or with the exposure factors derived from HAPEM modeling (i.e., ratios of EC to estimated ambient concentration). (See Section 4 for a more thorough discussion of AirToxScreen exposure modeling and estimates.) Because the exposure factors were applied at the tract level, each census block was assigned the tract-level EC or exposure factor and then the census block- level ECs are estimated. As was done with the ambient-level concentrations, the block-level ECs were used to estimate cancer and noncancer effects and to aggregate these concentrations up to larger spatial scales. To aggregate tract-level concentrations up to the county-, state- or national-level concentrations, the tract-level concentrations were population weighted. 6.5. The Risk Characterization Results That AirToxScreen Reports AirToxScreen provides a snapshot of the outdoor air quality and the risks to human health that would result if air toxic emission levels remain unchanged. The assessment was based on an inventory of air toxics emissions from 2018. Individuals were assumed to spend their entire lifetimes exposed to these air toxics. Therefore, we did not account for the reductions in emissions that have occurred since the year of the assessment, or those that might happen in the future due to regulations for mobile and industrial sources. The evaluation of national-scale results and comparison of risks among chemicals make it possible to estimate which air toxics pose the greatest potential risk to human health in the United States. AirToxScreen reports a summary of these findings. Cancer risks are presented as lifetime risks, meaning the risk of developing cancer due to inhalation exposure to each air toxic compound over a normal lifetime of 70 years. Noncancer hazards are presented in terms of the ratio between the exposure and an RfC for inhalation exposures (i.e., the HQ). As described previously in this section, HQs are combined across chemicals where a common target organ or system is expected to estimate HI (i.e., for respiratory). Using these quantitative results, AirToxScreen classifies certain pollutants as drivers or contributors at the national or regional scale based on certain criteria. Table 6-1 contains the criteria for classifying the air toxics included in AirToxScreen at the regional and national level. In general, drivers and contributors were defined as air toxics showing a particular level of risk or hazard for some number of people exposed. AirToxScreen 2018 Documentation 111 For example, for a pollutant to be categorized in AirToxScreen as a cancer contributor at the national level, the individual lifetime cancer risk for that pollutant must have been shown by the assessment to be at least 1-in-1 million and the number of people exposed to that pollutant must have been shown to be at least 25 million. For a pollutant to be categorized in AirToxScreen as a regional driver of noncancer health effects, the chronic hazard index for that pollutant must have been shown to exceed 1 and the number of people exposed to that pollutant must have been shown to be at least 10,000. AirToxScreen results indicate that most individuals’ estimated risk is between 1-in-1 million and 100-in-1 million, although a small number of localized areas show risks of higher than 100-in-1 million. Although individuals and communities may be concerned about these results, recall that AirToxScreen is not designed to assess specific risk values at local levels. The results are best used as a tool to prioritize pollutants, emissions sources and locations of interest for further investigation. Table 6-1. Criteria establishing AirToxScreen drivers and contributors of health effects for risk characterization Criterion (Criteria in both columns must be met) Risk-characterization Category Individual Health Risk or Hazard Minimum Number of People Index Exceeds… Exposed (in millions) is… Cancer Risk (value in first column represents individual lifetime cancer risk, in 1 million)a National cancer driver 10 25 Regional cancer driver 10 1 (either set of criteria can be used) 100 0.01 National cancer contributor 1 25 Regional cancer contributor 1 1 Hazard Index (value in first column represents chronic hazard index for any organ/organ system)b National noncancer driver 1 25 Regional noncancer driver 1 0.01 a Cancer risks are upper-bound lifetime cancer risks; that is, a plausible upper limit to the true probability that an individual will contract cancer over a 70-year lifetime as a result of a given hazard (such as exposure to a toxic chemical). This risk can be measured or estimated in numerical terms (e.g., one chance in a hundred). b Hazard index is the sum of the HQs for substances that affect the same target organ or organ system. Because different pollutants can cause similar adverse health effects, combining HQs associated with different substances is often appropriate to understand the potential health risks associated with aggregate exposures to multiple pollutants. Furthermore, the risks estimated by the assessment do not consider indoor sources of air toxics or ingestion exposure to any pollutants. Also, although AirToxScreen estimates cancer and noncancer risks for numerous pollutants, additional chemicals might exist that are not identified or for which toxicity information is unavailable. Therefore, these risk estimates represent only a subset of the total potential cancer and noncancer risk associated with air toxics. Analytical results (including modeled ambient concentrations, exposure and risks) for each AirToxScreen are also provided at the census tract, county and state levels for those who wish to do their own technical analyses using the most refined output available. In performing such analyses, users must be extremely mindful of the purposes for which AirToxScreen was developed. AirToxScreen was developed as a screening tool to inform both national and more localized efforts to collect air toxics information and characterize emissions (e.g., prioritize pollutants or areas of interest for more refined data collection such as monitoring). The results are most meaningful when viewed at the state or national level. Nevertheless, reported spatial patterns within a county likely represent actual variations in overall AirToxScreen 2018 Documentation 112 average population risks. Less likely, however, is that the assessment pinpoints the exact locations where higher risks exist or that the assessment captures the highest risks in a county. Using these results alone to draw conclusions about local concentrations and risk is inappropriate. This assessment does not attempt to identify areas or populations that have significantly higher risks than others. Rather, it focuses on characterizing geographic patterns and ranges of risk across the country. In general, however, spending time in larger urban areas tends to pose greater risks than spending time in smaller urban and rural areas because the emissions of air toxics tend to be higher and more concentrated in areas with more people. This trend is not universal, however, and can vary from pollutant to pollutant according to its sources. The trend also can be affected by exposures and risk from non-inhalation and indoor sources of exposure. Based on the AirToxScreen results, millions of people live in areas where air toxics pose potential health concerns. Although air quality continues to improve, more needs to be done to meet the CAA’s requirements to reduce the potential exposure and risk from these chemicals. EPA will continue to develop air toxic regulations and cost-effective pollution prevention and other control options to address indoor and urban pollutant sources that significantly contribute to risk. You can access all AirToxScreen results via the following website: https://www.epa.gov/AirToxScreen. 6.6. Summary  The purpose of AirToxScreen is to understand cancer risk and noncancer health effects to help EPA and others identify pollutants and source categories of greatest potential concern and to set priorities for collecting additional information to improve future assessments.  Cancer risk is expressed as a statistical probability that an individual will develop cancer. Cancer risks were assumed to be additive across chemicals for AirToxScreen.  Noncancer hazard is expressed as an HQ, which is the ratio of the EC to an RfC associated with observable adverse effects. Noncancer hazards were assumed to be additive across chemicals that effect the same target organ or organ system.  AirToxScreen estimates most individuals’ cancer risk to be between 1-in-1 million and 100-in-1 million, although a small number of localized areas show risk higher than 100-in-1 million.  Air toxics data for AirToxScreen are presented at the national, state, county and census tract levels. The results are most meaningful when used to identify patterns of risk over larger areas. Using these results in the absence of additional information to draw conclusions about local concentrations and risk is inappropriate. AirToxScreen 2018 Documentation 113 7. Variability and Uncertainty Associated with AirToxScreen 7.1. Introduction This section discusses the variability and uncertainty associated with AirToxScreen. Clearly understanding Key Definitions for this Section these two concepts – found in all broad-scale Variability represents the diversity or assessments that rely on models and data – will heterogeneity in a population or parameter help you understand which questions AirToxScreen (e.g., variation in heights of people). Variability can answer appropriately, and which it cannot. cannot be reduced by taking more (or better) measurements; however, it can be accounted for by a more detailed modeling approach As stated in Section 1, AirToxScreen results should (e.g., modeling peoples’ heights in terms of not be used for limited-scale or site-focused age will reduce the unexplained variability due applications. AirToxScreen results are intended to to variation in heights). Uncertainty refers to the lack of knowledge characterize broad-scale risk to help identify those regarding the actual values of model input air toxics and source types associated with the variables (parameter uncertainty) and of highest exposures and posing the greatest potential physical systems (model uncertainty). health risks. The results are intended to identify Uncertainty can be reduced through improved measurements and improved model geographic patterns and ranges of risks across the formulation. country. To avoid over-interpretation and misapplication of the results, users must first understand the concepts of variability and uncertainty and then must recognize the role that these elements play in AirToxScreen results. Air toxic emissions, air concentrations and exposures are not the same throughout the United States, and the risks associated with air toxics are not the same for all people. Some areas have higher concentrations than others. At certain times, the concentration is higher at a given location than at other times. The risks for some individuals are below the national average, while for others the risks are above the national average. For these reasons, understanding how the ambient (outdoor) air concentration, exposure and risks from air toxics vary throughout the United States is essential for understanding AirToxScreen. This information comes from a process called variability analysis. EPA seeks to protect health with reasonable confidence based on the best data available. Estimates of air concentrations, exposures and risks, however, must always involve assumptions. Assumptions are necessary to simplify the problem at hand. They make AirToxScreen possible given available information and resources. That said, assumptions introduce uncertainties into AirToxScreen results because we can never be fully confidence that the assumptions are entirely correct. Understanding the extent of these uncertainties, the level of confidence that can be placed in statements related to the assessment, and how this confidence affects the ability to make reasoned decisions is essential. This information comes from a process called uncertainty analysis. 7.2. How AirToxScreen Addresses Variability The AirToxScreen process focuses on the variation in ambient air concentrations, exposures and risks across the United States, Puerto Rico and the U.S. Virgin Islands. Included, for example, are variations in source locations and the amounts of pollutants these sources emit, variations in meteorological conditions across the country, and variations in the daily activities of people. This section presents AirToxScreen 2018 Documentation 114 information on the key components that drive variability in risks associated with air toxics and the variability components that AirToxScreen addresses. A brief explanation is also provided on how AirToxScreen results should be interpreted given variability. 7.2.1. Components of Variability AirToxScreen results show how air concentrations, exposures and risks vary across broad areas of the country. They do not fully characterize how concentration, exposure and risk vary among individuals, except to the extent these individuals live in different geographic regions and are affected by the values typical of a census tract in that region. AirToxScreen results also do not fully characterize how ambient air concentrations might vary temporally and they do not show how concentrations vary spatially within a census tract. The following list explains some of the components of variability that determine differences in ambient air concentrations and individual risks. Temporal. Sources do not emit pollutants at constant rates. Similarly, the meteorological conditions that affect dispersion in the atmosphere vary over time. Thus, the ambient air concentration at a given location can vary over time. Geographic. The influence of pollutant emissions on a location’s ambient concentration depends on the degree of atmospheric dispersion of the emissions as they travel from the source to the location. Dispersion depends on both meteorological conditions, which vary from place to place, and the travel distance from the source. As a result, the ambient air concentration can vary greatly among different locations. The AirToxScreen analysis accounts for some geographic variation by using available meteorology data representative of the location and by modeling ambient concentrations for census areas, but the spatial resolution of model predictions is limited. Individual location. Two individuals might live at different locations within the same census tract. The ambient concentration estimated for the tract is only an approximation of conditions at all locations in the tract. Different locations within that tract may have different average ambient concentrations. Therefore, exposures and risks may also vary. Individual activity patterns. Two people may live at the same location but engage in different activities (called an “activity pattern”) during each day. Concentrations of substances indoors often differ from concentrations outdoors. If one person spends more time indoors than the other person, the average air concentration to which the two are exposed will differ, even though the ambient air concentration is the same. Similarly, one person may spend more time in a car and be exposed to an air concentration that is typical near roads. The net effect would be that the concentration of each pollutant in the air inhaled by these two people would differ. In other words, their exposure differs. In addition, the amount of outdoor pollution that penetrates into buildings and vehicles varies due to differences in ventilation and structural integrity. Thus, two people who live in the same location and spend the same amount of time indoors or in vehicles can still be exposed to different pollutant concentrations. Susceptibility. Two individuals may live at the same location and engage in the same activities, but one person may be more susceptible than another. Susceptibility refers to the extent to which an individual takes a pollutant into the body, transports it into an organ or tissue that might be adversely affected by it, or develops an adverse effect. AirToxScreen 2018 Documentation 115 A more susceptible person may develop a higher concentration of a pollutant in his or her organs or tissues, or have a higher chance of developing an adverse health effect, than another individual, even when exposures for both individuals are the same. For example, people breathe at different rates; two people breathing the same air may bring different amounts of a pollutant into their bodies. The amount of a pollutant reaching an organ or tissue can also vary between individuals, even if both bring the same amount into their lungs. The amount of time the pollutant remains in the body may also differ. Finally, the innate sensitivity to the effect may vary, even at equal doses in the tissues. The net effect of these factors is that either the dose of the pollutant delivered to the organs or tissues of the body or the level of response (or both) can differ substantially between two people even if they are exposed to the same pollutant concentrations. The extent to which each factor described above influences variation in individual risk can depend on the age, gender or ethnic group to which an individual belongs, as well as on that individual’s lifestyle. These groups comprise different receptor populations, or cohorts, and the exposures and risks can differ among them. 7.2.2. Quantifying Variability EPA conducts AirToxScreen to understand how ambient air concentrations, exposures and risks vary geographically – not among specific individuals. EPA calculates the ambient air concentrations for each specific, discrete location (i.e., census block centroid or census tract centroid; see discussion below) based on the emission sources and meteorological conditions affecting those specific tracts. Some temporal variation is accounted for in AirToxScreen calculations. For example, meteorological data used for air quality modeling is temporally dynamic. The air quality modeling therefore captures important variations in ambient conditions on an hourly basis before the resulting modeled ambient air concentrations are time averaged. The ambient concentration inputs to HAPEM are stratified into eight 3-hour time blocks; HAPEM then calculates ECs for each 3-hour time block before calculating an overall, long-term average EC. Although this approach to air quality and exposure modeling considers some important temporal variations, these time-stratified model outputs are averaged prior to the risk characterization step and are not included in the AirToxScreen results reported by EPA. The AirToxScreen concentrations and risks, however, do reflect a degree of geographic variation. The smallest geographic area for which AirToxScreen results are reported is the census tract. Although results are reported at the census tract level, average risk estimates are far more uncertain at this level of spatial resolution than at the county or state level. Census tracts are small, relatively permanent statistical subdivisions of a county, typically having between 1,200 and 8,000 residents, with most having close to 4,000. Census tracts do not cross county boundaries. Their areas vary widely depending on the density of settlement. Census tracts tend to be small in densely populated areas but can be very large in sparsely populated areas. Within census tracts are census blocks, the smallest geographic areas that the Census Bureau uses. Blocks are bounded by visible or virtual features such as streets, streams, and city or town boundaries. Census blocks are typically small in area and population; for example, in an urban area, a census block might correspond to a block bounded by city streets. In remote areas, however, census blocks might be large and irregular, comprising many square miles. Blocks typically have approximately 50 residents. AirToxScreen 2018 Documentation 116 Air concentrations are estimated in AirToxScreen at various levels of resolution depending upon the source type modeled. Secondary formation, fires and biogenics (modeled in the CONUS) are at 12-km grid cell resolution. Other sources use census block resolution, though the emissions for some sources are at the grid cell level – these grid cell-level emissions originate from even broader geographic scales (county and national level) and are less certain at these finer scales, as discussed below. For a given source type and modeling approach, variation in ambient air concentrations within a grid cell or census block is not explicitly modeled. For estimates at the block level, a representative ambient air concentration is estimated for a single location near the center of the block (i.e., the centroid, which is typically, but not always, the geographic center of the block chosen by the U.S. Census Bureau as a reference point). EPA then averages ambient concentrations estimated at the block level for the encompassing census tract, with concentration and risk results reported at the tract level. Assessment results do not reflect variations in the susceptibility of people within a census tract because the focus is to compare typical exposures and risks in different tracts. As a result, individual exposures or risks might differ by as much as a factor of 10 in either direction. You should consider exposures or risks determined in AirToxScreen as representative of the geographic area where an individual lives, but not as that individual’s personal risk. Thus, the results of the AirToxScreen analysis do not allow for a comparison of ambient air concentrations, exposures or risks between two individuals. They do, however, enable you to understand the variation in typical values for these quantities among counties or states and to a lesser degree among census tracts. For an individual, however, the values may differ from the typical value for the county or state if that individual lives in a part of the area that has a higher or lower than typical value, has an activity pattern that causes a higher or lower exposure than is typical, or is more (or less) susceptible than a “typical” person used in this assessment. For the purposes of estimating and reporting risk, EPA assumes that individuals within a census tract have the same exposure and risk. This assumption allows AirToxScreen users to examine variation in individual exposure among census tracts, but not the variation within a census tract. Activity patterns are included for each of six cohorts defined by age. Even within a receptor population, some variability in activity patterns among individuals is considered. Differences in susceptibility, however, are not included in AirToxScreen. EPA took this approach for AirToxScreen for two primary reasons:  An overall purpose of AirToxScreen is to examine broad differences driven by geography. AirToxScreen considers only geographic differences in pollutant concentration, exposure and risk. The goal is to understand how these three factors differ among people living in different geographic areas. EPA assesses these differences, as mentioned above, by tracking differences in air concentration in different census tracts, producing differences in the typical pollutant concentrations, exposures and risks in different tracts. Differences in susceptibility, however, can produce differences in risk between two individuals in the same census tract, and reporting on these differences is not a purpose of AirToxScreen.  