1. 1 Summary of National Air Toxics Data Analyses - 2007 Hilary Hafner Mike McCarthy Sonoma Technology, Inc. Petaluma, CA Presented to webinar audience November 6, 2008 Excerpted from: Air Toxics Data Analysis Workshop Rosemont, IL October 4, 2007 908302.05-3485 905306-3236 908304

2. 2 Overview History of national air toxics data analysis Phase V analysis objectives Risk characterization Trends and accountability Data preparation Data factoids

3. 3 Background on National Data Analysis Phases I and II in 2000-2002 - Answered questions related to the design of the national air toxics monitoring network Bortnick and Stetzer, Sources of variability in ambient air toxics monitoring data, 2002, Atmos. Enviro. Bortnick and Stetzer, Sampling frequency guidance for ambient air toxics monitoring, 2002, JAWMA Phase III in 2004 – Addressed data quality, utility, and applicability for answering policy-relevant questions Kenski et al., Lessons learned from air toxics data: a national perspective, 2005, Enviro. Manag. McCarthy et al., Background concentrations of 18 air toxics for North America, 2006, JAWMA Phase IV in 2005-2006 – Assessed temporal and spatial variability in air toxics concentrations Touma et al., Spatial and temporal variability of ambient air toxics data, 2006, JAWMA McCarthy et al., Temporal variability of selected air toxics in the United States, 2007, Atmos. Enviro.

4. 4 Phase V Analysis Objectives The Phase V analysis focused on two broad topic areas Characterize concentrations using risk/hazard- weighting and categorical risk/hazard screening to better understand the relative importance of various pollutants to human health Assess trends in air toxics concentrations and develop methods to establish links with specific control measures for accountability

5. 5 Air Toxics Risk and Hazard Characterization Analyses Performed chronic risk and hazard screening Created national risk- and hazard-estimate distributions Created risk- and hazard-estimate maps Ranked our confidence in pollutants above risk levels Performed acute hazard screening Assessed spatial variability as it pertains to risk estimates Within-city variability NATTS urban-rural comparisons Cumulative risk estimates by site and city Hot and cold spot analysis for benzene and arsenic PM 2.5 Comparison with ozone and PM 2.5 nonattainment areas

6. 6 Air Toxics Trends and Accountability Analyses Investigated site level trends for all air toxics Three specific trend periods – 1990-2006, 1995-2006, 2000-2006 Longest trend record of 5+ years Prepared national distributions in trends for all air toxics with at least 10 trend sites in one of the three trend periods Identified MDL issues associated with national PM 2.5 metals trends Developed and applied two methods for identifying and associating control measures with changing ambient concentrations Top-down method – start with the ambient data and look for associated control measures; used to provide strong evidence on the efficacy of mobile source controls on air toxics concentrations. Bottom-up method – start with the control measure and look for changes in ambient concentrations to provide evidence of its efficacy; used to try to identify decreases in metals concentrations as a result of a MACT control on hazardous waste incinerators and cement kilns. Extrapolated current rates of change for key risk-driving toxics to project when they might go below the ambient concentration corresponding to a 1-in-a-million cancer risk level Reanalyzed and classified diurnal and seasonal patterns in air toxics concentrations

7. 7 Data Acquisition Acquired and processed data from 1990 through mid-2007  EPA’s Air Quality System (AQS) AMP501 data from 1995 through 2006 available in September 2006 –Updated with AMP501 data from 2000 through 2007 available in July 2007, which was used to update the trends and accountability analyses Archived AMP501 data from 1990 through 1994 requested directly from EPA in October 2005  IMPROVE speciated PM 2.5 data downloaded from VIEWS web site, in September 2006  SEARCH speciated PM 2.5 data downloaded from Atmospheric Research Analysis web site, in September 2006  The Legacy Air Toxics Archive from the Phase III national air toxics analysis project (Hafner and McCarthy, 2004)  Data acquired from local and state air quality agencies as part of the SESARM toxics data analysis project IMPROVE = Interagency Monitoring of Protected Visual Environments VIEWS = Visibility Information Exchange Web System SEARCH = SouthEastern Aerosol Research and Characterization Study SESARM = SouthEastern States Air Resources Managers

8. 8 Database As part of several projects, an air quality archive (AQA) was developed as an analysis-ready database that includes data from AQS (1990-2006), IMPROVE and SEARCH data, and data from the legacy air toxics archive. This database contains nearly 1 billion raw data records, 27 million raw air toxics records, and complete validated and temporally aggregated data sets. An excerpt of the data is available here: http://www.epa.gov/ttn/amtic/toxdat.html