The variability in susceptibility is difficult to model at the national scale. Very limited information is available on differences in susceptibility among individuals. Even if EPA were to choose to calculate and report differences among individuals in a census tract, scientifically reliable information necessary to produce these calculations is not available for many of the pollutants. Given current information, it may be possible to estimate variability in the rates at which people breathe, but this AirToxScreen 2018 Documentation 117 is only a small part of the overall variation in susceptibility. EPA therefore has chosen not to incorporate this source of variation between individuals. Considering these limitations, EPA elected to incorporate differences in emissions and meteorology (resulting in differences in ambient air concentration) and differences in location of typical individuals (resulting in differences in exposure) among census tracts. Variation in activity patterns for different age groups is reflected in the assessments to the degree that the age of residents varies by location. Variability in susceptibility is not included for the reasons given above. We address temporal variation in inputs by developing time-weighted averages of emissions characteristics, meteorological conditions and ECs. Temporal variation in the estimated ambient air concentrations, however, is not reflected in the results (only time-weighted annual averages are presented). 7.2.3. How Variability Affects Interpretation of AirToxScreen Results The AirToxScreen analysis illustrates how ambient air concentration, exposure and risk vary throughout the United States. The assessment does not focus on the variation in exposure and risk among individuals. It focuses on variation among well-defined geographic areas, such as counties or states, based on calculations of ambient air concentration, exposure and risk in various census tracts. To a lesser degree, variation among demographic groups is also addressed by AirToxScreen in that differences in activity patterns are considered in modeling ECs using HAPEM. Risk results, however, are not presented separately for individual demographic groups. The information contained in the maps, charts and tables produced in AirToxScreen display predictions of cancer risk and noncancer hazard. Cancer risk results include statements such as: “X percent of the census tracts in a given area are characterized by a typical lifetime excess cancer risk of less than R.” For this statement, if X is 25 percent and R is 1-in-1 million, the result would be: “25 percent of the census tracts are characterized by a typical risk of less than 1-in-1 million.” This statement does not necessarily mean that 25 percent of individuals in the specified area have a cancer risk of less than 1-in-1 million. Some people in these census tracts would be expected to have a risk above 1-in-1 million. Although a person may live in a census tract where the typical or average risk is less than 1-in-1 million, that person may live nearer the source than the average person in the census tract, may have an activity pattern that leads to greater exposure, or may be more susceptible. All these factors could cause that individual to experience a risk above the typical value for that census tract. Conversely, the individual could have a lower risk by living farther from the source, having an activity pattern that produces lower exposures, or being less susceptible. The important point to remember when interpreting the maps and charts of the AirToxScreen analysis is that they show variation among values of ambient air concentration, exposure or risk in census tracts or larger areas such as counties. This presentation allows the user to identify geographic regions (counties or states) where these values are higher or lower than the aggregated national average for all census tracts. It does not allow users to identify individuals who have higher or lower values of ambient air concentration, exposure or risk. Nevertheless, individuals with a high risk are more likely to be in geographic regions characterized by a high risk than in those geographic regions characterized by a low risk. The same can be said for exposure (i.e., individuals with a high exposure are more likely to be found AirToxScreen 2018 Documentation 118 in geographic regions characterized by high exposure than in those regions characterized by low exposure). 7.3. How AirToxScreen Addresses Uncertainty No scientific statement (in risk assessment or other areas of science) can be made with complete confidence. Risk estimates are always uncertain to some degree due to issues such as those discussed below. To maintain transparency and openness in the presentation of risk results, the party conducting a risk assessment must explain these uncertainties and how these uncertainties increase or decrease confidence. The AirToxScreen analysis produces statements about variability in ambient air concentrations, exposures and risks across geographic regions for typical individuals, as described in Section 7.2. In this section, the discussion of uncertainty is intended to address the confidence with which these statements regarding variability can be made. Note that uncertainty does not prevent EPA from making a statement of risk, nor does it prevent EPA from taking reasonable actions. Uncertainty does require, however, that the nature of the uncertainty, and the implications for decisions, be understood so the degree of support for the statement can be correctly and properly interpreted. 7.3.1. Components of Uncertainty Uncertainty arises from a variety of sources. To understand the sources of uncertainty affecting a risk assessment, it is instructive to consider the process by which a study such as AirToxScreen is performed, as described in the following sections. Problem formulation. We must first define the problem to be addressed. For example, a question that might help define the problem could include, “Is the occurrence of adverse human health effects correlated with emissions from industrial facilities?” What the study is intended to address and how the results will be used should be clear at the outset. This initial step in the analysis introduces problem- formulation uncertainty. The purpose of AirToxScreen is described in Section 1 of this document, where the question addressed in the assessment is defined as precisely as possible (e.g., that the study is limited to estimates of health effects in human populations) and the reader is informed of the limitations of the assessment. The issue of problem-formulation uncertainty is not considered further in this document. Defining the analysis components. This step describes what can influence the answer to the problem. In AirToxScreen, the multiple influences include emissions from a variety of sources (e.g., mobile, stationary, biogenic); atmospheric dispersion and chemistry; activity patterns for different cohorts; UREs and RfCs and other considerations. Where the science is poorly developed, the factors that must be included might not be clear. Resources may be limited, making the inclusion of all factors in the study infeasible. This step in the analysis, which results in the conceptual model for the assessment, introduces conceptual uncertainty. This issue is also addressed in the discussion of the limitations of AirToxScreen in Section 1, where the aspects of the problem that are (and are not) included in the study are addressed (e.g., that the study addresses inhalation of air toxics only). The issue of conceptual uncertainty is not considered further here. Selecting models. All risk assessments use models. The AirToxScreen analysis uses a series of mathematical models. EPA uses models in AirToxScreen to produce the emissions inventory, to calculate AirToxScreen 2018 Documentation 119 ambient air concentrations, to calculate exposures and to calculate risks (for cancer and noncancer effects). All scientific models involve uncertainties because a model reduces a (potentially very complex) set of chemical, biological, physical, social or other processes to manageable algorithms that can be used to perform calculations and make forecasts. The simplifications inherent in the development of a model introduce uncertainties. Typically, more than one model is available to apply to a problem, and those models can produce different results. Thus, uncertainty is introduced as to which model, and which model results, should be used. As a simple example, AirToxScreen uses a linear statistical model to relate EC and cancer risk: Cancer risk equals the exposure (air concentration) multiplied by a URE. Uncertainty analysis involves asking a series of questions: Are we certain this linear relationship is correct? Could the relationship be quadratic (i.e., risk equals exposure multiplied by the square of the dose)? Could the relationship have a threshold (i.e., no risk is apparent until the exposure becomes sufficiently large)? What are the implications for estimates of risk if these different models are used? What are the implications for decisions if a clear choice among the models cannot be made? This step in the analysis introduces model uncertainty. Judging model uncertainty can be both quantitative and qualitative. Qualitative issues involve the scientific plausibility of the model. Does the model include all important processes? Does it explain the phenomenon (e.g., atmospheric dispersion) well? Is the model well accepted in the scientific community – has it passed critical tests and been subject to rigorous peer review? Quantitative issues involve comparing model results against sets of data (although this also involves issues of parameter uncertainty discussed in the next bullet). Does the model generally predict these data accurately? Are the predictions accurate to within a factor of 2; a factor of 4? What is the effect of any approximation methods used in the model? Applying models. The models used in the AirToxScreen analysis require parameter inputs such as emission rates, stack heights, fractions of time spent indoors and UREs. Although models describe general relationships among properties of the real world (e.g., the linear relationship between exposure and cancer risk), parameters quantify these properties for specific cases (e.g., the numerical value of the URE for benzene). Parameters provide the numbers needed in the models. Various databases are available from which we can estimate these parameters, and the methods used to collect the data and to compile the databases introduce uncertainties. These factors all introduce parameter uncertainty. Although parameter uncertainty has both quantitative and qualitative aspects, common practice is to characterize this source of uncertainty quantitatively, with some qualitative caveats. For example, parameter uncertainty might be characterized by a confidence interval, which states that the true value of the parameter (such as the stack height for a facility) probably lies somewhere between 40 and 60 meters or that the stack height is “known to be within” a factor of 1.2, or that the stack height is “accurate to within” 20 percent. Attached to this quantitative characterization of uncertainty will be a qualitative caveat such as “the estimate of this uncertainty is based on measurements made in 1990 at facilities similar to the one considered in this study, but a change in the design of stacks might have been made since 1990.” This qualitative statement provides some idea of the confidence with which the quantitative assessment of uncertainty can be applied. AirToxScreen 2018 Documentation 120 7.3.2. Components of Uncertainty Included in AirToxScreen For this discussion, we have divided the uncertainties in AirToxScreen into three sources, based on the three steps AirToxScreen Components that Include leading from the estimate of emissions to the calculations Uncertainty of risk. Uncertainty in ambient air concentrations is due to • Ambient concentrations uncertainty in the emissions estimates and in the air • Exposure estimates quality models. Uncertainty in exposure is due to • Risk estimates uncertainty in activity patterns, locations of individuals within a census tract, and microenvironmental concentrations as reflected in the exposure model. Finally, uncertainty in risk is due to uncertainty in the shape of the relationship between exposure and effects, the URE and the RfC. These three sources of uncertainty are discussed below. Ambient air concentrations. Considering first the predictions of ambient air concentrations, the specific sources of uncertainty derive from the parameters for the following: emissions, stack data, particle sizes and reactivity, chemical speciation, terrain, boundary conditions, background concentration, meteorology and model equations. These sources of uncertainty are discussed briefly in this section. Emissions parameters, including emission rates and locations of sources, are taken from the NEI database. The NEI is a composite of estimates produced by state and local regulatory agencies, industry and EPA. Some of these data were further modified during the AirToxScreen review. We have not fully assessed the quality of specific emission rates and locations in the NEI and resultant AirToxScreen emissions (e.g., industrial emissions from a specific census tract), although we have conducted reviews. Some of the parameter values may be out of date, errors might have been introduced in transcribing raw data to a computer file, and other data-quality issues may be present. Emission estimates use a variety of methods such as emission factors, material balances, engineering judgment and source testing. Some release point locations use an average facility location instead of the location of each specific unit within the facility. Release point parameters may be defaulted for some situations. Fugitive release parameters are not required and are defaulted where missing. In addition, TRI data does not provide release-point parameters other than identifying sources as “stack” or “fugitive”; the release parameters used historical defaults from previous inventories or new defaults. Uncertainty also is inherent in the emission models used to develop inventory estimates. For example, we estimate county-level air toxic emissions from nonroad equipment by applying fractions of toxic total hydrocarbons to estimates of county-level hydrocarbons for gaseous air toxics and gas-phase PAHs and fractions of toxic particulate matter to estimates of county-level particulate matter for particle-phase PAHs. We use emission factors based on milligrams per mile for metals. The toxic fractions are derived from speciation data, based on limited testing of a few equipment types. The estimates of county-level total organic gases and particulates are derived from the EPA MOVES model. In MOVES, uncertainties are associated with emission factors, activity and spatial-allocation surrogates. National-level emissions for nonroad equipment in MOVES are allocated to the county level using surrogates, such as construction costs adjusted for geographic construction material cost (to allocate emissions of construction equipment) and employees in manufacturing (to allocate industrial equipment). Availability of more specific local data on equipment populations and usage will result in more accurate inventory estimates. For onroad sources, activity data either come from states or are allocated from states to counties using surrogates. Different surrogates are used for different emission types and geographic locations. These surrogates vary in quality from location to location. AirToxScreen 2018 Documentation 121 For mobile and nonpoint sources, we typically allocate emissions rates from the county level to grid cells through a surrogate such as traffic data, land cover or other land use types (such as truck stops for heavy duty diesel idling emissions). This allocation introduces additional uncertainty because the data on the surrogates also have uncertainty, and the correlations between the surrogates and the emissions are imperfect. The health effects of a pollutant depend on its chemical form when inhaled. For many sources, the emissions reported to the NEI or NEI database itself do not include information on chemical speciation of the pollutants of interest, but instead contains the total rate of pollutant emitted in all its forms. For example, chromium obtained from TRI or from some states is reported as chromium and are speciated into hexavalent and trivalent forms in an NEI augmentation step. We make assumptions about chemical speciation based on values estimated to be representative at such sources, considering information on source type, typical feedstock materials, knowledge of the process involved, or other relevant factors. Any one source, however, may have different values than the ones assumed. The dispersion, or movement, of pollutants in the atmosphere is influenced by the topography of the area surrounding a source, which is characterized by terrain parameters. Although CMAQ model estimates consider topography, the AERMOD model estimates as implemented for AirToxScreen do not in all cases. AERMOD estimates for point sources consider topography, but the estimates for the emissions sources modeled as census-tract area sources do not because considering topography in the model requires a single source elevation, which is not always possible for large grid cell sources. Not accounting for terrain introduces uncertainty into predictions of ambient air concentrations, particularly in areas with hills or mountains. Other sources of uncertainty in the modeling of ambient air concentrations are the values used for the boundary conditions in CMAQ and the background concentration estimates added to AERMOD concentrations for the non-CMAQ HAPs. These sources may include, for example, contributions from long-range transport of compounds from other counties and states. For more details on background concentrations, refer to the discussion in Section 3. The model equations used in the air quality models represent another source of uncertainty. The AERMOD dispersion model uses steady-state Gaussian equations, which make several assumptions that simplify plume dispersion. The CMAQ model is more complex in its treatment of pollutant dispersion and atmospheric dynamics than AERMOD; nevertheless, many assumptions underlie its Eulerian approach to dispersion, which are outlined further in the science documentation for the CMAQ model. While the hybrid approach of combining the CMAQ and AERMOD model output improves AirToxScreen’s treatment of chemistry and transport, there are uncertainties in the implementation. The approach requires emissions and meteorological inputs to be consistent between the models. While we treated emissions as consistently as possible, some simplifications were necessary. The main difference was in the temporal treatment of the emissions. The temporal allocation used in AERMOD was not exactly the same as in CMAQ for the county-level sources, though average profiles based on the CMAQ temporal approach were developed for use in AERMOD. There were also differences in the spatial treatment of CMVs, though for the large vessels, AERMOD source characterization parameters were developed based on summaries of the CMAQ vertical distribution of emissions. In addition to inconsistencies in model inputs, the hybrid approach uses an AERMOD grid cell average to normalize the individual AERMOD concentrations within a grid cell. The AERMOD values are less representative of the AirToxScreen 2018 Documentation 122 true AERMOD average in grid cells where there are fewer gridded receptors (i.e., nine gridded receptors were used in less-populated areas). To help characterize the aggregate uncertainty of the predictions of the air quality models, EPA compared modeled concentrations to available monitoring data on ambient air quality. For each monitor-pollutant combination, we compared the predicted annual average concentrations at the monitor location to the sampled annual average concentrations. We present these comparisons in the model evaluations found in the Supplemental Data folder on the AirToxScreen website. Measured concentrations were taken from EPA’s Ambient Monitoring Archive for HAPs, which includes National Air Toxics Trends Stations, state and local monitors reported to the Air Quality System, and other monitoring data collected from sources outside of AQS. For AirToxScreen, the exact locations of the monitors were used for the model-to-monitor comparison, an approach that increases accuracy over previous assessments. For more details about the model-to-monitor analyses for previous assessments, see Comparison of 1996 ASPEN Modeling System Results to Monitored Concentrations (EPA 2002c), Comparison of 1999 Model-Predicted Concentrations to Monitored Data (EPA 2006b), Comparison of 2002 Model-Predicted Concentrations to Monitored Data (EPA 2009) and Comparison of 2005 Model-Predicted Concentrations to Monitored Data (EPA 2010b). Discrepancies between model predictions and concentration measurements can be attributed to five sources of uncertainty:  emission characterization (e.g., specification of source location, emission rates and release characterization);  meteorological characterization (e.g., representativeness);  model formulation and methodology (e.g., characterization of dispersion, plume rise, deposition, chemical reactivity);  monitoring; and  boundary conditions/background concentrations. Underestimates for some pollutants could be a result of the following:  The NEI may be missing specific emission sources.  Emission rates may be underestimated or overestimated due to emission-estimation techniques and/or spatial allocation of national estimates to county, and county estimates to grid cells.  The accuracy of the monitor averages is uncertain; the monitors, in turn, have their own sources of uncertainty. Sampling and analytical uncertainty, measurement bias and temporal variation all can cause the ambient concentrations to be inaccurate or imprecise representations of the true atmospheric averages. Exposure. Sources of uncertainty in the relationship between ambient air concentrations and ECs include those associated with microenvironmental factors and activity patterns. HAPEM calculates the EC in various microenvironments (e.g., indoors at home, in a car) based on inputs of predicted ambient AirToxScreen 2018 Documentation 123 air concentrations and microenvironmental factors. The factors are characterized as probability distributions to reflect the variability found in air toxics measurements more fully. For many air toxics, the measurement studies needed to estimate microenvironmental factors are not available, so the values used are based on measurement studies of similar compounds in similar situations. This practice introduces uncertainty into the estimation of ECs for such compounds. In addition, even for air toxics with measurement studies, the estimated microenvironmental factors have some uncertainty because the number of such studies is limited. Furthermore, the uniform application of the microenvironmental factors to all census tracts introduces uncertainty by not accounting for possible geographic differences among tracts (e.g., different window-opening behavior, different levels of building integrity). The activity-pattern sequences for individuals used in HAPEM are based on the Consolidated Human Activity Database (CHAD). As explained in Section 4.3.3, HAPEM algorithms consider the variability in activity patterns among individuals within a cohort-tract combination. They do this largely by addressing correlation between subsequent activity patterns assumed to occur for each cohort-tract combination. The representativeness of the daily diaries in CHAD is uncertain because they are a compilation of many studies, including some older studies and some for which the data are based on non-random sampling. It is also uncertain how well the model algorithms represent actual daily autocorrelation between types of activity. This latter issue, however, pertains only to the variability of the ECs across the demographic group and not the median EC, which is the concentration reported by AirToxScreen. The commuting data used in HAPEM are based on an EPA analysis of information from a special study by the U.S. Census Bureau (2010). HAPEM uses this information, reflecting 2010 data, in coordination with the activity-pattern data to place an individual in either the home tract or work tract at each time step. These data introduce some uncertainty because they simplify commuting patterns to a pair of home and work census tracts and may not reflect certain details of some commutes (e.g., the additional census tracts encountered by commuters who travel to non-adjacent tracts or more complex commuting patterns that are not point to point). An additional important consideration is that the commuting- pattern data included in HAPEM do not account for the movement of school-age children who travel (or commute) to a school located outside the tracts in which they reside. Risk. Concerning the predictions of risk, the specific sources of uncertainty in dose-response relationships (in addition to those considered for ambient air concentration and exposure) are hazard identification, dose-response models for carcinogens, UREs and RfCs. One component of predicting risk is hazard identification. AirToxScreen’s cancer-risk estimates assume that a compound either is a carcinogen or produces a noncancer effect. We base this on the results of a hazard-identification stage that assesses the evidence that an air toxic produces either cancer or a noncancer effect. Because the evidence for either judgment is never unequivocal, a compound labeled as a carcinogen or one deemed to produce noncancer effects may produce neither effect in humans. This possibility introduces uncertainty into the calculation of risk – the risk could be zero. As the evidence for the original conclusion (i.e., that the compound produces the effect) increases, this uncertainty decreases. AirToxScreen’s cancer-risk estimates assume that the relationship between exposure and probability of cancer is linear. In other words, the probability of developing cancer is assumed proportional to the exposure (equal to the exposure multiplied by a URE). This type of dose-response model is used routinely in regulatory risk assessment because it is believed to be conservative; that is, if the model is incorrect, it is more likely to lead to an overestimate of risks than to an underestimate. Other AirToxScreen 2018 Documentation 124 scientifically valid, biologically based models are available. These produce estimates of cancer risk that differ from those obtained from the linear model. Uncertainty in risk estimates is therefore introduced by the inability to justify completely the use of one model or the other (because each model has some scientific support). An essential consideration is that this uncertainty is, to some extent, one-sided. In other words, conservatism when uncertainty exists allows more confidence in the conclusion that true risks are less than predicted than in the conclusion that risks are greater than predicted. URE parameters have associated uncertainty. In some cases, the UREs are based on maximum-likelihood estimates of the slope of the dose-response relationship derived from reliable data. In other cases, the UREs are based on “upper-bound” estimates (i.e., the slope is not the best estimate, but is a conservative value that is likely to lead to overestimates of risk) derived from less reliable data. For some compounds, the UREs are derived from human-exposure studies, but for others they are from animal exposures. These considerations introduce uncertainty into the URE values, and the amount of uncertainty varies among pollutants. Another source of uncertainty in estimating risk derives from the values chosen for the RfC parameters used to calculate an HQ for noncancer health risk. The RfC, which (like the URE) is based on limited information, is uncertain; as a result, the value of HQ is uncertain. As is the case for UREs, the uncertainty in the RfC is generally one-sided, and the risk is unlikely to be greater than predicted. 7.4. Summary of Limitations in AirToxScreen EPA developed this assessment to inform both national and more localized efforts to collect information and characterize or reduce air toxics emissions (e.g., to prioritize pollutants or areas of interest for monitoring and community assessments). As described above, uncertainty and variability characterize many of the elements in the assessment process for AirToxScreen, as in other assessments that derive results from environmental data and modeling of environmental data. Because of this, EPA suggests exercising caution when using the results of these assessments, as the overall quality and uncertainty of each assessment vary from location to location and from pollutant to pollutant. More localized assessments, using local-scale monitoring and modeling, are often needed to better characterize local- level risks. Recognizing the specific limitations in AirToxScreen results is critical to interpreting and using them properly, including that the results:  apply to geographic areas, not specific locations;  do not include comprehensive impacts from sources in Canada or Mexico;  are restricted to the year to which the assessment pertains (because the assessment uses emissions data only from that year);  do not reflect exposures and risk from all compounds;  do not reflect all pathways of exposure;  reflect only compounds released into the outdoor air; AirToxScreen 2018 Documentation 125  do not fully capture variations in background ambient air concentrations;  may underestimate or overestimate ambient air concentrations for some compounds due to spatial uncertainties;  are based on default, or simplifying, assumptions where data are missing or of poor quality; and  may not accurately capture sources with episodic emissions or other uncertainties. The results apply to geographic areas, not specific locations. The assessment focuses on variations in air toxics concentrations, exposures and risks among areas such as census tracts, counties and states. All questions asked, therefore, must focus on the variations among different areas. They cannot be used to identify “hot spots” where concentrations, exposures or risks may be significantly higher than other locations. Furthermore, this type of modeling assessment cannot address the kinds of questions an epidemiology study might, such as the relationship between cancer risks and proximity of residences to point sources, roadways and other sources of pollutant emissions. The results do not include comprehensive impacts from sources in Canada or Mexico. The AirToxScreen results for states that border these countries do not thoroughly reflect these potentially significant sources of transported emissions. The results apply to groups, not to specific individuals. Within a census tract, all individuals are assigned the same ambient concentration, chosen to represent a typical concentration. Similarly, the exposure assessment uses activity patterns that do not fully reflect variations among individuals. As a result, the exposures and risks in a census tract should be interpreted as typical values rather than as means, medians or some other statistical average. The values are likely to be in the midrange of values for all individuals in the census tract. The results for the current AirToxScreen are restricted to 2018 because the assessment used emissions data from that year. Also, the assumption regarding emissions in the assessment is that the levels remain constant throughout one’s lifetime (the emissions are not today’s levels, nor are they projected levels). Emissions continue to decrease, however, as (1) mobile-source regulations are phased in over time, (2) EPA-issued air toxics regulations for major industrial sources reach compliance due dates, (3) state and industry initiatives to reduce air pollutants continue and (4) some facilities are closed or have made process changes or other changes that have significantly reduced their emissions since 2018. The results do not reflect exposures and risk from all compounds. Only 138 of the 181 air toxics (180 CAA HAPs plus diesel PM) modeled in AirToxScreen have dose-response values. The remaining air toxics are not considered in the aggregate cancer risk or target-organ-specific hazard indexes. Of significance is that the assessment does not quantify cancer risk from diesel PM, although EPA has classified diesel PM as likely to be carcinogenic to humans by inhalation from environmental exposures. Currently, a URE for diesel PM has not yet been derived; therefore, a quantitative estimate of the cancer risks was included in AirToxScreen. An IRIS RfC for diesel PM allows AirToxScreen to include a quantitative estimate of its noncancer effects. The results do not reflect all pathways of exposure. The assessment includes only risks from direct inhalation of the emitted pollutants. It does not consider pollutants that might then deposit onto soil AirToxScreen 2018 Documentation 126 and into water and food, and therefore enter the body through ingestion or skin contact. Consideration of these routes of exposure could increase estimates of exposure and risk. The assessment results reflect only compounds released into the outdoor air. The assessment does not include exposure to pollutants produced indoors, such as from stoves or out-gassing from building materials, or evaporative benzene emissions from cars in attached garages. For some compounds such as formaldehyde, these indoor sources can contribute significantly to the total exposure for an individual, even if only inhalation exposures are considered. In addition, the assessment does not consider pollutants released directly to water and soil. It does include secondary formation, the transformation of one pollutant into another in the atmosphere. The assessment does not utilize CMAQ in all areas (i.e., not in Alaska, Hawaii, Puerto Rico and the U.S. Virgin Islands) and therefore does not estimate fires, biogenics and secondary formation based on location-specific data in these areas. It also does not use CMAQ for all pollutants and hence may not appropriately estimate long-range transport for these non-CMAQ pollutants. For pollutants not estimated in CMAQ, the assessment uses background ambient air concentrations based on remote concentration estimates, but these would not account for variations due to regional transport of these pollutants. AirToxScreen may underestimate or overestimate ambient air concentrations for some compounds in some locations due to spatial uncertainty in mobile and nonpoint emissions, which are more uncertain at finer geographic scales. The assessment uses default, or simplifying, assumptions where data are missing or of poor quality. Data for some variables used in the modeling for emissions and dispersion of pollutants (such as stack height and facility location) may be unavailable or flawed. In such instances, these values are replaced by default assumptions. For example, a stack height for a facility might be set equal to stack heights at comparable facilities or the location of the release points within a facility might be placed at the center of the facility. These substitutions introduce uncertainty into the final predictions of ambient concentrations, exposures and risks. AirToxScreen may not accurately capture sources with episodic emissions except for those with continuous emissions monitoring (CEMS) data, which use hourly emissions. AirToxScreen also does not include any short-term (a few days or weeks) deviations from a facility’s typical emissions pattern, such as during startups, shutdowns, malfunctions and upsets. AirToxScreen modeling uses temporal profiles for sources without CEMS that would not capture non-routine emissions spikes. AirToxScreen 2018 Documentation 127 8. References Akhtar, F., Henderson, B., Appel, W., Napelenok, S., Hutzell, B., Pye, H. and Foley, K. 2012. Multiyear Boundary Conditions for CMAQ 5.0 from GEOS-Chem with Secondary Organic Aerosol Extensions. 11th Annual Community Modeling and Analysis System Conference, Chapel Hill, NC, October 2012. Appel, K.W., Gilliam, R.C., Davis, N., Zubrow, A. and Howard, S.C. 2011. Overview of the Atmospheric Model Evaluation Tool (AMET) v1.1 for Evaluating Meteorological and Air Quality Models. Environ. Modell. Softw., 26(4): 434–443. Available online at http://www.cmascenter.org/. Appel, K.W., Napelenok, S., Hogrefe, C., Pouliot, G., Foley, K.M., Roselle, S.J., Pleim, J.E., Bash, J., Pye, H.O.T., Heath, N., Murphy, B., Mathur, R. 2018. Overview and evaluation of the Community Multiscale Air Quality Model (CMAQ) modeling system version 5.2. In Mensink C., Kallos G. (eds), Air Pollution Modeling and its Application XXV. ITM 2016. Springer Proceedings in Complexity. Springer, Cham. Available online at https://doi.org/10.1007/978-3-319-57645-9_11. ATSDR. 2012. (Agency for Toxic Substances & Disease Registry). Toxicological Profile for Manganese. Available online at http://www.atsdr.cdc.gov/toxprofiles/tp.asp?id=102&tid=23. ATSDR. 2015. Toxic Substances Portal Minimal Risk Levels (MRLs) for Hazardous Substances. Available online at http://www.atsdr.cdc.gov/mrls/index.asp. Last updated 28 October 2015. Last accessed 10 December 2015. Bey, I., Jacob, D.J., Yantosca, R.M., Logan, J.A., Field, B.D., Fiore, A.M., Li, Q., Liu, H.Y., Mickley, L.J., Schultz, M.G. 2001. Global modeling of tropospheric chemistry with assimilated meteorology: Model description and evaluation. Journal of Geophysical Research 106, 23073. https://doi.org/10.1029/2001JD000807. Caldwell, J.C., Woodruff, T.J., Morello-Frosch, R. and Axelrad, D.A. 1998. Application of Health Information to Hazardous Air Pollutants Modeled in EPA’s Cumulative Exposure Project. Toxicology and Industrial Health, 14(3): 429–454. Cimorelli, A.J., Perry, S.G., Venkatram, A., Weil, J.C., Paine, R.J., Wilson, R.B., Lee, R.F., Peters, W.D. and Brode, R.W. 2005. AERMOD: A Dispersion Model for Industrial Source Applications. Part I: General Model Formulation and Boundary Layer Characterization. Journal of Applied Meteorology, 44: 682–693. Cook, R., Phillips, S., Houyoux, M., Dolwick, P., Mason, R., Yanca, C., Zawacki, M., Davidson, K., Michaels, H., Harvey, C., Somers, J. and Luecken, D. 2011. Air Quality Impacts of Increased Use of Ethanol under the United States’ Energy Independence and Security Act. Atmospheric Environment, 45: 7714–7724. Dorman DC, Struve MF, Wong BA, Marshall MW, Gross EA and Willson GA, 2008. Respiratory tract responses in male rats following subchronic acrolein inhalation. Inhal Toxicol 20(3): 205-16. EPA (U.S. Environmental Protection Agency). 