9. 9 Data Preparation 24-hr duration samples are the most common for air toxics measurements. 24-hr averages were aggregated to quarterly, annual, and site averages. For the rare sites with subdaily duration samples (1-hr or 3-hr duration samples), daily averages were created using a 75% diurnal completeness criterion Quarterly averages were created from daily averages –75% completeness for expected sampling frequency (e.g., 12 out of 15 samples for 1-in-6-day sampling frequency) –Minimum of six samples required (1-in-12-day sampling frequency) Annual averages were created from quarterly averages – 3 out of 4 quarterly averages were required (75% completeness) When concentrations were reported at or below the method detection limit (MDL), quarterly and annual averages were created using MDL/2 substituted values. –For sites and pollutants with fewer than 15% of records reported above MDL, annual averages were treated as less certain. Annual mean concentrations for these sites and pollutants are likely below MDL, but quantification is not feasible.

10. Data Availability by Year * Data available in July 2007 may not contain full annual averages for data collected in 2006 (i.e., late reporting to AQS appears to be common). *

11. 2003-2005 Data Quality and Quantity 85 % below MDL 100 sites

12. 12 How Representative Is the Current Monitoring Network of the United States? Air toxics are primarily measured in urban counties:  Urban areas are where most people live and where most air toxics are emitted (VOCs are representative of top 10% of populated counties; metals are representative of top 20%).  The current monitoring network is not representative of the rural U.S. VOC = volatile organic compound

13. Toxics Monitoring Sites 2003-2005

14. Google Earth file: http://www.epa.gov/airexplorer/monitor_kml.htm

15. 15 Summary of Risk and Hazard Characterization This summary provides an overview of an assessment of the importance of species; analysis performed using risk- or hazard- estimates of air toxics data collected from 1990-2005, focusing on the most recent three years of data (2003-2005). Data preparation and other method details are minimally described here in order to focus on the results and implications.

16. 16 Scientific Questions  Which air toxics have the highest risk- and hazard-estimates nationally?  Which toxics are possibly problems but are not measured well enough to be quantified?  Which toxics that are not measured are identified as problems by models and emissions inventories?  How do spatial variations of concentrations change relative to levels of concern?  Do concentrations vary across risk levels (e.g., 1-in-a-million, 10-in- a-million cancer risk)?  Which areas appear to have higher cumulative risk estimates based on commonly measured pollutants?  Do concentrations vary substantially within cities relative to chronic cancer risk or noncancer hazard levels?  Do cumulative risk estimates drop off in rural and remote areas? Is the drop-off large enough to reduce risk below levels of concern? Which pollutants contribute the most to risk in these areas?  Are ozone, PM, and toxics cumulative risk estimates high in the same areas?

17. 17 Data Preparation For most of the analyses described here, toxics concentrations were aggregated to include data from 2003-2005; these three years had the largest number of valid annual averages. Site averages are the mean of annual mean concentrations from 2003-2005 (1-3 years of data). These averages were used to minimize meteorological and data reporting artifacts in the data set. Sites and pollutants with < 15% of records reported above MDL were treated as less certain. Annual mean concentrations for these sites and pollutants are likely below MDL, but quantification is not feasible. These sites are usually included in the analyses shown, but are colored or marked to indicate that the values shown are only known to be less than the MDL value. For all risk- and hazard-estimates, EPA OAQPS chronic and acute dose-response values were used. These values are available at http://www.epa.gov/ttn/atw/toxsource/table1.pdf

18. 18 What is the National Picture of Ambient Risk Estimates? The following slide provides a summary of the range of risk estimates for air toxics with cancer risk organized by highest risk to lowest – first for toxics with >15% of data above detection nationally and then for toxics with most data below detection (>85%). Risk estimates are computed by multiplying ambient concentrations by unit risk estimates (URE). For example, a benzene ambient concentration of 2 μg/m 3 and a URE of 7.8 x 10 -6 (μg/m 3 ) -1 provides an estimate of risk of 1.6 x 10 -5 or “16 in a million”. These slides aim to answer the following questions:  Are concentrations above the 1-in-a-million cancer risk levels?  Are concentrations characterized well enough to assess health risks? Note: The 2005 version of EPA OAQPS UREs were used (not July 2007 version).