1986. Guidelines for Mutagenicity Risk Assessment. EPA/630/R-98/003. EPA, Washington, DC. Available online at AirToxScreen 2018 Documentation 128 http://www2.epa.gov/risk/guidelines- mutagenicity-risk assessment. Last accessed 10 December 2015. EPA. 1991. Guidelines for Developmental Toxicity Risk Assessment. EPA/600/R-91/001. EPA, Washington, DC. Available online at http://www2.epa.gov/risk/guidelines-developmental- toxicity-risk assessment. Last accessed 26 October 2015. EPA. 1993. Integrated Risk Information System Review of Manganese. Available online at https://cfpub.epa.gov/ncea/iris2/chemicalLanding.cfm?substance_nmbr=373. EPA. 1994. Methods for Derivation of Inhalation Reference Concentrations and Application of Inhalation Dosimetry. EPA/600/8-90/066F. EPA Office of Research and Development (ORD), Washington, DC. Available online at http://www2.epa.gov/risk/methods-derivation-inhalation-reference- concentrations- and-application-inhalation-dosimetry. Last accessed 27 October 2015. EPA. 1996. Guidelines for Reproductive Toxicity Risk Assessment. EPA/630/R-96/009. EPA, Washington, DC. Available online at http://www2.epa.gov/risk/guidelines-reproductive-toxicity-risk assessment. Last accessed 10 December 2015. EPA. 1997. Health Effects Assessment Summary Tables (HEAST). EPA NCEA, Washington, DC. Available online at http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=2877. EPA. 1998. Guidelines for Neurotoxicity Risk Assessment. EPA/630/R-97/0. EPA, Washington, DC. Available online at http://www2.epa.gov/risk/guidelines-neurotoxicity-risk assessment. Last accessed 10 December 2015. EPA. 2000. Supplementary Guidance for Conducting Health Risk Assessment of Chemical Mixtures. EPA/630/R-00/002. Risk Assessment Forum, Washington, DC. Available online at http://ofmpub.epa.gov/eims/eimscomm.getfile?p_download_id=4486. Last accessed 8 December 2015. EPA. 2001a. NATA – Evaluating the National-scale Air Toxics Assessment 1996 Data – An SAB Advisory. EPA/SAB/EC/ADV-02/001. Science Advisory Board, Washington, DC. Available online at http://archive.epa.gov/airtoxics/nata/web/pdf/sabrept1201.pdf. Last accessed 29 November 2015. EPA. 2001b. National-scale Air Toxics Assessment for 1996. Draft for EPA Science Advisory Board Review: January 18, 2001. EPA-453/R-01-003. EPA Office of Air Quality Planning and Standards (OAQPS), Research Triangle Park, NC. Available online at http://archive.epa.gov/airtoxics/nata/web/html/sabrev.html. Last accessed 29 November 2015. EPA. 2002a. Health Assessment Document for Diesel Engine Exhaust. EPA/600/8-90/057F. EPA ORD/ National Center for Environmental Assessment (NCEA), Washington, DC. Available online at http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=29060. Last accessed 4 December 2015. EPA. 2002b. The HAPEM User’s Guide Hazardous Air Pollutant Exposure Model, Version 4. EPA OAQPS, Research Triangle Park, NC. Available online at http://archive.epa.gov/airtoxics/nata/web/zip/hapem4guide4.zip. Last accessed 29 November 2015. AirToxScreen 2018 Documentation 129 EPA. 2002c. Comparison of ASPEN Modeling System Results to Monitored Concentrations. EPA. Available online at http://archive.epa.gov/airtoxics/nata/web/html/mtom_pre.html. Last updated 22 October 2015. Last accessed 30 November 2015. EPA. 2003a. Integrated Risk Information System Review of Acrolein. Available online at https://cfpub.epa.gov/ncea/iris2/chemicalLanding.cfm?substance_nmbr=364. EPA. 2003b. Framework for Cumulative Risk Assessment. EPA/630/P-02/001F. EPA ORD/NCEA, Washington, DC. Available online at http://www2.epa.gov/risk/framework-cumulative-risk assessment. Last accessed 26 October 2015. EPA. 2004a. Air Toxics Risk Assessment Reference Library. Volume 1: Technical Resource Manual. EPA- 453/K-04-001A. EPA OAQPS, Research Triangle Park, NC. Available online at http://www2.epa.gov/sites/production/files/2013-08/documents/volume_1_reflibrary.pdf. Last accessed 2 December 2015. EPA. 2004b. Air Toxics Risk Assessment Reference Library. Volume 2: Facility-Specific Assessment. EPA- 453/K-04-001B. EPA OAQPS, Research Triangle Park, NC. Available online http://www2.epa.gov/sites/production/files/2013-08/documents/volume_2_facilityassess.pdf. Last accessed 2 December 2015. EPA. 2005a. Guidelines for Carcinogen Risk Assessment. EPA/630/P-03/001F. EPA, Washington, DC. Available online at http://www2.epa.gov/risk/guidelines-carcinogen-risk assessment. Last accessed 26 October 2015. EPA. 2005b. Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens. EPA/630/R-03/003F. EPA, Washington, DC. Available online at http://www2.epa.gov/osa/memoranda-about-implementation-cancer-guidelines-and- accompanying- supplemental-guidance-science. Last accessed 4 December 2015. EPA. 2005c. The HAPEM User’s Guide Hazardous Air Pollutant Exposure Model, Version 5. EPA OAQPS, Research Triangle Park, NC. Available online at http://www2.epa.gov/sites/production/files/2013-08/documents/hapem5_guide.pdf. Last accessed 29 November 2015. EPA. 2006a. Air Toxics Risk Assessment Reference Library. Volume 3: Community-Scale Assessment. EPA- 453/K-06-001C. EPA OAQPS, Research Triangle Park, NC. Available online at http://www2.epa.gov/sites/production/files/2013- 08/documents/volume_3_communityassess.pdf. Last accessed 2 December 2015. EPA. 2006b. Comparison of 1999 Model-Predicted Concentrations to Monitored Data. EPA OAQPS. Available online at http://archive.epa.gov/airtoxics/nata1999/web/html/99compare.html. Last updated 13 September 2015. Last accessed 30 November 2015. EPA. 2007a. Control of Hazardous Air Pollutants from Mobile Sources Regulatory Impact Analysis. EPA420-R-07-002. EPA Office of Transportation and Air Quality (OTAQ) Assessment and Standards Division, Ann Arbor, MI. Available online at https://nepis.epa.gov/Exe/ZyPdf.cgi?Dockey=P1004LNN.PDF. Last accessed 16 July 2018. AirToxScreen 2018 Documentation 130 EPA. 2007b. The HAPEM User’s Guide Hazardous Air Pollutant Exposure Model, Version 6. EPA OAQPS, Research Triangle Park, NC. Available online at http://www2.epa.gov/sites/production/files/2013-08/documents/hapem6_guide.pdf. Last accessed 29 November 2015. EPA. 2009. Comparison of 2002 Model-Predicted Concentrations to Monitored Data. EPA OAQPS. Available online at http://archive.epa.gov/nata2002/web/html/compare.html. Last accessed 30 November 2015. EPA. 2010a. Development of a Relative Potency Factor (RPF) Approach for Polycyclic Aromatic Hydrocarbon (PAH) Mixtures: In Support of Summary Information of the Integrated Risk Information System (IRIS) (External Review Draft). EPA/635/R-08/012A. EPA, Washington, DC. Available online at http://cfpub.epa.gov/ncea/iris_drafts/recordisplay.cfm?deid=194584. Last accessed 19 October 2015. EPA. 2010b. Results of the 2005 NATA Model-to-Monitor Comparison, Final Report. Prepared by Eastern Research Group for EPA OAQPS. Available online at http://www3.epa.gov/ttn/atw/nata2005/05pdf/nata2005_model2monitor.pdf. Last accessed 30 November 2015. EPA. 2014. Toxicological Review of Trichloroethylene. EPA NCEA, Washington, DC. Available online at http://cfpub.epa.gov/ncea/iris/search/index.cfm?keyword=trichloroethylene. Last updated 12 September 2014. Last accessed 10 December 2015. EPA. 2015a. Consolidated Human Activity Database (CHAD). EPA, Washington, DC. Available online at http://www2.epa.gov/healthresearch/consolidated-human-activity-database-chad-use-human- exposure- and-health-studies-and. Last updated 30 September 2015. Last accessed 17 November 2015. EPA. 2015b. The HAPEM User’s Guide Hazardous Air Pollutant Exposure Model, Version 7. EPA OAQPS, Research Triangle Park, NC. Available online at http://www.epa.gov/fera/hazardous-air- pollutant-exposure-model-hapem-users-guides. Last accessed 29 November 2015. EPA. 2015c. What Are the Six Common Air Pollutants? EPA, Washington, DC. Available online at http://www3.epa.gov/airquality/urbanair/. Last updated 18 September 2015. Last accessed 2 December 2015. EPA. 2015d. Risk Assessment Guidance and Tools. EPA, Washington, DC. Available online at http://www.epa.gov/risk/guidance.htm. Last updated 1 December 2015. Last accessed 2 December 2015. EPA. 2015e. User’s Guide for the AMS/EPA Regulatory Model – AERMOD. EPA-454/B-03-001. Addendum June 2015. EPA, Research Triangle Park, NC. EPA. 2015f. Community Multiscale Air Quality (CMAQ). EPA, Washington, DC. Available online at http://www.epa.gov/air-research/community-multi-scale-air-quality-cmaq-modeling-system- air-quality- management. Last updated 8 December 2015. Last accessed 10 December 2015. AirToxScreen 2018 Documentation 131 EPA. 2015g. MOVES 2014. EPA OTAQ, Washington, DC. Available online at http://www3.epa.gov/otaq/models/moves/moves-docum.htm. Last updated 5 November 2015. Last accessed 10 December 2015. EPA. 2015h. Overview by Section of CAA. EPA OAR, Washington, DC. Available online at http://www.epa.gov/ttn/atw/overview.html. Last updated 10 September 2015. Last accessed 10 December 2015. EPA. 2016. Products and Publications Relating to Risk Assessment Produced by the Office of the Science Advisor (OSA). EPA OSA, Washington, DC. Available online at http://www.epa.gov/osa/products- and-publications-relating-risk-assessment-produced-office-science- advisor. Last updated 23 August 2016. Last accessed 16 August 2018. EPA. 2017. Revisions to the Guideline on Air Quality Models: Enhancements to the AERMOD Dispersion Modeling System and Incorporation of Approaches To Address Ozone and Fine Particulate Matter. 82 Federal Register 10 (17 January 2017), pp. 5182-5231. EPA. 2018. Initial List of Hazardous Air Pollutants with Modifications. EPA Office of Air and Radiation (OAR), Washington, DC. Available online at h https://www.epa.gov/haps/initial-list-hazardous- air-pollutants-modifications. Last updated 16 March 2017. Last accessed 18 June 2018. ERG. 2017. Improvement of Default Inputs for MOVES and SMOKE-MOVES: CRC Project A-100. February 28, 2017. Prepared for: The Coordinating Research Council. ERG No. 4020.00.001.001. EPA. 2022. Guidance on the Use of the Mesoscale Model Interface Program (MMIF) for AERMOD applications. EPA-454/B-22-011. June 2022. EPA, Research Triangle Park, NC. Eyth, Alison, J. Vukovitch, M. Strum, P. Dolwick, G. Pouliot, C. Allen, J. Beidler, B. Baek and Z. Adeleman. 2016. “Development of 2011 Hemispheric Emissions for CMAQ,” Presented at the 2016 CMAS Conference, October 24, 2016, University of North Carolina, Chapel Hill. Feron, V.J., Kryusse, A., Til H.P., et al. 1978. Repeated exposure to acrolein vapor: subacute studies in hamsters, rats and rabbits. Toxicology 9:47-57. Friedman, C.L., Y. Zhang, and N.E. Selin, 2014. Climate change and emissions impacts on atmospheric PAH transport to the Arctic, Environ. Sci. Technol., 48 (1), 429-437, doi:10.1021/es403098w. Garshick, E., Laden, F., Hard, J.E., Davis, M.E., Eisen, E.A. and Smith, T.J. 2012. Lung Cancer and Elemental Carbon Exposure in Trucking Industry Workers. Environmental Health Perspectives, 120:1301–1306. Available online at http://dx.doi.org/10.1289/ehp.1204989. Last accessed 10 December 2015. Gilliam, R.C., Pleim, J.E., 2010. Performance Assessment of New Land Surface and Planetary Boundary Layer Physics in the WRF-ARW. Journal of Applied Meteorology and Climatology 49, 760-774. Heath, Nicholas K., Pleim, J.E., Gilliam, R., Kang, D., 2016. A simple lightning assimilation technique for improving retrospective WRF simulations. Journal of Advances in Modeling Earth Systems. 8. 10.1002/2016MS000735. AirToxScreen 2018 Documentation 132 HEI (Health Effects Institute). 2015. Diesel Emissions and Lung Cancer: An Evaluation of Recent Epidemiological Evidence for Quantitative Risk Assessment. HEI, Boston, MA. Available online at http://pubs.healtheffects.org/view.php?id=446. Last updated 24 November 2015. Last accessed 10 December 2015. Henderson, B.H., Akhtar, F., Pye, H.O.T., Napelenok, S.L. and Hutzell, W.T. 2014. A Database and Tool for Boundary Conditions for Regional Air Quality Modeling: Description and Evaluation. Geosci. Model Dev., 7:339–360. Hoesly, R. M., Smith, S. J., Feng, L., Klimont, Z., Janssens-Maenhout, G., Pitkanen, T., Seibert, J. J., Vu, L., Andres, R. J., Bolt, R. M., Bond, T. C., Dawidowski, L., Kholod, N., Kurokawa, J.-I., Li, M., Liu, L., Lu, Z., Moura, M. C. P., O'Rourke, P. R., and Zhang, Q., 2018: Historical (17502014) anthropogenic emissions of reactive gases and aerosols from the Community Emissions Data System (CEDS), Geosci. Model Dev., 11, 369-408, doi: 10.5194/gmd-11-369-2018. IARC (International Agency for Research on Cancer). 2014. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, Volume 105 (2014). Available online at http://monographs.iarc.fr/ENG/Monographs/vol105/index.php. Last accessed 10 December 2015. Isakov, V., Irwin., J. and Ching, J.K. 2007. Using CMAQ for Exposure Modeling and Characterizing the Sub- grid Variability for Exposure Estimates. Journal of Applied Meteorology and Climatology, 46:1354– 1371. Janssens-Maenhout, G., Dentener, F., Van Aardenne, J., Monni, S., Pagliari, V., Orlandini, L., Klimont, Z., Kurokawa, J., Akimoto, H., Ohara, T., others, 2012. EDGAR-HTAP: a harmonized gridded air pollution emission dataset based on national inventories. European Commission Publications Office, Ispra (Italy). JRC68434, EUR report No EUR 25, 299–2012. Keller, C.A., M.S. Long, R.M. Yantosca, A.M. Da Silva, S. Pawson, and D.J. Jacob, 2014. HEMCO v1.0: A versatile, ESMF-compliant component for calculating emissions in atmospheric models, Geosci. Model Devel., 7, 1409-1417, doi:10.5194/gmd-7-1409-2014. Ma, L-M. and Tan, Z-M, 2009. Improving the behavior of the Cumulus Parameterization for Tropical Cyclone Prediction: Convection Trigger. Atmospheric Research 92 Issue 2, 190-211. http://www.sciencedirect.com/science/article/pii/S0169809508002585. National Centers for Environmental Prediction/National Weather Service/NOAA/U.S. Department of Commerce. 2015, updated daily. NCEP GFS 0.25 Degree Global Forecast Grids Historical Archive. Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory. https://doi.org/10.5065/D65D8PWK. NRC (National Research Council). 1983. Risk Assessment in the Federal Government: Managing the Process. Committee on the Institutional Means for Assessments of Risk to Public Health, Commission on Life Sciences. NRC. National Academy Press, Washington, DC. NRC. 1994. Science and Judgment in Risk Assessment. Committee on Risk Assessment of Hazardous Air Pollutants, Board on Environmental Sciences and Technology, Commission on Life Sciences. NRC. National Academy Press, Washington, DC. AirToxScreen 2018 Documentation 133 NUATRC (Mickey Leland National Urban Air Toxics Research Center). 2011. Available online at http://www.sph.uth.tmc.edu/mleland/. Last updated 30 December 2011. Last accessed 10 December 2015. OEHHA (Office of Environmental Health Hazard Assessment, California). 2008. California Environmental Protection Agency. Technical Support Document for the Derivation of Noncancer Reference Exposure Levels. Available online at https://oehha.ca.gov/media/downloads/crnr/appendixd1final.pdf. OEHHA. 2015. Hot Spots Guidelines. OEHHA, Sacramento, CA. Available online at http://www.oehha.ca.gov/air/hot_spots/index.html. Contents last updated 6 March 2015. OEHHA. 2016. Air Toxicology and Epidemiology. OEHHA, Sacramento, CA. Available online at http://www.oehha.ca.gov/air/allrels.html. Values last updated 28 June 2016. Otte T.L. and Pleim, J.E. 2010. The Meteorology-Chemistry Interface Processor (MCIP) for the CMAQ Modeling System: Updates through v3.4.1. Geoscientific Model Development, 3:243–256. Pye, H.O. and Pouliot, G.A. 2012. Modeling the Role of Alkanes, Polycyclic Aromatic Hydrocarbons, and Their Oligomers in Secondary Organic Aerosol Formation. Environ. Sci. Technol, 46, 11, 6041- 6047. Roels HA, Ghyselen P, Buchet JP, et al. 1992. Assessment of the permissible exposure level to manganese in workers exposed to manganese dioxide dust. Br J Ind Med 49:25-34. Selin, N.E., D.J. Jacob, R.J. Park, R.M. Yantosca, S. Strode, L. Jaeglé, and D. Jaffe, 2007. Chemical cycling and deposition of atmospheric mercury: Global constraints from observations, J. Geophys. Res., 112, D02308, doi:10.1029/2006JD007450. Silverman, D.T., Samanic, C.M., Lubin, J.H., Blair, A.E., Stewart, P.A., Vermeulen, R., Coble, J.B., Rothman, N., Schleiff, P.L., Travis, W.D., Ziegler, R.G., Wacholder, S. and Attfield, M.D. 2012. The Diesel Exhaust in Miners Study: A Nested Case-Control Study of Lung Cancer and Diesel Exhaust. Available online at http://dx.doi.org/10.1093/jnci/djs034. Last accessed 10 December 2015. Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M., Duda, M.G., Huang, X., Wang, W. and Powers, J.G. 2008. A Description of the Advanced Research WRF Version 3. Available online at http://www2.mmm.ucar.edu/wrf/users/pub-doc.html. Last updated 5 December 2014. Last accessed 16 December 2015. Stammer, D., F.J. Wentz, and C.L. Gentemann, 2003, Validation of Microwave Sea Surface Temperature Measurements for Climate Purposes, J. Climate, 16, 73-87. U.S. Census Bureau. 2010. Decennial Census of Population and Housing. Available online at https://www.census.gov/programs-surveys/decennial-census/decade.2010.html. Last accessed 20 August 2018. Wesson, K., Fann, N., Morris, M., Fox, T. and Hubbell, B. 2010. A Multi-pollutant, Risk-based Approach to Air Quality Management: Case Study for Detroit. Air Pollut. Res., 1:296–304. doi:10.5094/APR.2010.037. AirToxScreen 2018 Documentation 134 Yarwood, G., J. Jung, G. Whitten, G. Heo, J. Mellberg and M. Estes. 2010: Updates to the Carbon Bond Chemical Mechanismfor Version 6 (CB6). Presented at the 9th Annual CMAS Conference, Chapel Hill, NC. Available at https://www.cmascenter.org/conference/2010/abstracts/emery_updates_carbon_2010.pdf. AirToxScreen 2018 Documentation 135 This page intentionally left blank. Appendix A. Glossary "N"-in-1 million cancer risk: A risk level of “N”-in-1 million implies that up to “N” people out of one million equally exposed people may contract cancer if exposed continuously (24 hours per day) to the specific concentration over 70 years (an assumed lifetime). For example, a 1-in-1 million risk means that one person may develop cancer for every 1 million people exposed. A 10-in-1 million risk means 10 people may develop cancer per 1 million exposed, etc. These risks would be in addition to cancer cases that would normally occur in one million unexposed people. Note that AirToxScreen looks at lifetime cancer risks. This shouldn’t be confused with or compared to annual cancer risk estimates. To compare an annual cancer risk estimate with AirToxScreen results, multiply the annual estimate by 70 (or divide the lifetime risk by 70). Activity-pattern data: Data that depict actual human physical activity, the location of the activity and the time of day it takes place. The Hazardous Air Pollution Model (HAPEM) uses activity-pattern data from EPA's Comprehensive Human Activity Database (CHAD). Adverse health effect: A change in body chemistry, body function or cell structure that could lead to disease or health problems. Air Toxics Screening Assessment (AirToxScreen): EPA's ongoing thorough review of air toxics in the United States. AirToxScreen results help scientists focus on pollutants, emission sources and places that may need further study to better understand risks. AirToxScreen also spurs improvements in what we know about U.S. air toxics. This includes expanding air toxics monitoring, improving and updating emission inventories, improving air quality modeling, driving research on health effects and exposures to both ambient and indoor air, and improving assessment tools. AMS/EPA Regulatory Model (AERMOD): EPA’s preferred model to simulate near-field (i.e., within 50 km) dispersion of emissions. AERMOD models near- surface (boundary-layer) air turbulence in simple and complex terrain. This allows AERMOD to simulate how pollutants move and disperse in the air. It calculates pollutant concentrations from surface and elevated point, area, line and volume sources at many discrete points (receptors). Air toxics: Pollutants known to cause or suspected of causing cancer or other serious health effects. Air toxics are also known as toxic air pollutants or hazardous air pollutants.* Health concerns are linked to both short- and long-term exposures to these pollutants. Many air toxics cause respiratory, neurological, immune or reproductive effects, particularly for more susceptible or sensitive groups such as children. Five important air pollutants are not included in the list of air toxics because the Clean Air Act addresses them separately as “criteria pollutants.” These are particulate matter (PM), nitrogen oxides (NOx), sulfur oxides (SOx), ozone and carbon monoxide. Lead is both a criteria pollutant and an air toxic. Criteria pollutants are not addressed in AirToxScreen. *Diesel particulate matter is not a hazardous air pollutant but is included in the AirToxScreen air toxics. Ambient: Surrounding, as in the surrounding environment. In AirToxScreen, ambient air refers to the outdoor air surrounding a person through which pollutants can be carried. Therefore, the ambient concentrations estimated by AirToxScreen are concentrations estimated in the outdoor environment. AirToxScreen also estimates exposure concentrations that result when a person moves through various microenvironments, including the indoor environment. Ambient air monitoring: Process of collecting outdoor air samples to determine how much of an air pollutant is present at a location. Monitoring is used to: AirToxScreen 2018 Documentation A-1 • assess the extent of pollution; • provide air pollution data to the public in a timely manner; • help implement air quality goals or standards; • evaluate whether emissions control strategies are effective; • provide data on air quality trends; • help evaluate air quality models or modeling results; and • support research (for example, long-term studies of the health effects of air pollution). State, local and tribal agencies often monitor near places where screening tools (such as AirToxScreen) suggest the chance of high concentrations or risks. Area and other sources: Sources of air pollution that, by themselves, generally have lower emissions than “major sources” of air pollution (like factories). Area sources are often too small or too widespread to be inventoried as individual sources. They include facilities with air toxics emissions below the major source threshold as defined in the Clean Air Act (less than 10 tons of a single toxic air pollutant or less than 25 tons of multiple toxic air pollutants emitted in any one year). Area sources include smaller facilities, such as dry cleaners. As a separate definition, area sources in air quality modeling refer to those modeled in two dimensions (with length and width), as compared to point sources modeled at a single location. Assessment System for Population Exposure Nationwide (ASPEN): A computer model used to estimate toxic air pollutant concentrations. The ASPEN model includes: • rate of pollutant release; • location of and height from which the pollutants are released; • wind speeds and directions from the meteorological stations nearest to release; • breakdown of the pollutants in the atmosphere after release (i.e., reactive decay); • settling of pollutants out of the atmosphere (i.e., deposition); • transformation of one pollutant into another (i.e., secondary formation or decay). Atmospheric transformation (secondary formation): The process by which chemicals are transformed into other chemicals in the air (atmosphere). When a chemical is transformed, the original pollutant no longer exists; it is replaced by one or more new chemicals. Compared to the original chemical, the transformed chemical can have more, less or the same toxicity. Transformations and removal processes affect both the fate of the chemical and how long it stays in the air, called its persistence. Persistence is important because human exposure to a chemical depends on the length of time the chemical remains in the air. In AirToxScreen, we use both “atmospheric transformation” and “secondary formation”; they mean the same thing. Background concentrations: The amount of a pollutant that exists in the air that does not come from a specific source. These pollutants may come from a natural source or from distance sources. Background concentrations can explain pollutant concentrations found even without recent human-caused emissions. In AirToxScreen, we add background concentrations to AERMOD concentrations but not to CMAQ concentrations, which include background already. Most risk from AirToxScreen background concentrations is from carbon tetrachloride, a common pollutant that has few emission sources but is persistent due to its long half-life. Biogenic emissions: Biogenic emissions are emissions from natural sources, such as plants and trees. These sources emit formaldehyde, acetaldehyde and methanol; formaldehyde and acetaldehyde are key risk drivers in AirToxScreen. Biogenic sources also emit large amounts of other nonhazardous VOCs. We estimate biogenic emissions with a model that uses vegetation and land use data with temperature and solar radiation data. In addition to being a primary source of air toxics, compounds emitted by biogenic sources sometimes react with human-caused pollutants to form secondary pollutants. The AirToxScreen biogenics source group includes only the primary emissions. AirToxScreen 2018 Documentation A-2 Cancer Risk: The probability of contracting cancer over the course of a lifetime, assuming continuous exposure (assumed in AirToxScreen to be 70 years). Carcinogen: A chemical, physical or biological agent that can cause cancer. Carcinogenicity: Ability to produce cancer cells from normal cells. Chemical Abstracts Service (CAS) number: A unique number assigned to a chemical by the Chemical Abstracts Service, a service of the American Chemical Society. The purpose is to make database searches easier, as chemicals often have many names. Census tracts: Land areas defined by the U.S. Census Bureau. Tracts usually contain from 1,200 to 8,000 people, with most having close to 4,000 people. Census tracts are usually smaller than 2 square miles in cities, but are much larger in rural areas. Cohort: A group of people assumed to have identical exposures during a certain period. Using cohorts makes modeling exposures of a large population easier to manage. In AirToxScreen, we divide the entire population into a set of cohorts. Each person is assigned to one and only one cohort, and all the cohorts combined equal the entire population. Community Multiscale Air Quality (CMAQ) modeling system: An air quality model used in AirToxScreen. CMAQ estimates how pollutants move and disperse in the air. It includes the effect of atmospheric chemistry – how pollutants react in the air – a unique feature of the model. CMAQ’s structure allows it to calculate concentrations over a very large area, including many emission sources. Concentration: A way to describe how much of a pollutant is in the air. Concentration is usually shown as an amount, or mass, of pollutant per certain volume of air. In AirToxScreen, most concentrations are in micrograms (µg) of air pollutant per cubic meter (m3) of air (a “box” of air one meter on each side). Consolidated Human Activity Database (CHAD): An in-depth EPA database of human activity. CHAD includes data from over 20 activity studies dating to 1982. It also includes data from other assessments of human exposure, intake dose and risk. Diesel particulate matter: A mixture of particles that is part of diesel exhaust. EPA lists diesel exhaust as a mobile-source air toxic due to the cancer and noncancer health effects linked to exposure to whole diesel exhaust. Diesel PM (expressed as grams diesel PM/m3) has been used as a surrogate exposure measure for whole diesel exhaust. Dispersion model: A computerized set of equations that uses emissions and meteorological data to simulate how air pollutants behave and move in the air. A dispersion model estimates outdoor concentrations of individual air pollutants at chosen locations (called receptors). Emission Inventory System (EIS): An EPA information system for collecting emission inventory data and generating emission inventories. Emissions: Pollutants released into the air. AirToxScreen 2018 Documentation A-3 Emission inventory: A listing, by source, of the location and amount of air pollutants released into the air during some period (in AirToxScreen, a single year). Exposure assessment: An exposure assessment is part of an air toxics risk assessment such as AirToxScreen. The assessment determines (or estimates): • how a person may be exposed to chemicals (for example, by breathing); • how much of a chemical to which a person is likely to be exposed; • how long and/or how often they will be exposed; and • how many people are likely to be exposed. HAP: Hazardous air pollutant; another name for air toxics. Hazard index (HI): The sum of hazard quotients for toxics that affect the same target organ or organ system. Because different air toxics can cause similar adverse health effects, combining hazard quotients from different toxics is often appropriate. As with the hazard quotient, exposures below an HI of 1.0 likely will not result in adverse noncancer health effects over a lifetime of exposure. An HI equal to or greater than 1.0, however, doesn’t necessarily suggest a likelihood of adverse effects. Hazard quotient (HQ): The ratio of the potential exposure to a substance and the level at which no adverse effects are expected (calculated as the exposure divided by the appropriate chronic or acute value). A hazard quotient less than or equal to 1.0 indicates that adverse noncancer effects are not likely to occur, and thus can be considered to have negligible hazard. For HQs greater than 1.0, the potential for adverse effects increases, but we do not know by how much. Hazardous Air Pollutant Exposure Model (HAPEM): A computer model designed to estimate inhalation exposure for specified population groups and air toxics. The model uses census data, human-activity patterns, ambient air quality levels, and indoor/outdoor concentration relationships to estimate an expected range of inhalation exposure concentrations for groups of people. Human Exposure Model (HEM): A computer model used primarily for conducting inhalation risk assessments for sources emitting air toxics to ambient air. Inhalation exposure: Introducing air toxics (or other pollutants) into the body via breathing. Once inhaled, air toxics can be deposited in the lungs, taken into the blood, or both. Integrated Risk Information System (IRIS): An EPA program that identifies and characterizes the health hazards of chemicals found in the environment. IRIS is EPA’s preferred source of toxicity information. Lifetime cancer risk: The probability of contracting cancer over the course of a lifetime (assumed to be 70 years for the purposes of AirToxScreen). Major sources: Defined by the Clean Air Act as those stationary facilities that emit or have the potential to emit 10 tons of any one toxic air pollutant or 25 tons of more than one toxic air pollutant per year. Median: The middle value of a set of ordered values (i.e., half the numbers are less than or equal to the median value). A median is the 50th percentile of the data. AirToxScreen 2018 Documentation A-4 Metropolitan statistical area (MSA): A region with a relatively high population density at its core and close economic ties throughout the area. As defined by the U.S. Census Bureau, an MSA must have at least one urban area of 50,000 or more inhabitants. Microenvironment: A small space in which human contact with a pollutant takes place. AirToxScreen models cohort activities in indoor, outdoor and in-vehicle microenvironments: 1. Indoor locations: • Residence • Office • Store • School • Restaurant • Church • Manufacturing facility • Auditorium • Healthcare facility • Service station • Other public building • Garage 2. Outdoor locations: • Parking lot/garage • Near road • Motorcycle • Service station • Construction site • Residential grounds • School • Sports arena • Park/golf course 3. In-vehicle locations: • Car • Bus • Truck • Train/subway • Airplane • Other Microgram: One-millionth of a gram. One gram is about one twenty-eighth of an ounce, or about the weight of a raisin or paper clip. Mobile source: Air pollution sources that can move from place to place, like cars or trucks. Mobile sources are divided into two categories: on-road and nonroad vehicles/engines. Monitoring: See Ambient air monitoring. Motor Vehicle Emission Simulator (MOVES): An emissions modeling system that estimates emissions for mobile sources at the national, county, and project level for criteria air pollutants, air toxics and greenhouse gases. AirToxScreen 2018 Documentation A-5 National Emissions Inventory (NEI): A national database of air emissions data. EPA prepares NEI with input from many state and local air agencies, from tribes and from industry. This database contains information on stationary and mobile sources that emit criteria air pollutants and their precursors, as well as hazardous air pollutants. NEI includes estimates of annual emissions, by source, of air pollutants in each area of the country. NEI includes emission estimates for all 50 states, the District of Columbia, Puerto Rico and the U.S. Virgin Islands. National Mobile Inventory Model (NMIM): Computer tool containing EPA’s NONROAD model for estimating county-level inventories of nonroad mobile emissions. Noncancer risks: Risks associated with health effects other than cancer. Nonroad mobile sources: Mobile sources not used on roads and highways for transportation of passengers or freight. Nonroad sources include: • aircraft; • heavy equipment; • locomotives; • marine vessels; • recreation vehicles (snowmobiles, all-terrain vehicles, etc.); and • small engines and tools (lawnmowers, etc.). On-road mobile sources: Mobile sources used on roads and highways for transportation of passengers or freight. On-road sources include: • passenger cars and trucks • commercial trucks and buses; and • motorcycles. Percentile: Any one of the points dividing a set of values into parts that each contain 1/100 of the values. For example, the 75th percentile is a value such that 75 percent of the values are less than or equal to it. Polycyclic organic matter (POM): A broad class of compounds that includes polycyclic aromatic hydrocarbons. Polycyclic organic matter (POM) compounds form mainly from combustion and are present in the air as particles. Sources of POM emissions include: • vehicle exhaust; • forest fires and wildfires; • asphalt roads; • coal; • coal tar; • coke ovens; • agricultural burning; • residential wood burning; and • hazardous waste sites. Not all POM reported to EPA's National Emission Inventory is broken down by pollutant. So, we make some simplifying assumptions to model and assess the risk from the different pollutants that make up polycyclic organic matter. Reference concentration (RfC): An estimate of a continuous inhalation exposure unlikely to cause adverse health effects during a lifetime. This estimate includes sensitive groups such as children, asthmatics and the elderly. AirToxScreen 2018 Documentation A-6 Risk: The probability that adverse effects to human health or the environment will occur due to a given hazard (such as exposure to a toxic chemical or mixture of toxic chemicals). We can measure or estimate some risks in numerical terms (for example, one chance in a hundred). Rural: A county is considered “rural” if it does not contain a metropolitan statistical area with a population greater than 250,000 and the U.S. Census Bureau designates 50 percent or less of the population as “urban.” Note that this definition does not necessarily apply for any regulatory or implementation purpose. It is consistent with the definition EPA used in the analyses to support the Integrated Urban Air Toxics Strategy. Science Advisory Board (SAB): A panel of scientists, engineers and economists who provide EPA with independent scientific and technical advice. Secondary formation: See "Atmospheric transformation (Secondary Formation)" Secondary sources: See "Atmospheric transformation (Secondary Formation)" Sparse Matrix Operator Kernel Emissions (SMOKE): A modeling system that processes emissions data for use in air quality models. It uses the Biogenic Emission Inventory System (BEIS) to model biogenic emissions. It also has a feature to use MOVES emission factors, activity data and meteorological data to compute hourly gridded on-road mobile emissions. Stationary sources: Sources of air emissions that do not move. Stationary sources include large industrial sources such as power plants and refineries, smaller industrial and commercial sources such as dry cleaners, and residential sources such as residential wood combustion and consumer products usage. Stationary sources may be “major” or “area” sources based on definitions in the Clean Air Act. In AirToxScreen, we present sources as “point” and “nonpoint” rather than “major” and “area” sources. “Point” and “nonpoint” reflect how we modeled each emission source. Some smaller sources that are area sources in the inventory (based on the amount of their emissions) are modeled as point sources because their location was identified with latitude and longitude coordinates. Susceptibility: The increased likelihood of an adverse effect. Susceptibility is often discussed in terms of relationship to a factor describing a human population (for example, life stage, demographic feature or genetic trait). Toxicity weighting: A way to prioritize pollutant emissions based on risk. To calculate toxicity-weighted emissions, we multiply emissions from a facility or source being assessed by a toxicity factor for each pollutant. Pollutants that are more harmful (for the same emission rate) have a higher toxicity factor. By weighting the amount of a pollutant released to its toxicity, we can compare relative risk from different pollutants in emission inventories. Toxicity weighting is very useful if the number of pollutants is large, helping risk assessors focus on pollutants that contribute the most to risk. Typical: Describes a hypothetical person living at a census tract centroid (a reference point usually but not always located at the geographic center of a census tract) and doing the types of things (indoors and outdoors) that most people living in that tract would do. To describe that person’s risk, AirToxScreen divides the population into cohorts (groups assumed to have the same exposures each day) based on where they live, how old they are and their daily activity patterns. Fora census tract, we select age-appropriate activity patterns to model the range of exposure conditions for residents of the tract. We can then calculate a population-weighted typical exposure estimate for each cohort. We use this value to estimate risks for a “typical” individual residing in that tract. AirToxScreen 2018 Documentation A-7 Unit risk estimate (URE): The upper-bound excess lifetime cancer risk estimated to result from continuous exposure to an air toxic at a concentration of 1 microgram per cubic meter (µg/m3) in air. You can interpret the URE as follows: If the URE = 3 x 10-6 per µg/m3, as many as three more people might be expected to develop cancer per one million people exposed daily for a lifetime to 1 microgram (µg) of the chemical in 1 cubic meter (m3) of air. UREs are considered upper-bound estimates designed to keep us from underestimating risks. The true risk may be lower and is considered unlikely to be higher. In AirToxScreen, we multiply the model-output concentrations for an air toxic by that pollutant’s URE to calculate exposure risks from that air toxic. Upper-bound: A likely upper limit to the true value of a quantity. This is usually not a true statistical confidence limit. Upper-bound lifetime cancer risk: A likely upper limit to the true probability that a person will contract cancer over a 70-year lifetime due to a given hazard (such as exposure to a toxic chemical). This risk can be measured or estimated in numerical terms (for example, one chance in a hundred). Urban: A county is considered “urban” if it either includes a metropolitan statistical area with a population greater than 250,000 or the U.S. Census Bureau designates more than 50 percent of the population as “urban.” Note that this definition does not necessarily apply for any regulatory or implementation purpose. It is consistent with the definition EPA used in the analyses to support the Integrated Urban Air Toxics Strategy. Volatile organic compounds (VOCs): Chemicals emitted as gases from certain solids or liquids. VOCs are known for being common indoor air pollutants. EPA regulates VOCs in the ambient air because some cause adverse health effects and because they can react with other pollutants to form ozone and secondary air toxics. Cars and trucks, some industries, and even plants and trees emit VOCs. Weight-of-evidence for carcinogenicity (WOE): A system used by the EPA for characterizing the extent to which available data support the hypothesis that an agent causes cancer in humans. The approach, outlined in EPA’s Guidelines for Carcinogen Risk Assessment (2005), considers all scientific information in determining the WOE. Five standard descriptors are used as part of the WOE narrative: 1. Carcinogenic to humans. 2. Likely to be carcinogenic to humans. 3. Suggestive evidence of carcinogenic potential. 4. Inadequate information to assess carcinogenic potential. 