19. 19 Risk Estimates (per million)

20. 20 Findings Category A – pollutants with a majority of sites with risk estimates above the one-in-a- million risk level: ethylene oxide, acrylonitrile, carbon tetrachloride, benzene, arsenic, 1,3 butadiene, acetaldehyde, 1,4-dichlorobenzene, tetrachloroethylene, naphthalene, and the larger particulate size fractions of nickel. Some pollutants were monitored at 100s of locations while others are monitored at few sites. Pollutants measured at fewer monitoring locations may be a poorer representation of the national distribution. Many pollutants in category A are above levels of concern at all locations. Category B – pollutants with most of the data below MDL, but detection limits above the one-in-a-million risk level. These pollutants potentially have concentrations above the one-in-a-million level: ethylene dibromide, cadmium, 1,1,2,2-tetrachloroethane, benzyl chloride, hexachlorobutadiene, ethylene dichloride, 1,1,2 trichloroethane, and 1,2-dichloropropane. Concentration distributions reflect typical MDL/2 values, rather than actual ambient concentrations. The concentration distributions shown are likely to be upper-limit estimates of risk for these pollutants. Category C – pollutants with the majority of monitoring sites reporting concentrations below the one-in-a-million risk level including those usually above MDL, such as chromium VI, dichloromethane, and formaldehyde, and those usually below the MDL, such as vinyl chloride and trichloroethylene. While the actual concentrations may be poorly quantified because ambient concentrations are below MDL, it is sufficient to say that concentrations are below the one-in-a-million level.

21. 21 Risk Estimates: Spatial Distribution The following slides show maps of the site-level data included in the previous summaries of national concentration and risk estimate ranges. These maps enable assessments of the spatial representativeness of the toxics, show where concentrations or risk estimates are highest and lowest, and show where concentrations are typically below detection.

22. 22 Maps of Risk Estimates: Acrylonitrile Acrylonitrile’s reliable risk estimates are above 1-in-a-million, and often above 10- in-a-million risk.  Method detection limits (MDLs) are insufficient to quantify risk at many sites.  Emissions sources include the petrochemical industry (emitted as an intermediate product) and cigarette smoke. Questionable

23. 23 Maps of Risk Estimates: Benzene The highest concentrations are often not associated with mobile sources. Benzene is above 1-in-a-million risk everywhere, and often above 10-in-a-million risk. Questionable

24. 24 Maps of Risk Estimates: Carbon Tetrachloride Carbon tetrachloride’s risk-weighted concentrations should be 9-in-a-million.  Carbon tetrachloride concentrations should be the same everywhere (e.g., McCarthy et al., 2006) with almost no U.S. exceptions; this air toxic provides an excellent monitoring QC check.  Sites in gray have detection limit issues.  Sites in red have concentrations that are too high. Some sites in orange have concentrations that are too low. Questionable

25. 25 Maps of Risk Estimates: Arsenic PM 2.5 Note that STN and IMPROVE sites are not directly comparable due to an MDL issue. IMPROVE MDLs will be increasing by at least a factor of five for most toxics metals (Hyslop and White, 2007). Arsenic PM 2.5 is typically above the 1-in-a-million risk level throughout the eastern United States. Questionable

26. 26 Categorical Risk Screening: Approach Is 85% of data for this site- pollutant below MDL? Yes No Is level of concern above MDL? Yes No Pollutant concentration is below level of concern Site-pollutant is uncertain Is site-average concentration above level of concern? Yes No Pollutant concentration is below level of concern Pollutant concentration is above level of concern

27. 27 Pollutants Above or Possibly Above 10 -5 Risk levels at >20% of Sites Pollutant Number of sites with concentrations above 10 -5 cancer risk Percent of sites with concentrations above 10 -5 cancer risk Number of uncertain sites (85% of data below MDL and 10 -6 cancer risk below MDL) Percent of sites above or possibly above 10 -5 cancer risk Benzene113370 Acrylonitrile43354168 Ethylene Oxide13813100 Carbon Tetrachloride29102720 Benzyl Chloride985659 1,1,2,2-Tetrachloroethane12516980 Hexachlorobutadiene645942 Arsenic TSP454965 1,4-Dichlorobenzene633520 Ethylene Dibromide63227100 Arsenic PM 10 251238 Ethylene Dichloride425423 Cadmium PM 2.5 1018672 Counts of sites with <10 -5 risk are not explicitly shown in this table.

28. 28 Pollutants Above or Possibly Above 10 -6 Risk at >50% of Sites Counts of sites with <10 -6 risk are not explicitly shown in this table.

29. 29 Quantifying Confidence in Toxics of Concern A weighted method was created to quantify the level of confidence that a specific pollutant is usually measured at or above a level of concern. Using the 10 -6 risk screening results, the values above 10 -6 are weighted as 1, the values below 10 -6 as -1, and uncertain values as 0. The results are summed across sites. Higher magnitudes (positive or negative) indicate higher confidence levels.