5. Not likely to be carcinogenic to humans. Each of these descriptors is explained in its own glossary entry. You can read more details about WOE narratives in the AirToxScreen Technical Support Document. Carcinogenic to humans: This descriptor indicates strong evidence of human carcinogenicity. It covers different combinations of evidence. This descriptor is appropriate when there is convincing epidemiologic evidence of a link between human exposure and cancer. It might also be appropriate when a lesser weight of epidemiologic evidence is strengthened by other lines of evidence. This descriptor can be used when all of the following conditions are met: • there is strong evidence of an association between human exposure and either cancer or the key precursor events of the agent's mode of action, but not enough for a causal association; • there is extensive evidence of carcinogenicity in animals; • the mode(s) of carcinogenic action and associated key precursor events have been identified in animals; and AirToxScreen 2018 Documentation A-8 • there is strong evidence that the key precursor events that precede the cancer response in animals are anticipated to occur in humans and progress to tumors, based on available biological information. Likely to be carcinogenic to humans: This descriptor is appropriate when the weight of evidence is enough to show the potential to cause cancer in humans but does not meet all conditions necessary to be called “carcinogenic to humans.” Adequate evidence consistent with this descriptor covers a broad spectrum. At one end of the spectrum is a plausible association between human exposure to the agent and cancer and strong experimental evidence of carcinogenicity in animals. At the other, with no human data, the weight of experimental evidence shows animal carcinogenicity by a mode or modes of action that are relevant or assumed to be relevant to humans. The use of the term “likely” as a WOE descriptor does not correspond to a quantifiable probability. Moreover, additional information, for example, on mode of action, might change the choice of descriptor for the illustrated examples. Suggestive evidence of carcinogenic potential: This descriptor is appropriate when the weight of evidence suggests carcinogenicity, raising concern for potential carcinogenic effects in human, but the data are judged insufficient for a stronger conclusion. This descriptor covers a spectrum of evidence ranging from a positive cancer result in the only study on an agent to a single positive cancer result in an extensive database that includes negative studies in other species. Depending on the extent of the database, additional studies might or might not provide further insights. Inadequate information to assess carcinogenic potential: This descriptor is appropriate when available data are judged inadequate for applying one of the other descriptors. Additional studies generally would be expected to provide further insights. Not likely to be carcinogenic to humans: This descriptor is appropriate when the available data are considered strong enough for deciding that there is no basis for cancer concerns for humans. In some cases, there can be positive results in experimental animals, but the evidence is strong and consistent that each mode of action does not operate in humans. In other cases, the evidence can be convincing that the agent is not carcinogenic in humans or animals. “Not likely” applies only to the circumstances supported by the data. For example, an agent might be “not likely to be carcinogenic” by one route but not necessarily by another. In cases having positive animal experiment(s), but the results are judged not to be relevant to humans, the narrative discusses why the results are not relevant. Weather Research and Forecasting (WRF) model: A mesoscale numerical weather-prediction system for atmospheric research and weather forecasting. It can generate atmospheric conditions using real input data or idealized conditions. AirToxScreen 2018 Documentation A-9 Appendix B. Air Toxics Modeled in AirToxScreen A master pollutant list for AirToxScreen in spreadsheet format, “2018AirToxScreen_Pollutants.xlsx,” is provided in the Supplemental Data folder on the AirToxScreen website (https://www.epa.gov/AirToxScreen). This file includes all air toxics modeled in AirToxScreen and indicates the inventory types(s) reporting them. The file also includes the toxicity values used in AirToxScreen. The names shown in this spreadsheet match the terminology used in the 1990 Clean Air Act (CAA) Amendments; for example, the file lists “chromium compounds” but does not indicate which individual compounds containing chromium were modeled, and it lists four forms of xylenes (o-, m-, p- and mixed isomers), but these were grouped and modeled as a single entity. The file also contains indications about whether cancer risks and chronic noncancer hazard quotients were estimated for each air toxic. The spreadsheet also contains an air toxic names crosswalk and metal speciation factors used to conduct the modeling of emissions. This crosswalk contains a link between lists of air toxic names in two data bases used for AirToxScreen:  the names used in the National Emissions Inventory (NEI); and  the names used for AirToxScreen. In addition, the file contains the speciation of metal chemicals based on their metal mass fractions. The metal speciation factor was used to adjust modeled mass emissions prior to modeling and conducting risk calculations because metal toxicity is usually evaluated relative to the amount of metal ion present rather than the total mass of the metal compound. Most metal and cyanide compounds are reported in the NEI as just the metal or cyanide parts; consequently, most fractions are 1, including the two cyanide compounds. If the NEI data reporters did not adjust the emissions downward to account for just the metal part, a more health-protective (higher risk) result would be obtained. Table B-1 contains the air toxics that were not modeled for AirToxScreen and why. Note that although diesel PM was modeled for AirToxScreen and is included in the list of air toxics modeled, it is not categorized as a HAP in the CAA. Diesel PM emissions were computed based on PM10 emissions from on- road and nonroad mobile sources burning diesel or residual fuels. Note that NEI = National Emissions Inventory. AirToxScreen 2018 Documentation B-1 Table B-1. Pollutants excluded from AirToxScreen NEI Pollutant Pollutant Code (CAS Reason for Exclusion In Previous Number) a Assessments? Chromium III – In building the NEI, chromium VI is n emphasized, and chromium III may be missing data. 2,3,7,8-Tetrachlorodibenzo-p- 1746016 Dioxins and furans are not in the NEI due to n dioxin uncertainty in the completeness or accuracy of the S/L/T agency data for this group of pollutants. In addition, the most significant Other dioxins/furans multiple exposure route for dioxin is ingestion, not n inhalation, so dioxin’s relative contribution to AirToxScreen’s inhalation risk estimates likely would not be large. Radionuclides Radionuclides are not in the NEI due to n uncertainty in the completeness or accuracy of the S/L/T agency data for this group of pollutants. In addition, the NEI currently is not compatible with emissions reported in units other than mass, and therefore suitable emissions data have not been compiled for these substances on a national scale. DDE 72559 This pollutant was not reported to the 2018 n incorrectly NEI. referred to in the Section 112(b) list as 3547-04-4 Fine mineral fibers (including Fine mineral Rockwool has zero emissions in the 2018 NEI. n rockwool, slagwool and fine fibers: 383 Slagwool and fine mineral fibers are excluded mineral fibers) Rockwool: 617 from previous assessments. Slagwool: 616 Asbestos 1332214 Air concentrations of asbestos are often n measured in terms of numbers of fibers per unit volume, but the NEI provides tons, which cannot be converted. Diazomethane 334883 This pollutant has 0 emissions in the 2018 n NEI. Beta-propiolactone 57578 This pollutant has 0 emissions in the 2018 y NEI. aIn most cases, the NEI pollutant code is the same as the CAS number. In a few cases (e.g., coke oven emissions) a CAS number has not been assigned, and NEI uses a unique pollutant code. Appendix C. Estimating Background Concentrations for AirToxScreen The memo in this appendix contains the methods we used to estimate background concentrations for AirToxScreen. AirToxScreen 2018 Documentation C-2 AirToxScreen 2018 Documentation C-3 AirToxScreen 2018 Documentation C-4 AirToxScreen 2018 Documentation C-5 AirToxScreen 2018 Documentation C-6 AirToxScreen 2018 Documentation C-7 AirToxScreen 2018 Documentation C-8 AirToxScreen 2018 Documentation C-9 AirToxScreen 2018 Documentation C-10 AirToxScreen 2018 Documentation C-11 Appendix D. Model Evaluation Summaries EPA performed model evaluations for other AirToxScreen pollutants. These evaluations, including graphics, can be found in the Supplemental Data folder on the AirToxScreen website. AirToxScreen 2018 Documentation D-1 Appendix E. Exposure Factors for AirToxScreen The memorandum contained within this Appendix (below) describes in detail how EPA developed exposure factors for each chemical assessed in AirToxScreen. These calculated exposure factors can be found in the Supplemental Data folder accompanying this TSD. AirToxScreen 2018 Documentation E-1 MEMORANDUM To: Ted Palma, U.S. EPA Office of Air Quality Planning and Standards From: John Hader and Chris Holder, ICF Date: June 25, 2018 Re: HAPEM7 Modeling for AirToxScreen AirToxScreen is a nationwide EPA modeling assessment of the air concentrations, exposure concentrations,1 and potential human health cancer and noncancer risks and hazards associated with inhalation exposure to Hazardous Air Pollutants (HAPs) emitted by manmade and natural sources of outdoor origin. As part of AirToxScreen, the Hazardous Air Pollutant Exposure Model (HAPEM), which is a U.S. Environmental Protection Agency (EPA) model, is used to perform screening-level assessments of long-term inhalation exposures to HAPs. For AirToxScreen, ICF (“we”) used Version 7 of HAPEM (HAPEM7), with modeled air- concentration data provided by EPA, to model exposure concentrations for seven selected HAPs, stratified by census tract and emission source type. These HAPs also are used as surrogates for the many other HAPs included in AirToxScreen, as described in detail below. We used the annual-average exposure concentrations by age group, provided by the HAPEM modeling, to estimate lifetime-average exposure concentrations. We then divided the lifetime exposure concentrations by the corresponding annual-average air concentrations creating “exposure factors”. EPA can then estimate exposure to each AirToxScreen HAP in each census tract by multiplying the air concentration for the HAP by the exposure factor derived for the chemical’s surrogate HAP. In this memorandum, we discuss HAPEM7, how we assigned the gas or particulate phase of the AirToxScreen HAPs, how EPA selected the seven HAPs modeled in HAPEM, how we set up the HAPEM runs, and how we developed the exposure factors. 1 Exposure concentrations are time-averaged air concentrations to which a simulated individual is exposed. The time averaging considers the air concentrations in each place where the individual is simulated to spend time, and how long he or she spends there. AirToxScreen 2018 Documentation E-1 1. Introduction to HAPEM and its Use in AirToxScreen HAPEM is a model used by EPA to perform screening-level assessments of long-term inhalation exposures to HAPs. We updated HAPEM and its default, ancillary files in 2015, as discussed in the 2011 NATA documentation.2 This latest version of HAPEM (HAPEM7)3 estimates exposure concentrations using • demographic and behavior data from the 2010 U.S. Census (covering all 50 U.S states, the District of Columbia, Puerto Rico, and the U.S. Virgin Islands),4 • compiled activity data from a recent version of EPA’s Consolidated Human Activity Database (CHAD),5 and • updated data on population proximities to major roadways. • HAPEM7 stratifies exposure concentrations by • location (U.S. Census tract), • time of day, • age group, and • the individual emission source types and HAPs being modeled. HAPEM is a probabilistic model that selects some model inputs from distributions of data to characterize exposure. These elements include commuting patterns, activity patterns, employment work location, microenvironmental factors, and (if desired) air concentrations. Including more simulated individuals (termed “replicates” in HAPEM) can increase the range of values selected for these probabilistic elements, across the simulated population. 2. Air Quality Inputs to HAPEM7 2.1. HAPs Assessed for AirToxScreen For AirToxScreen, EPA compiled an inventory of the annual mass of HAPs emitted from four source types, namely • point, • non-point, • on-road mobile, and 2 Appendix G of the Technical Support Document of EPA’s 2011 National Air Toxics Assessment. December 2015. EPA Office of Air Quality Planning and Standards. Available at https://www.epa.gov/sites/production/files/2015- 12/documents/2011-nata-tsd.pdf. 3 As of October 23, 2017, HAPEM7 is available for download from https://www.epa.gov/fera/human-exposure- modeling-hazardous-air-pollutant-exposure-model-hapem. The HAPEM7 User’s Guide also is available from that link, and it contains detailed discussions of the model’s functionality and input files. 4 For additional information on the census data parameterized in HAPEM, see Appendix B of The HAPEM User’s Guide, Version 7 (July, 2015), available at https://www.epa.gov/sites/production/files/2015- 12/documents/hapem7usersguide.pdf. 5 For additional information on CHAD and how its data were parameterized in HAPEM, see https://www.epa.gov/healthresearch/consolidated-human-activity-database-chad-use-human-exposure-and- health-studies-and and Appendix B of The HAPEM User’s Guide, Version 7 (July, 2015), available at https://www.epa.gov/sites/production/files/2015-12/documents/hapem7usersguide.pdf. AirToxScreen 2018 Documentation E-2 • non-road mobile. These emissions were nationwide, typically at spatial resolutions of 12-km grids down to specific points. The assessed HAPs are shown in the far left column of Table (under the “Chem” heading). In some cases, emissions of several chemicals were aggregated and modeled as a group, as indicated on the right side of Table (under the “AirToxScreen” heading; the columns under the “AirToxScreen” heading provide the chemicals and chemical groups as they were modeled for AirToxScreen). Using these estimated emissions, temporal emission profiles based on source type, and hourly meteorology data selected by proximity, EPA used AERMOD6 and CMAQ7 to estimate ambient air concentrations. The EPA air-modeling results provided for HAPEM exposure modeling were annual-average air concentrations at the level of census tracts, stratified by HAP, source type, and at three-hour increments across the day. 2.2. Specifying Chemical Phases for HAPEM Modeling For all source types, HAPEM7 models concentrations for three chemical phases. These phases are • gaseous (“G”), • particulate (“P”), and • variable for typical atmospheric conditions (“G/P”). HAPEM estimates concentrations for user-specified microenvironments (ME). An ME is a generic location, such as indoors at home or outdoors near roadway, where a person spends time and breathes air, leading to chemical exposure. In HAPEM, MEs are treated as having well-mixed air concentrations at a given time, and the model estimates those ME concentrations relative to the outdoor air concentrations provided as input to the model. HAPEM7 is parameterized for 18 MEs, including eight indoor MEs (including residential, school, office, bar/restaurant, and similar), two in-vehicle MEs (personal vehicle and public transit), six outdoor MEs near roadways, and two outdoor MEs not near roadways. Starting with outdoor ambient chemical concentrations, the model uses phase-specific penetration and proximity factors to estimate the ME chemical concentrations. A penetration factor describes how much of the chemical in the outdoor air penetrates the indoor or in-vehicle air. The penetration factor is a ratio in the form of [chemical air concentration in an indoor/in- vehicle ME] to [chemical air concentration directly outdoors of the ME]. A proximity factor parameterizes the heterogeneity of air concentrations across a Census tract. It describes how much higher or lower we expect the outdoor air concentration to be in the immediate vicinity of the ME relative to the Census tract outdoor concentration supplied in the input file, and it is a ratio in the form of [outdoor chemical air concentration in the immediate vicinity of the ME] to [tract-level outdoor chemical air concentration input to the model]. HAPEM7 has one set of penetration and proximity factors for mobile on-road sources, and a second set for all other source types. For mobile on-road sources, there are four categories of factors: three are for 6 AERMOD: the atmospheric dispersion model developed by the American Meteorological Society and the EPA Regulatory Model Improvement Committee. See https://www.epa.gov/scram/air-quality-dispersion-modeling- preferred-and-recommended-models#aermod 7 CMAQ: EPA’s Community Multiscale Air Quality model. See https://www.epa.gov/cmaq AirToxScreen 2018 Documentation E-3 specific HAPs (benzene, 1,3-butadiene, and diesel particulate matter [DPM]8) and a fourth category representing all other HAPs. To prepare for AirToxScreen exposure modeling, we categorized each of the AirToxScreen HAPs as G, P, or G/P based on available boiling-point data, as defined in Table . We provide each HAP’s boiling point and assigned HAPEM7 phase in Table (left side under the “Chem” heading). If chemicals were modeled as a group, we assigned the group a G, P or G/P designation, as indicated in Table (right side under the “AirToxScreen” heading) and as discussed later in this section. We obtained the vast majority of boiling-point values from • the Centers for Disease Control (CDC),9 • the National Institutes of Health (NIH),10 or • the Royal Society of Chemistry using their ChemSpider web site.11 These were judged to be the most reputable, comprehensive, user-friendly, and readily available sources of chemical boiling-point data. Each source allows the user to search by chemical name or Chemical Abstract Service (CAS) Registry Number. For HAPs whose boiling points we could not identify using these three sources, we searched a variety of additional data sources. We provide the source of each chemical’s boiling-point value in the “Source” column of Table . We could not identify empirical boiling-point data for some of the HAPs. For 25 HAPs (9 percent of all evaluated HAPs), we used predictive boiling points in order to categorize their chemical phases. These values came from ChemSpider, which generates estimated boiling points using three software modules: EPA’s EPIsuite, the ACD/Labs Percepta Platform - PhysChem Module, and ChemAxon’s predictive software platform. We typically selected the ACD/Labs values when available because these values were presented with confidence intervals and the conditions under which the boiling-point values were predicted (typically standard temperature and pressure); the other two platforms did not provide such information. If an ACD prediction was not available, we used the EPA EPIsuite value. Those boiling points that are predictive rather than empirical are flagged with a “P” in Table . Note that the boiling-point ranges in Table have imprecise endpoints (e.g., the high end of boiling points for G HAPs covers a range of 240 to 260°C). Forty of AirToxScreen HAPs have boiling points within these imprecise endpoints, meaning there was some uncertainty associated with assigning the phases for these HAPs. In order to make appropriate phase designations, we conducted a literature review for each of these HAPs to identify relevant information regarding its typical physical state. For example, 1-nitropyrene has a boiling point of 445°C, within the overlap of G/P and P boiling points. A review of the literature yielded several studies and reports identifying 1-nitropyrene as a particulate at typical atmospheric conditions, leading us to assign a designation of “P” to this HAP with a high degree of confidence. Where literature searches were uninformative, we assigned HAP phase based on the categorizations used for HAPEM5 to support the 1999 NATA. The combination of the additional literature review and consultation of 8 DPM is not a HAP as defined under Section 112(b) of the Clean Air Act; however, for ease of discussion we refer to it as a HAP in this memorandum. 9 CDC: http://www.cdc.gov/niosh/npg. 10 NIH: http://pubchem.ncbi.nlm.nih.gov/. 11 CS: http://www.chemspider.com. AirToxScreen 2018 Documentation E-4 the HAPEM5 designations allowed us to make phase designations for these HAPs. Seventeen HAPs have boiling points within the 240–260°C range; based on the literature review and HAPEM5 designations, we categorized 16 as G and 1 as G/P. Twenty HAPs have boiling points within the 400–480°C range; we categorized 16 as P and 4 as G/P. Three HAPs have wide- ranging boiling-point values; we categorized the two with higher ranges of boiling points (extending into the 400–480°C range) as G/P, and we categorized the one with a lower range of boiling points (extending only into the 240–260°C range) as G. In addition to the above 40 HAPs, boiling-point values were specified by the CDC or EPA as “variable” for three HAPs (see the “V” designations in the boiling-point-value column in Table ). We categorized two of these HAPs (coke oven emissions and cyanide) as G/P. We categorized coke oven emissions based on an EPA characterization of this pollutant as consisting of a mixture of particulates, volatiles, and semi-volatiles.12 We characterized cyanide based on the fact that cyanide is not typically found in isolation in nature, but rather in a variety of compounds, some of which are typically solid (e.g., calcium cyanide, sodium cyanide) and some of which are typically gaseous (e.g., hydrogen cyanide).13 The third HAP with “variable” boiling points was DPM and we assumed that it is typically a particulate. For five HAPs, boiling-point data were either unavailable or were ill-defined (see the “NA”, “D”, and “S” designations in the boiling-point-value column in Table ); one was Fine Mineral Fibers, which we categorized as P, and we left the remaining four (Extractable Organic Matter (EOM), propoxur, quinone, and toxaphene) uncategorized. For these four HAPs, EPA was health protective and set exposure concentrations equal to ambient outdoor concentrations per census tract, as indicated by “EF=1” designations in the “Matching Chemical Modeled in HAPEM” columns in Table . A setting of EF=1 is health-protective because it assumes that people always breathe outdoor ambient air, receiving no protection that may otherwise be afforded by the barriers and filtration systems of a building or vehicle (that is, a penetration factor of 1 rather than below 1). Without indoor sources of emissions (which are not included in AirToxScreen), outdoor air concentrations necessarily are the same or higher than those of indoor or in-vehicle MEs. Where EPA modeled several HAPs as one group, we assigned the group’s chemical phase based on the phase most frequently assigned to the group’s component chemicals. Two AirToxScreen chemical groups did not easily accommodate this criterion, as discussed in the bullets below. • For AirToxScreen chemical group CHROMHEX, its two components are Chromium (VI) and Chromium Trioxide, which are P and G/P phases, respectively. We assumed that Chromium (VI) was the primary chemical in the group (the one emitted in higher amounts) and assigned the group to P accordingly. • For AirToxScreen chemical group CYANIDE, its two components are cyanide and hydrogen cyanide (HCN). Cyanide has a variable boiling point, while HCN’s boiling point is 26ºC (making it a G chemical). Because HCN has the higher level of emissions, we assigned the group to G accordingly. 12 https://www.epa.gov/sites/production/files/2016-01/documents/coke_rra.pdf 13 http://www.atsdr.cdc.gov/ AirToxScreen 2018 Documentation E-5 2.3. HAPs Modeled in HAPEM7 for the AirToxScreen As noted previously, except for air-concentration data, HAPEM7 does not use any settings or inputs that are specific to each individual AirToxScreen HAP. Penetration and proximity factors vary only by the phase of the HAP, not by each HAP individually (except for mobile on-road sources, where benzene, 1,3-butadiene, and DPM have specific penetration and proximity factors). Therefore, except for those three HAPs for mobile on-road sources, the model inputs for HAPs of a given phase vary only by air concentration, and so their exposure concentrations output by HAPEM (i.e., the time-averaged air concentrations across the MEs where individuals are simulated to spend time) vary only by their air concentrations. This means that we can estimate the exposure concentration of one HAP of a given phase in HAPEM, and then in post- processing we can calculate the relationship between its air and exposure concentrations and apply this relationship to other HAPs of the same phase, significantly reducing the level of effort needed to conduct the exposure modeling. EPA identified a small subset of AirToxScreen HAPs for which to conduct HAPEM exposure modeling. AirToxScreen uses the modeling results of this subset as-is, and uses the results as surrogate values for the remaining AirToxScreen HAPs not modeled in HAPEM. EPA used the following decision criteria in identifying this subset of HAPs to model. 1. The subset must include at least one HAP per phase (at least one G HAP, one P HAP, and one G/P HAP). 2. Each emission source type must be represented for each phase (for example, G HAPs must collectively cover the point, non-point, on-road mobile, and non-road mobile source types, and so on for P and G/P HAPs). 3. It is preferred that the selected HAPs have the potential to pose higher risks to human health nationwide, relative to other HAPs. 4. It is preferred that the selected HAPs be emitted in many spatially-diverse locations across the United States. Using these criteria, EPA selected seven HAPs for exposure modeling. We list these HAPs below and in Table . The right four columns of Table indicate how these modeled HAPs were mapped to the other AirToxScreen HAPs not modeled in HAPEM. • Benzene and 1,3-butadiene are emitted by many processes (and all four modeled source types) in nearly all U.S. locations. They each also have specific penetration and proximity factors for emissions from mobile on-road sources. Benzene was selected to be the surrogate for all other G HAPs. • Unspeciated, generic PAHs (“PAH, total”) are emitted by all four source types and from a wide variety of processes, so it was selected to be the surrogate for all G/P HAPs except coke oven emissions, which is a special case that was modeled by itself for point sources but not used as a surrogate for any other HAPs. • Chromium (VI) is a highly toxic HAP that was selected as the surrogate for all P HAPs emitted by point and non-point sources, except for DPM, which was modeled as itself for non-point and mobile sources. DPM has specific penetration and proximity factors for emissions from mobile on-road sources. Note that the AirToxScreen air- concentration modeling included chromium (VI) emissions from all four source types, but its use as an exposure surrogate included only point and non-point sources because those are its major emitters. AirToxScreen 2018 Documentation E-6 • Nickel was selected as the surrogate for P HAPs emitted by mobile sources, except for DPM, which was modeled as itself for non-point and mobile sources. Nickel is emitted by a variety of processes across the United States. Note that AirToxScreen air- concentration modeling included nickel emissions from all four source types, but its use as an exposure surrogate included only mobile sources because chromium (VI) was designated as the more appropriate surrogate for point and non-point sources. We used the air-concentration modeling outputs provided by EPA for these seven HAPs, stratified by source type, three-hour increment of the day, and census tract, as the air-quality inputs for seven HAPEM7 runs. 3. HAPEM7 Runs 3.1. Design For each of the seven HAPEM7 runs (one run for each modeled chemical), we used the HAPEM7 default census- and CHAD-based files. Each run assessed the 18 HAPEM7 MEs and all populated census tracts in the United States, Puerto Rico, and the U.S. Virgin Islands. The air-quality inputs were in three-hour periods for each HAP, tract, and source type (that is, for each day, average air concentrations for the first three hours of the day, the second three hours, etc., totaling eight values per day for each HAP, tract, and source type). We used the six default HAPEM7 age groups (discussed below in Section 3.3) and three day-types (summer weekdays, non-summer weekdays, and weekends). We linked each HAP to its appropriate HAPEM7 penetration- and proximity-factors files, and we used 30 replicates (simulated individuals) per age group evaluated per tract. ICF has previously shown that the tract-mean exposure concentrations of an age group are reasonably stable when using 30 replicates, striking a good balance between model runtime and the stability of model outputs (that is, at a 95-percent probability level, 30 replicates are needed to estimate the tract-mean exposures to within 20 percent).14 3.2. Quality Assurance and Quality Control We reviewed the HAPEM7 control files (“parameters” files) for accuracy, and then we reviewed the log, “counter,” and “mistract” HAPEM7 output files to identify any potential errors in the modeling. One tract identifier was modified by the U.S. Census after 2010 (after the HAPEM7 population files were developed using the 2010 census)—the AirToxScreen air-concentration data used the tract’s latest identifier of 51019050100, which we renamed to its old identifier (51515050100) for HAPEM7 modeling (its geography did not change). We switched it back to the current identifier in the results delivered to EPA. We detected no other errors in the inputs or outputs. We present below additional information gleaned from the QA/QC activities. • EPA modeled all U.S. census tracts to develop air concentration estimates. However, 584 tracts (less than 1 percent of U.S. tracts) were not modeled in HAPEM because the 2010 Census indicated zero population. 14 Source: “Benzene Case Study for PO3-NTA006-ICF”, a technical memorandum from Arlene Rosenbaum, Michael Huang, and Jonathan Cohen (ICF) to Ted Palma (U.S. EPA), September 30, 2002. AirToxScreen 2018 Documentation E-7 • A total of 73,450 tracts were modeled for exposure concentrations for the AirToxScreen. 3.3. Post-processing for Exposure Factors We utilized the HAPEM7 outputs for the seven modeled HAPs to estimate exposure factors that EPA then applies to all HAPs assessed in AirToxScreen, based on HAP phase and source type. For each tract and source type in a HAP-specific HAPEM7 run, we calculated the estimated 70- year lifetime-average exposure concentration for each modeled replicate. We calculated these average concentrations as the time-weighted average of exposures for the six HAPEM7 age groups, as shown below. Lifetime average exposure conc. = [ages 00–01 exposure conc. × (2/70)] + [ages 02–04 exposure conc. × (3/70)] + [ages 05–15 exposure conc. × (11/70)] + [ages 16–17 exposure conc. × (2/70)] + [ages 18–64 exposure conc. × (47/70)] + [ages 65+ exposure conc. × (5/70)] We then calculated the population-median lifetime-average exposure concentration in each tract (median calculated from the 30 replicates in a tract), stratified by HAP and source type. For a given HAP and tract, summing these median concentrations across source types yields an estimated “typical” lifetime-average HAP exposure concentration from all contributing sources. For each assessed HAP in a tract, we calculated exposure factors by dividing these median lifetime-exposure concentrations by the corresponding annual-average outdoor air concentrations (stratified by source type as well as the all-source total). EPA can then multiply these exposure factors by the air concentrations of any HAP of the same phase- type, resulting in estimated lifetime-exposure concentrations for that HAP. For example, for a given tract, to estimate the exposure concentrations of fluorene (a G/P HAP) emitted by non-point sources, EPA will multiply its non-point air concentrations by the non-point exposure factors for “PAH, total” (the surrogate for G/P HAPs). The exposure factors typically were between roughly 0.4 and 1.0 (larger factors generally for on-road mobile sources and gases; smaller factors typically for the other source types and particulates). The primary limitation of this exposure-factor method is related to commuting. HAPEM7 simulates the movement of replicates to and from work. Some replicates will work in the same tract where they live, so that the outdoor air concentrations of their “home tract” equal those of their “work tract”. For these replicates, the exposure-factor method is accurate, because the exposure factors account only for home-tract exposure to the modeled HAP, and the air- concentration denominator in the factor calculation also corresponds only to home-tract air concentrations of the unmodeled HAP. For replicates who commute outside their home tract, they work in a tract with different air concentrations than in their home tract, affecting their exposure during work hours (usually less than half the day). When the work air tract concentrations are similar to the home tract air concentrations, then the exposure factor method provides a good approximation of exposure for unmodeled HAPs. However, when work tract and home tract air concentrations are highly dissimilar, the exposure factors method estimates are less accurate because the exposure factors account for some home-tract exposures and some very different work-tract exposures to the modeled HAP, while the air-concentration AirToxScreen 2018 Documentation E-8 denominator in the factor calculation corresponds only to home-tract air concentrations of the unmodeled HAP. Instances of highly dissimilar work tract and home tract concentrations are most easily detected when exposure factors are above 1, indicating the work-tract air concentrations are higher than those of the home tract. For a relatively small number of tracts, exposure factors were larger than 10 and, for seven tracts, exposure factors were larger than 100. In an effort to improve the representativeness of the modeled exposure factors and their application to unmodeled HAPs, we limited exposure factors to be no larger than the values shown in Table (exposure factors that were above these values were set to the maximum values). To define these upper limits on the exposure factors, we calculated the median exposure factor and standard deviation of exposure factors across all tracts for each combination of HAP and source type; the upper limit was then defined as median+standard deviation. The calculations for medians and standard deviations did not include exposure factors of 100 or larger, which we considered to be outliers. They also did not consider instances where air concentrations or exposure concentrations were zero, which occurred at 584 tracts as discussed in Section 3.2. In these instances, we set the exposure factors equal to 1.0, indicating that the exposure concentration equals the air concentration. All applications of the exposure factors for the AirToxScreen use these truncated values, including exposure factors initially set equal to 1.0 and including the HAPs explicitly modeled in HAPEM7. 4. Outputs We performed quality assurance checks of the post-processing by thoroughly reviewing and testing the post-processing R code in order to ensure calculations were being performed correctly, and we also performed other broad checks to ensure all records were properly read in and that flagged records were properly processed. AirToxScreen 2018 Documentation E-9 Table E-1. HAPs Assessed in AirToxScreen, with their HAPEM7 HAP Phases and Surrogate Chemical Assignments  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Acetaldehyde 20 CDC G ACETALD G Benz Benz Benz Benz Acetamide 165 CDC G ACETAMIDE G Benz Benz Benz Benz Acetonitrile 82 CDC G ACETONIT G Benz Benz Benz Benz Acetophenone 202 NIH G ACETOPHEN G Benz Benz Benz Benz Acrolein 53 CDC G ACROLEI G Benz Benz Benz Benz Acrylic acid 141 CDC G ACRYLCACID G Benz Benz Benz Benz Acrylamide 175 CDC G ACRYLMID G Benz Benz Benz Benz Acrylonitrile 77 CDC G ACRYLONITRL G Benz Benz Benz Benz 2-Acetylaminofluorene 400 P CS G/P ACTYALFLUR2 G/P PAH PAH PAH PAH Allyl chloride 45 CDC G ALLYLCHLORD G Benz Benz Benz Benz 4-Aminobiphenyl 302 CDC G/P AMNOBIPNYL4 G/P PAH PAH PAH PAH Aniline 184 CDC G ANILINE G Benz Benz Benz Benz Antimony 1,587 CDC P ANTIMONY P Cr6 Cr6 Ni Ni Arsenic 612 CDC P ARSENIC P Cr6 Cr6 Ni Ni Benzene 80 CDC G BENZENE G Benz Benz Benz Benz Benzidine 400 CDC G/P BENZIDINE G/P PAH PAH PAH PAH Benzotrichloride 221 NIH G BENZOTRICHL G Benz Benz Benz Benz Benzyl chloride 179 NIH G BENZYLCHLO G Benz Benz Benz Benz BERYLLIUM 2,500 NIH P BERYLLIUM P Cr6 Cr6 Ni Ni Biphenyl 256 NIH G BIPHENYL G Benz Benz Benz Benz Bis(chloromethyl)ether 106 CDC G BISCHLROME G Benz Benz Benz Benz Bromoform 149 CDC G BROMOFORM G Benz Benz Benz Benz Bis(2-ethylhexyl)phthalate 386 CDC G/P BS2TYLHXYLP G/P PAH PAH PAH PAH beta-Propiolactone 162 http://www.cdc.gov/niosh/ G BT-PRPLACTN G Benz Benz Benz Benz docs/81- 123/pdfs/0528.pdf 1,3-Butadiene 138 NIH G BUTADIE G Buta Buta Buta Buta Cadmium 765 CDC P CADMIUM P Cr6 Cr6 Ni Ni Calcium cyanamide >2,444 CDC P CALCIUMCYA P Cr6 Cr6 Ni Ni Captan 314 CS G/P CAPTAN G/P PAH PAH PAH PAH Carbaryl 315 CS G/P CARBARYL G/P PAH PAH PAH PAH Carbon disulfide 47 CDC G CARBNDISULF G Benz Benz Benz Benz Carbonyl sulfide -50 NIH G CARBNYLSUL G Benz Benz Benz Benz Carbon tetrachloride 77 CDC G CARBONTET G Benz Benz Benz Benz Catechol 245 CDC G CATECHOL G Benz Benz Benz Benz Chloroform 62 CDC G CHCL3 G Benz Benz Benz Benz Chloroacetic acid 106 CDC G CHLACETACD G Benz Benz Benz Benz 2-Chloroacetophenone 244 CDC G CHLACETPH2 G Benz Benz Benz Benz Chloramben 312 CS G/P CHLORAMBEN G/P PAH PAH PAH PAH Chlordane 175 NIH G CHLORDANE G Benz Benz Benz Benz Chlorine -33 CDC G CHLORINE G Benz Benz Benz Benz Chloroprene 59 CDC G CHLOROPRENE G Benz Benz Benz Benz Chlorobenzilate 146 NIH G CHLROBZLAT G Benz Benz Benz Benz Chlorobenzene 132 CDC G CHLROBZNE G Benz Benz Benz Benz Chloromethyl methyl ether 59 CDC G CHLROMME G Benz Benz Benz Benz Chromium (VI) 2,642 CDC P CHROMHEX P Cr6 Cr6 Ni Ni Chromium Trioxide 250 CDC G/P AirToxScreen 2018 Documentation E-10  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Chromium III 2,672 http://books.google.com/b P CHROMTRI P Cr6 Cr6 Ni Ni ooks?id=SFD30BvPBhoC &pg=PA123&lpg=PA123 &dq=chromium+III+meltin g+point&source=bl&ots=u pHljDrKMy&sig=dISMKFL 5z0sVI0z8Z4NhlsFHggE &hl=en&sa=X&ei=4nklVP LvJ4LS8AGbiYD4DA&ve d=0CFkQ6AEwCQ#v=on epage&q=chromium%20II I%20melting%20point&f=f alse Ethylene dichloride 83 CDC G CL2_C2_12 G Benz Benz Benz Benz Trichloroethylene 87 CDC G CL3_ETHE G Benz Benz Benz Benz Cobalt 3,100 CDC P COBALT P Cr6 Cr6 Ni Ni Coke oven emissions V CDC G/P COKEOVEN G/P Coke Coke Coke Coke Cresol/Cresylic Acid (Mixed Isomers) 202 CDC G CRESOLS G Benz Benz Benz Benz m-Cresol 202 CDC G o-Cresol 191 CDC G p-Cresol 202 CDC G Cumene 152 CDC G CUMENE G Benz Benz Benz Benz Cyanide V CDC G/P CYANIDE G Benz Benz Benz Benz Hydrogen cyanide 26 CDC G 2,4-Dichlorophenoxy Acetic Acid 345 P CS G/P D24SALTS G/P PAH PAH PAH PAH Diethyl sulfate 210 NIH G DETHLSULFAT G Benz Benz Benz Benz Diazomethane -23 CDC G DIAZOMETHAN G Benz Benz Benz Benz Dibenzofuran 287 NIH G/P DIBENZOFUR G/P PAH PAH PAH PAH 1,2-Dibromo-3-chloropropane 196 CDC G DIBRM3CHLPR G Benz Benz Benz Benz Dibutyl Phthalate 340 CDC G/P DIBUTYLPHTH G/P PAH PAH PAH PAH 3,3'-Dichlorobenzidine 400 NIH G/P DICHLBZD33P G/P PAH PAH PAH PAH Dichlorvos 140 at 40 NIH G DICHLORVOS G Benz Benz Benz Benz mmHG 1,4-Dichlorobenzene 173 CDC G DICHLRBNZN G Benz Benz Benz Benz Dichloroethyl ether 177 CDC G DICHLROETET G Benz Benz Benz Benz 1,3-Dichloropropene 108 NIH G DICLPRO13 G Benz Benz Benz Benz Diesel PM V http://www.epa.gov/region P DIESEL_PM10 P DPM DPM DPM DPM 1/eco/airtox/diesel.html Diethanolamine 268 NIH G/P DIETHNLAMIN G/P PAH PAH PAH PAH Dimethylcarbamoyl chloride 167 NIH G DIMTHYCARB G Benz Benz Benz Benz N,N-Dimethylformamide 153 CDC G DIMTHYFORM G Benz Benz Benz Benz 4-Dimethylaminoazobenzene 371 CS G/P DIMTHYLAMAZ G/P PAH PAH PAH PAH Dimethyl phthalate 284 NIH G/P DIMTHYLPHTH G/P PAH PAH PAH PAH Dimethyl sulfate 188 NIH G DIMTHYLSULF G Benz Benz Benz Benz 3,3'-Dimethylbenzidine 300 CDC G/P DIMTYLBZ33P G/P PAH PAH PAH PAH 1,1-Dimethyl Hydrazine 64 CDC G DIMTYLHYD11 G Benz Benz Benz Benz 3,3'-Dimethoxybenzidine 391 CS G/P DIMTYOXB33P G/P PAH PAH PAH PAH 2,4-Dinitrotoluene 300 NIH G/P DINITOTOL24 G/P PAH PAH PAH PAH 4,6-Dinitro-o-cresol 312 CDC G/P DINITROCRES G/P PAH PAH PAH PAH 2,4-Dinitrophenol 312 CS G/P DINITROPH24 G/P PAH PAH PAH PAH 1,2-Diphenylhydrazine 293 NIH G/P DIPYLHYZ12 G/P PAH PAH PAH PAH N,N-dimethylaniline 192 CDC G DMTYLANILNN G Benz Benz Benz Benz Epichlorohydrin 118 NIH G EPICHLORO G Benz Benz Benz Benz 1,2-Epoxybutane 63 NIH G EPOXYBUTA12 G Benz Benz Benz Benz Ethylene dibromide 131 CDC G ETHDIBROM G Benz Benz Benz Benz Ethyleneimine 56 CDC G ETHENIMINE G Benz Benz Benz Benz AirToxScreen 2018 Documentation E-11  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Ethylene glycol 197 CDC G ETHGLYCOL G Benz Benz Benz Benz Ethylidene dichloride -17 CDC G ETHIDDICHLD G Benz Benz Benz Benz Ethylene thiourea 230 CDC G ETHTHUREA G Benz Benz Benz Benz Ethyl acrylate 99 CDC G ETHYLACRYL G Benz Benz Benz Benz Ethyl benzene 136 CDC G ETHYLBENZ G Benz Benz Benz Benz Ethyl carbamate 185 NIH G ETHYLCARBA G Benz Benz Benz Benz Ethyl chloride -139 NIH G ETHYLCHLRD G Benz Benz Benz Benz Ethylene oxide 11 CDC G ETOX G Benz Benz Benz Benz Formaldehyde -21 CDC G FORMALD G Benz Benz Benz Benz 1,2-Dimethoxyethane 82 NIH G GLYCOLETHR G Benz Benz Benz Benz 2-(Hexyloxy)ethanol 258 NIH G 2-Butoxyethyl acetate 192 CDC G 2-Propoxyethyl Acetate 184 CS G Butyl carbitol acetate 245 NIH G Carbitol acetate 219 NIH G Cellosolve Acetate 145 CDC G Cellosolve Solvent 124 CDC G Diethylene glycol diethyl ether 189 E CS G Diethylene glycol dimethyl ether 161 E CS G Diethylene glycol ethyl methyl ether 168 P CS G Diethylene glycol monobutyl ether 230 NIH G Diethylene glycol monoethyl ether 196 NIH G Diethylene glycol monomethyl ether 194 E CS G Ethoxytriglycol 256 NIH G Ethylene glycol methyl ether 124 CDC G Ethylene Glycol Monomethyl Ether Acetate 145 CDC G Ethylene glycol mono-sec-butyl ether 192 CS G Glycol ethers 120–240 