30. 30 Determining which Pollutants are Most Likely above the 10 -6 Level Nationally

31. 31 Confidence that National Risk-weighted Concentrations are >10 -6

32. 32 Noncancer Hazard The same basic analyses as shown for the cancer risk were performed for chronic hazard using OAQPS reference concentrations. An additional acute and subchronic hazard screening was performed on daily and subdaily measurements. Minimum risk levels (MRLs) were used for daily screening values.

33. 33 Pollutants Above or Possibly Above 1.0 Hazard Estimates at >1% of Sites Pollutant Number of sites Number of sites above reference concentration Percent of sites above reference concentration Uncertain sites (85% of data below MDL and reference concentration below MDL) Percent of sites above or possibly above reference concentration Acrolein53417712100 Manganese TSP9688615 Manganese PM 10 261404 Acetonitrile632303 Formaldehyde1614202 Acrylonitrile1243202 1,3-Butadiene2764101 Nickel TSP1011167 Counts of sites with <1.0 hazard quotient are not explicitly shown in this table.

34. 34 Using new canister methods, acrolein concentrations are above the MRL for all regions in which it has been measured. Samples collected in canisters from 10/2005 to 9/2006 grouped by EPA region. Note Region 9 measurements were reported using a different method code and were not included. Figure taken from: Wade et al., Analysis of ambient air quality after Hurricane Katrina, STI-3159, 2007 More than half of the daily samples were above the MRL screening level for most regions.

35. 35 National Toxics of Importance: Summary Chronic cancer risk estimates above 10 -6 nationally Chronic noncancer hazard estimates above 1.0 nationally

36. 36 Comparison with NATA 1999 results Pollutants identified as contributors to chronic risk or hazard in NATA 1999 that are not measured directly or reported to AQS:  Particulate organic matter (usually reported individually, rather than as a sum)  Diesel Particulate Matter (not directly measured; surrogate measurements such as BC are available)  Coke oven emissions  Quinoline  Triethylamine  Hydrazine  Maleic Anhydride

37. 37 Trends: Approach and Methods Create trends for (a) three trend periods: 1990-2006, 1995-2006, and 2000-2006; (b) longest trend possible at each site 75% completeness for trend period required Data from 1990 or 1991 required for 1990-2006 trend period Trends were created at the site level Trends required consistent site, parameter, and method codes over time (POC can float between years) Individual trends were plotted along with MDL values and standard deviation in annual average Linear regressions were fitted to the trends F-test was performed to identify if trend was statistically significant (i.e., non-zero) at 95% confidence level

38. 38 Example Individual Site Trend A tool was developed to produce site level trend statistics and graphics by pollutant. Visual inspection is key to properly classifying the trends. Benzene concentrations at site 245100040 in Baltimore, Maryland Error bars are the standard deviation of the annual average -4.7% decrease per year >95% statistical significance

39. Counts of Trend Sites by Pollutant All other pollutants had fewer than 20 trend sites available in these three trend periods.

40. 40 National Picture of Trends Data were aggregated at the national level. Sites with more than 85% of data below MDL were not included in aggregation (because trends at these sites are unlikely to be identified using simple methods). The distribution of trends nationally was plotted for each pollutant with at least 15 monitoring sites with trends in one of the three trend periods.

41. 41 National distribution of the trends in VOCs for any site with 5+ years of monitoring data It is also possible to plot the distribution in trends for the longest 5+ year trend period at all sites in the United States. While the trend periods are not consistent across sites, this way of inspecting the data captures a larger number of sites where data were collected that did not meet the criteria for the other trend periods. This data set is entirely consistent with the results from the other time periods.

42. 42 National distribution of the trends in chlorinated VOCs for three trend periods At the sites where chlorinated VOCs are measured reliably, the median sites show decreasing concentrations. However, it is important to recognize the sparseness of the data set as a result of sites being excluded from the analysis. Most of these pollutants had a large fraction of their sites excluded because more than 85% of data were below MDL.

43. 43 National distribution of the trends in TSP metals for any 5+ year trend period Adding the 5+ year trend period greatly enhances the number of monitoring sites available for assessment. If we weight the data towards the 5+ year trend period, the Nickel and Cadmium decreasing trends are relatively convincing. Manganese and chromium trends remain balanced between increasing and decreasing trends, with a slight bias towards increasing values.

44. 44 National distribution of the trends in PM 2.5 metals for 2000-2006 and any 5+ year trend period Additional investigation of trends of these pollutants shows they track MDL even when >15% of data is above MDL. Therefore, these trends are not reliable.

45. 45 PM 2.5 Arsenic Concentrations and MDLs Even though the concentrations are above MDLs, they still track MDL over time. Trends are therefore unreliable. 1 2 3

46. 46 Other PM 2.5 Metal Examples ChromiumLead Nickel All have concentrations that track MDLs in suspicious ways. While these trends may not necessarily result from MDLs, we are skeptical of using PM 2.5 metals data for trends without additional assessment.

47. 47 Summary of National Trends Hydrocarbon concentrations are decreasing at consistent rates of 4 to 6% per year nationally over multiple trend periods. The distribution of trends for carbonyl compounds is centered at zero; approximately equal numbers of sites have increasing or decreasing trends Chlorinated VOCs are declining where measured reliably; however, many of these compounds are not measured reliably (i.e., data are usually below MDL). Lead TSP concentrations are declining nationally over the earlier trend periods. Other metal TSP concentrations have too few trend sites to draw national conclusions. PM 2.5 metal concentration trends are suspect.

48. 48 Accountability Analysis Identifying and characterizing trends in air toxics does not specifically explain which, if any, control measures are contributing to those trends. We have developed methods to identify specific control measures using data at either the national or local scale. Top-down approach: Use existing trends to identify control measures. Bottom-up approach: Identify known control measures and determine if trends in pollutants meet expectations.

49. 49 Top-down Accountability Approach Hypothesis: If pollutants are emitted by the same source, emissions should covary over long-time scales. In other words, trends should be parallel if normalized. Identify covariant trends in Mobile Source Air Toxics (MSATs) as an indicator of sites dominated by mobile source emissions. Characterize MSAT trend “signature.” Screen sites with trends >5 years to identify : –Mobile source emission-dominated sites and signature –Sites where other emissions sources may be important Identify spatial differences in trends, if any. Provide evidence that mobile source controls are indeed reducing concentrations of air toxics.

50. 50 Top-down Accountability Method Identify sites with long-term (6+ years) records of measurements of selected MSATs (primary emissions only): Benzene, toluene, 1,3-butadiene, xylenes, and ethylbenzene Carbon tetrachloride as internal tracer (non-MSAT) Require only the site and parameter to be consistent over the trend period (method and POC can float between years). Normalize annual average concentrations at each site using maximum concentration over the trend period for each pollutant (i.e., divide each year by highest value measured). Plot trends and visually screen them for covarying linear trends.

51. 51 Top-down Approach Example: Site 060371002, Burbank, CA Maximum concentration is always 1 Slopes of the MSATs regressions are all virtually parallel. Carbon tetrachloride has a very different slope, as expected from a pollutant with a different emissions source.

52. 52 Other Types of Emissions Signatures Four major categories of trend signatures were observed:  Mobile source signature (shown in previous slides, yellow on following map)  Mobile source signature with shallow or increasing 1,3-butadiene  Mobile source signature with shallow or increasing benzene and/or 1,3-butadiene  Other noncovariant signatures (usually have at least one increasing trend)

53. 53 Spatial Characterization of Trend Profile “Signatures” Visual inspection of the slopes of trends provides useful information on the covariance of pollutant concentrations over time. The percentage change in concentrations per year was plotted on maps for each of the pollutants shown in the scatter plots to spatially investigate the trends profiles. Mobile source signatures have MSAT profiles of similar magnitudes; other signatures have increasing or varying magnitudes among the pollutants.

54. 54 Identifying Spatial Patterns in Trend Signatures

55. 55 Accountability Summary (1 of 2) Most sites in the United States conform to our expected mobile source trend profile signature. California and the Northeast had the largest number of these sites Two mobile source-like signatures accounted for most of the rest of the sites 1,3-butadiene signature sites had shallow or increasing 1,3-butadiene (possible measurement issues?). These were concentrated in Minnesota and New York. Benzene signature sites had shallow or increasing benzene (likely explained by nearby point-source emissions for some sites; others are not clear). These were concentrated in Texas and Indiana. Some sites had increasing trends or noncovariant trends in multiple MSATs. These sites may have nearby emissions sources that are influencing trends. These sites may be good candidates for case-study analyses of other emissions sources. These sites were concentrated in Texas. The top-down approach may be applicable to other pollutants from mobile sources (CO, NO x, black carbon) or other emissions sources with multiple co- emitted pollutants.

56. 56 Accountability Summary (2 of 2) In general, MSATs showed a 4 to 6% decrease per year in spite of increases in vehicle miles traveled (VMT) of 3 to 4% per year. Harley et al., (2006) has shown that a 4% per year decline in benzene, for example, was attributable to fleet turnover.