http://msdssearch.dow.co G m/PublishedLiteratureDO WCOM/dh_012d/0901b8 038012d976.pdf?filepath =oxysolvents/pdfs/noreg/ 110- 00977.pdf&fromPage=Ge tDoc Methoxytriglycol 249 http://msdssearch.dow.co G m/PublishedLiteratureDO WCOM/dh_012d/0901b8 038012d976.pdf?filepath =oxysolvents/pdfs/noreg/ 110- 00977.pdf&fromPage=Ge tDoc N-Hexyl carbitol 260 E CS G Phenyl cellosolve 245 E CS G Propyl cellosolve 150 http://msdssearch.dow.co G m/PublishedLiteratureDO WCOM/dh_012d/0901b8 038012d976.pdf?filepath =oxysolvents/pdfs/noreg/ 110- 00977.pdf&fromPage=Ge tDoc Triethylene Glycol Dimethyl Ether 215 E CS G Triglycol monobutyl ether 278 NIH G/P Hydrochloric acid -85 CDC G HCL G Benz Benz Benz Benz Heptachlor 392 CS G/P HEPTACHLOR G/P PAH PAH PAH PAH AirToxScreen 2018 Documentation E-12  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Hexamethylene Diisocyanate 212 NIH G HEXAMTHLE G Benz Benz Benz Benz Hexane 69 CDC G HEXANE G Benz Benz Benz Benz Hexachloroethane 187 CDC G HEXCHLETHN G Benz Benz Benz Benz Hexachlorocyclopentadiene 238 CDC G HEXCHLPNTD G Benz Benz Benz Benz Hexachlorobutadiene 215 CDC G HEXCHLRBT G Benz Benz Benz Benz Hexachlorobenzene 325 NIH G/P HEXCHLROBZ G/P PAH PAH PAH PAH Hexamethylphosphoramide 233 NIH G HEXMTHPHO G Benz Benz Benz Benz Hydrogen Fluoride 19 CDC G HF G Benz Benz Benz Benz Mercury 356 CDC G/P HGSUM G/P PAH PAH PAH PAH 1,2,3,4,5,6-Hexachlorocyclohexane 323 CDC G/P HXCCL123456 G/P PAH PAH PAH PAH Hydrazine 113 CDC G HYDRAZINE G Benz Benz Benz Benz Hydroquinone 285 CDC G/P HYDROQUIN G/P PAH PAH PAH PAH Isophorone 215 CDC G ISOPHORONE G Benz Benz Benz Benz LEAD 1,740 CDC P LEAD P Cr6 Cr6 Ni Ni Maleic anhydride 202 CDC G MALANHYD G Benz Benz Benz Benz Manganese 1,962 CDC P MANGANESE P Cr6 Cr6 Ni Ni 4,4'-Methylenebis(2-Chloraniline) 209 NIH G MB2CLRAN44P G Benz Benz Benz Benz Methylene chloride 39 CDC G MECL G Benz Benz Benz Benz Methanol 64 CDC G METHANOL G Benz Benz Benz Benz Methoxychlor 89 NIH G METHOXYCHL G Benz Benz Benz Benz Methyl bromide 3 CDC G METHYLBROM G Benz Benz Benz Benz Methyl isobutyl ketone 116 CDC G MIBK G Benz Benz Benz Benz Fine Mineral Fibers NA http://www.usg.com/conte P MINFIB P Cr6 Cr6 Ni Ni nt/dam/USG_Marketing_ Communications/united_s tates/product_promotional _materials/finished_asset s/usg-mineral-wool-300a- msds-en-75850002.pdf Methyl methacrylate 101 CDC G MMETACRYLAT G Benz Benz Benz Benz Methyl tert-butyl ether 55 NIH G MTBE G Benz Benz Benz Benz 4,4'-Methylenedianiline 397 CDC G/P MTHYDIAN44P G/P PAH PAH PAH PAH Methyl chloride -98 CDC G MTHYLCHLRD G Benz Benz Benz Benz Methyl Chloroform 74 CDC G MTHYLCHLRF G Benz Benz Benz Benz 4,4'-Methylenediphenyl Diisocyanate 313 CS G/P MTHYLDPNLDS G/P PAH PAH PAH PAH Methylhydrazine 88 CDC G MTHYLHYZIN G Benz Benz Benz Benz Methyl iodide 43 CDC G MTHYLIODIDE G Benz Benz Benz Benz Methyl isocyanate 39 CDC G MTHYLISOCY G Benz Benz Benz Benz Naphthalene 260 CDC G NAPHTH G Benz Benz Benz Benz Nickel 2,913 CDC P NICKEL P Cr6 Cr6 Ni Ni Nickel oxide 1,955 NIH P Nickel refinery dust 2,730 http://www.cdc.gov/niosh/ P docs/81- 123/pdfs/0445.pdf 4-Nitrobiphenyl 340 CDC G/P NITROBIPHL4 G/P PAH PAH PAH PAH Nitrobenzene 211 CDC G NITROBNZNE G Benz Benz Benz Benz 4-Nitrophenol 279 NIH G/P NITROPHENL4 G/P PAH PAH PAH PAH 2-Nitropropane 121 CDC G NITROPROPA2 G Benz Benz Benz Benz N-Nitrosodimethylamine 152 CDC G NNITROSDIM G Benz Benz Benz Benz N-Nitrosomorpholine 224 NIH G NNITROSMPH G Benz Benz Benz Benz N-Nitroso-n-methylurea 164 P CS G NNITROSURE G Benz Benz Benz Benz o-Anisidine 225 CDC G o-Anisidine G Benz Benz Benz Benz o-Toluidine 200 CDC G O-TOLUIDINE G Benz Benz Benz Benz Anthracene 342 NIH G/P PAH_000E0 G/P PAH PAH PAH PAH Phenanthrene 340 NIH G/P AirToxScreen 2018 Documentation E-13  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Pyrene 404 NIH P 3-Methylcholanthrene 178 http://www.speclab.com/c G PAH_101E2 G Benz Benz Benz Benz ompound/c50328.htm 7,12-Dimethylbenz[a]anthracene 122 http://www.sigmaaldrich.c G PAH_114E1 G Benz Benz Benz Benz om/catalog/product/supel co/442425?lang=en&regi on=US Dibenzo[a,h]pyrene 308 E CS G/P PAH_176E2 P Cr6 Cr6 Ni Ni Dibenzo[a,i]pyrene 552 P CS P Dibenzo[a,l]pyrene 552 P CS P 5-Methylchrysene 449 P CS P PAH_176E3 P Cr6 Cr6 Ni Ni 7H-Dibenzo[c,g]carbazole 544 P CS P Benzo[a]pyrene 360 http://www.speclab.com/c G/P ompound/c50328.htm Coal tar >250 http://www.inchem.org/do G/P cuments/icsc/icsc/eics141 5.htm Dibenzo[a,e]pyrene 552 P CS P Methylchrysene 449 P CS P 1-Nitropyrene 445 P CS P PAH_176E4 P Cr6 Cr6 Ni Ni Benz[a]anthracene 438 NIH P Benzo[b]fluoranthene 4,665 P CS P Benzo[j]fluoranthene 480 E CS P Dibenz[a,h]acridine 534 P CS P Dibenzo[a,j]Acridine 534 P CS P Indeno[1,2,3-c,d]pyrene 530 http://www.speclab.com/c P ompound/c193395.htm Benzo[k]fluoranthene 480 http://www.speclab.com/c P PAH_176E5 P Cr6 Cr6 Ni Ni ompound/c207089.htm Carbazole 355 http://www.sigmaaldrich.c G/P om/catalog/product/sigma /c5132?lang=en&region= US Chrysene 448 http://www.speclab.com/c P ompound/c218019.htm Dibenzo[a,h]anthracene 262 http://www.sigmaaldrich.c G/P PAH_192E3 G/P PAH PAH PAH PAH om/catalog/product/supel co/48574?lang=en&regio n=US 12-Methylbenz(a)anthracene 410 P CS P PAH_880E5 G/P PAH PAH PAH PAH 1-Methylnaphthalene 240 NIH G 1-Methylphenanthrene 359 http://www.nature.nps.gov G/P /hazardssafety/toxic/phen 1met.pdf 1-Methylpyrene 372 http://www.chemicalbook. G/P com/ChemicalProductPro perty_EN_CB7421679.ht m 2-Chloronaphthalene 256 http://www.chemicalbook. G com/ChemicalProductPro perty_EN_CB8854627.ht m 2-Methylnaphthalene 241 http://www.speclab.com/c G ompound/c91576.htm 2-Methylphenanthrene 339 CS G/P 9-Methyl Anthracene 196 CS G Acenaphthene 279 NIH G/P Acenaphthylene 265 NIH G/P Benzo(a)fluoranthene 295 NIH G/P AirToxScreen 2018 Documentation E-14  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Benzo(c)phenanthrene 430 P CS P Benzo(g,h,i)fluoranthene 406 P CS P Benzo[e]pyrene 465 P CS P Benzo[g,h,i,]Perylene 550 NIH P Benzofluoranthenes 406 P CS P Extractable Organic Matter (EOM) NA NA NA Fluoranthene 384 NIH G/P Fluorene 295 NIH G/P Methylanthracene 360 P CS G/P Methylbenzopyrene 479 P CS G/P PAH, total 240–400 http://www.epa.gov/reg3h G/P wmd/bf- lr/regional/analytical/semi- volatile.htm Perylene 276 http://www.sigmaaldrich.c G/P om/catalog/product/aldric h/394475?lang=en&regio n=US Parathion 375 CDC G/P PARATHION G/P PAH PAH PAH PAH 2,4,4'-Trichlorobiphenyl (PCB-28) 164 CS G PCB G/P PAH PAH PAH PAH 4,4'-Dichlorobiphenyl (PCB-15) 144 CS G Decachlorobiphenyl (PCB-209) 460 P CS P Heptachlorobiphenyl 415 P CS P Hexachlorobiphenyl 396 P CS G/P Pentachlorobiphenyl 365 E CS G/P Polychlorinated biphenyls 365 E CS G/P Tetrachlorobiphenyl 360 P CS G/P p-Dioxane 101 CDC G P-DIOXANE G Benz Benz Benz Benz Tetrachloroethylene 121 CDC G PERC G Benz Benz Benz Benz Phenol 182 CDC G PHENOL G Benz Benz Benz Benz p-Phenylenediamine 267 CDC G/P PHNYLNDIAMI G/P PAH PAH PAH PAH Phosgene 8 CDC G PHOSGENE G Benz Benz Benz Benz Phosphine 88 CDC G PHOSPHINE G Benz Benz Benz Benz Phosphorus 280 CDC G/P PHOSPHORS G/P PAH PAH PAH PAH Phthalic anhydride 295 CDC G/P PHTHANHYDR G/P PAH PAH PAH PAH Pentachloronitrobenzene 328 NIH G/P PNTCHLNBZ G/P PAH PAH PAH PAH Pentachlorophenol 309 CDC G/P PNTCLPHENOL G/P PAH PAH PAH PAH Propylene dichloride 97 CDC G PROPDICLR G Benz Benz Benz Benz Propionaldehyde 48 NIH G PROPIONAL G Benz Benz Benz Benz 1,3-Propanesultone 180 NIH G PROPNESLT13 G Benz Benz Benz Benz Propoxur D CDC NA PROPOXUR NA EF=1 EF=1 EF=1 EF=1 Propylene oxide 34 CDC G PROPYLENEOX G Benz Benz Benz Benz 1,2-Propylenimine 66 NIH G PROPYLENIMI G Benz Benz Benz Benz Quinoline 238 NIH G QUINOLINE G Benz Benz Benz Benz Quinone S CDC NA QUINONE NA EF=1 EF=1 EF=1 EF=1 Selenium 685 CDC P SELENIUM P Cr6 Cr6 Ni Ni Styrene 145 CDC G STYRENE G Benz Benz Benz Benz Styrene oxide 194 http://www.sigmaaldrich.c G STYROXIDE G Benz Benz Benz Benz om/catalog/product/aldric h/s5006?lang=en&region =US Titanium tetrachloride 136 http://www.sigmaaldrich.c G TITATETRA G Benz Benz Benz Benz om/catalog/product/aldric h/697079?lang=en&regio n=US 2,4-Toluene diisocyanate 251 NIH G TOL_DIIS G Benz Benz Benz Benz AirToxScreen 2018 Documentation E-15  Chem.  AirToxScreen Boiling Point Matching Chemical Modeled in HAPEMd Name  Name Value Sourceb       (ºC)a Phasec Phasec P NP M-OR M-NR Toluene-2,4-Diamine 292 CDC G/P TOLDIAM24 G/P PAH PAH PAH PAH Toluene 111 CDC G TOLUENE G Benz Benz Benz Benz Toxaphene D CDC NA TOXAPHENE NA EF=1 EF=1 EF=1 EF=1 2,4,5-Trichlorophenol 247 NIH G TRCLPHNL245 G Benz Benz Benz Benz 2,4,6-Trichlorophenol 246 NIH G TRCLPHNL246 G Benz Benz Benz Benz 1,2,4-Trichlorobenzene 213 CDC G TRICBZ124 G Benz Benz Benz Benz 1,1,2-Trichloroethane 114 CDC G TRICLA112 G Benz Benz Benz Benz Triethylamine 89 CDC G TRIETHLAMN G Benz Benz Benz Benz Trifluralin 140 http://www.speclab.com/c G TRIFLURALIN G Benz Benz Benz Benz ompound/c1582098.htm 2,2,4-Trimethylpentane 99 NIH G TRMEPN224 G Benz Benz Benz Benz 1,1,2,2-Tetrachloroethane 147 CDC G TTCLE1122 G Benz Benz Benz Benz Vinyl chloride -14 CDC G VINYCHLRI G Benz Benz Benz Benz Vinyl acetate 72 CDC G VINYLACET G Benz Benz Benz Benz Vinyl bromide 16 CDC G VINYLBROM G Benz Benz Benz Benz Vinylidene chloride 32 CDC G VINYLIDCLOR G Benz Benz Benz Benz m-Xylene 139 CDC G XYLENES G Benz Benz Benz Benz o-Xylene 144 CDC G p-Xylene 138 CDC G Xylenes (Mixed Isomers) 139 NIH G Note: Information under the “Chem” header indicates chemicals for which EPA developed emissions data; we identified boiling points and HAPEM phase designations for these chemicals. When EPA ultimately conducted air-concentration modeling, for use then in exposure and risk modeling, they grouped some chemicals together as indicated in the information under the “AirToxScreen” header (note the occasional many-to-one relationship between the “Name” column under “Chem” and the “Name” column under “AirToxScreen”). Assigning phase designations to groups of chemicals required professional judgment, as discussed in Section 2.2. a D=decomposes; E=experimental; NA=not available; P=predicted; S=sublimes; V=varies depending on compound. b CDC=http://www.cdc.gov/niosh/npg; CS=http://www.chemspider.com; NIH=http://pubchem.ncbi.nlm.nih.gov/. c G=gaseous; G/P=gaseous or particulate depending on conditions; P=particulate; NA=unknown. d Source Types: P =point; NP=non-point; M-OR=mobile on-road; M-NR=mobile non-road. Matching Chemicals: Benz=benzene; Buta=1,3-butadiene; Coke=Coke Oven Emissions; PAH=PAH, total; Ni=Nickel; Cr6=Chromium (VI); DPM = diesel particulate matter; “EF=1” means no surrogate assignment, so the exposure factor is set to 1 (before any truncations are then applied), meaning all exposure concentrations are set equal to ambient outdoor concentrations. Table E-2. Boiling-point Definitions Used to Classify HAPs for HAPEM7 Modeling for the 2018 AirToxScreen Boiling-point HAPEM7 HAP Phase Range (°C) G (Gaseous) < 240-260 G/P (Either gaseous or particulate depending on conditions) 240-260 to 400-480 P (Particulate) > 400-480 Source: Adapted from the “Classification of Inorganic Pollutants” table at EPA’s Volatile Organic Compound web page (available as of 2 February 2023 at https://www.epa.gov/indoor-air-quality-iaq/technical-overview-volatile-organic- compounds), as adapted from: World Health Organization, 1989. “Indoor air quality: organic pollutants.” Report on a WHO Meeting, Berlin, 23-27 August 1987. Euro Reports and Studies 111. Copenhagen, World Health Organization Regional Office for Europe. AirToxScreen 2018 Documentation E-16 Table E-3. HAPs Modeled in HAPEM7 for AirToxScreen HAPEM7 Modeled in HAPEM7b AirToxScreen HAP HAP Phasea P NP M-OR M-NR Benzene G     1,3-butadiene G     Coke Oven Emissions G/P  PAH, Total G/P     (polycyclic aromatic hydrocarbons; aggregate mass of unspecified congeners) Chromium (VI) P   (compounds of hexavalent chromium) DPM P    (diesel particulate matter) Nickel P   Note: For chromium (VI), DPM, and nickel, EPA modeled air concentrations corresponding to emissions from all four source types, but per EPA direction we only generated exposure concentrations for the subset of source types shown in the table. See discussion at the bottom of Section 2.3. a G=gaseous; G/P=gaseous or particulate depending on conditions; P=particulate. b P =point; NP=non-point; M-OR=mobile on-road; M-NR=mobile non-road. Table E-4. Truncations Applied to Exposure Factors for AirToxScreen Source Type Total (Aggregate of all HAP Point Non-point Mobile On-road Mobile Non-road Source Types) Benzene 1.00 0.89 1.09 0.89 0.97 1,3-butadiene 0.99 0.92 1.18 0.92 1.03 Coke oven emissions 0.87 NA NA NA 0.87 PAH, total 0.70 0.66 0.83 0.66 0.72 Chromium (VI) 0.60 0.94 NA NA 0.45 Diesel PM NA 1.27 0.60 0.48 0.53 Nickel NA NA 0.56 0.48 0.52 AirToxScreen 2018 Documentation E-17 United States Office of Air Quality Planning and Standards Publication No. EPA-452/B-22-002 Environmental Protection Health and Environmental Impacts Division August 2022 Agency Air Quality Assessment Division Research Triangle Park, NC (pg. 25, a-2); https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=88282&Lab=NERL;
https://www.epa.gov/sites/default/files/2020-07/documents/nata_2014_summary_of_results.pdf2014 NATA Summary of Results Overview On August 22, 2018, the U.S. Environmental Protection Agency (EPA) released the sixth National Air Toxics Assessment (NATA). NATA provides information on potential health risks from breathing air toxics, also known as hazardous air pollutants. NATA is a screening tool, used to help EPA and state, local and tribal air agencies find out if areas, pollutants or types of pollution sources need to be looked at further to better understand risks to public health. NATA provides broad estimates of the risk of developing cancer and serious noncancer health effects over census tracts across the country. It does not estimate any person’s individual risk. This NATA is based on emissions of air toxics for the calendar year 2014. It includes exposure estimates for 180 air toxics that EPA regulates under the Clean Air Act, calculating cancer risk and noncancer health effect estimates for about 140 of these. It also estimates exposure and noncancer health effects for diesel particulate matter. For more details on NATA, please review the 2014 NATA Fact Sheet. You can also learn more at the NATA website, online at https://www.epa.gov/nata. National Results The 2014 NATA estimates that the nationwide average cancer risk from breathing air toxics is 30 in 1 million. This means that, on average, about 30 people per 1 million would develop cancer if they breathe air with 2014 levels of air toxics over a lifetime of 70 years. This risk would be in addition to the cancer risk a person would have without being exposed to air toxics. This is a national average – some areas of the country have lower risks, and some have higher risks. About half of that nationwide average risk comes from formaldehyde. This air toxic comes mostly from other pollutants that chemically react in the air. This is known as secondary formation. Emissions from industries, mobile sources, and natural sources combine in this way. The rest of the nationwide cancer risk comes mostly from pollution that is directly emitted to the air. We’ll go into more detail on how we determine which pollutants drive health risks later in this summary. The map below shows NATA’s estimate of cancer risks from air toxics across the country. Lighter colors indicate lower estimated risks. Darker colors show areas where NATA estimates relatively higher risks. 1 U.S. cancer risks from air toxics as estimated by the 2014 NATA. Local Results While the national results above help EPA describe general trends from air toxics, NATA can also help us identify local areas we may need to look at closer to better understand risks. The 2014 NATA does show some local areas that may have elevated risks of cancer from air toxics. NATA presents results at a local geographic level called a census tract. Census tracts are small subdivisions of a county or parish that represent about 4,000 people. NATA estimates that a few census tracts have a cancer risk from air toxics of 100 in 1 million or higher. Less than 1 percent of all census tracts in the country show these levels. Industrial emissions of three pollutants – ethylene oxide, chloroprene and coke oven emissions – contribute to most of the risk in these tracts. The NATA Fact Sheet and a special EPA ethylene oxide website have more information on local risks. These also describe how best to interpret NATA results at these local levels. 2 To quickly find out what cancer risk and noncancer health effects the 2014 NATA estimates for your area, you can use the NATA Map app. The map app’s search tool lets you “zoom” to places of interest anywhere in the country. You can click on the map to display risks, emissions and other NATA data and results. Identifying NATA’s Risk Drivers EPA uses several measures to determine which pollutants may contribute most to long-term cancer risks and noncancer health hazards, both nationally and regionally (for example, in certain metropolitan areas). This information helps EPA and state, local and tribal air agencies to better understand which pollutants and sources of pollution that we may need to look at more closely. For example, if a certain important air toxic comes mostly from vehicles, this may mean we need to look closer at how to control these vehicle emissions. To do this, we define “drivers” of risk – pollutants that contribute most to risks and hazards – and “contributors” – pollutants that may also have an impact but are generally less important at these scales, nationally and regionally, than drivers. To classify risk drivers and contributors, we use the criteria shown in the following table. 2014 NATA Health Effects Drivers and Contributors Risk Characterization Category Cancer Risk Hazard Number of People (or Exceeds Index2 Greater) Exposed (in 1 million)1 (in millions) National Cancer Driver 10 25 Regional Cancer Driver 10 1 Regional Cancer Driver 100 0.01 National Cancer Contributor 1 25 Regional Cancer Contributor 1 1 National Noncancer Driver > 1 25 Regional Noncancer Driver > 1 0.01 1 Cancer risks are upper-bound lifetime cancer risks (i.e., a plausible upper limit to the true probability that a person will contract cancer over a 70-year lifetime from a given hazard (such as exposure to a toxic chemical). 2 A hazard index of 1 or lower means adverse noncancer effects are unlikely. You can read more about hazard indexes and how they are calculated on the NATA website. 3 Using these guidelines, we can list the air toxics that show up as drivers and contributors to cancer risks and noncancer health hazards in the 2014 NATA: • National cancer risk driver: formaldehyde • Regional cancer risk drivers: ethylene oxide, chloroprene • National cancer risk contributors: 1,3-butadiene, acetaldehyde, benzene, carbon tetrachloride, naphthalene • Regional cancer risk contributors: 1,4-dichlorobenzene, arsenic compounds, chromium VI compounds, coke oven emissions, ethylbenzene • National noncancer hazard drivers: None • Regional noncancer hazard drivers: chlorine, hexamethylene diisocyanate To learn about the health effects of these air toxics, please visit EPA’s Health Effects Notebook for Hazardous Air Pollutants. This website presents fact sheets for these and all other hazardous air pollutants specified in the Clean Air Act. How EPA Uses NATA Results As noted above, we designed NATA as a screening tool. This means that EPA does not use NATA results as the only source of information on whether any particular place has – or does not have – unacceptable risks from air toxics. Instead, we intend NATA results to be a place to start looking at possible risks from air toxics. NATA helps EPA and our partner air agencies at the state, local and tribal levels identify which pollutants, emissions sources and places we may need to study in more detail. What NATA tells us about national and regional pollutant drivers and contributors helps in this focus. We look at ways we can broadly reduce emissions of national-level drivers and contributors. Regional drivers may help call attention to more localized air issues. How You and Your Community Can Use NATA Results NATA can help you learn which air toxics and sources of these emissions may be of concern to you and your community. You can also use NATA results to open a dialogue about air toxics in your community. You may wish to contact your state, local or tribal air agency with any questions or concerns you have about NATA results. These agencies usually set and enforce local air toxics rules. Often, they also carry out any localized air quality studies needed to get a more precise picture of the air quality in an area. Depending on the source, location and pollutant of interest, these local assessments may include updated emissions information, local weather data, actual measurements of air toxics, and other data that’s more detailed than NATA’s broad, national scale allows us to include. 4;
https://www.epa.gov/AirToxScreen/airtoxscreen-assessment-methods.