Trends in air toxics cancer risk in Southern California, 1998-2018

Air toxics are an important category of air pollutants that are known to cause adverse health effects, including increased cancer risk. Regulatory efforts at federal, state, and local levels have aimed to decrease air toxics emissions over the past several decades. This study evaluated trends in air toxics cancer risks in Southern California using data from 1998 to 2018. We estimated air toxics cancer risk for each of four iterations of the South Coast Air Quality Management District’s Multiple Air Toxics Exposure Study, which included at least one year of measurements at 10 stations and air toxics modeling for each iteration. Cancer risks were calculated using the measured and modeled air toxics concentrations averaged over a one to two year period and multiplied by the corresponding cancer potency factor and combined exposure factor that accounted for multiple exposure pathways and children’s increased sensitivity to the health effects of air pollution. We examined temporal trends in overall air toxics cancer risks and evaluated changes in the air toxics species that contributed most to cancer risk in the region. Both measurement and modeling results show that air toxics cancer risk in Southern California decreased by more than 80% between 1998 and 2018, including a decrease of about 50% from 2012 to 2018. Diesel particulate matter was the main risk driver, followed by benzene, 1,3-butadiene, and formaldehyde. We found that more densely populated communities showed larger decreases than sparsely populated areas. The substantial decrease in air toxics levels over this 20-year period points to the success of air pollution policies aimed at addressing air toxics emissions and can inform future policy efforts to further reduce air toxics health impacts.


Introduction
Under the federal Clean Air Act, the U.S. Environmental Protection Agency (U.S. EPA) has identified 188 substances as hazardous air pollutants (HAPs), which are defined as pollutants that are known or suspected to cause cancer or other serious health effects [1].Similarly, the state of California identifies toxic air contaminants, which include the federal HAPs and additional air pollutants that have been found to cause or contribute to negative health effects in humans [2].These lists of air toxics designations are updated over time as additional scientific information becomes available.Unlike the criteria air pollutants for which national and state ambient air quality standards have been established, the regulatory structure for air toxics focuses on facility-specific emission limits rather than ambient concentration standards.Air toxics concentrations depend heavily on proximity to sources and meteorological factors, which results in steeper spatial gradients compared to criteria air pollutants, making ambient concentration standards more difficult to apply.Therefore, an evaluation of the health impacts of air toxics in a densely populated region would ideally include detailed measurement and modeling information to ascertain long-term health impacts of pollution from air toxics.U.S. national trends for HAPs have been examined in previous studies [3][4][5][6][7][8], including one recent publication that estimated air toxics cancer risk at seven urban sites in the U.S. to be above 75 in one million and reported there was little change in most pollutant concentrations between 2013 to 2017 [7].However, this study relied on monitoring data from 27 stations spread across 20 states and Washington, DC to estimate exposures, so the exposure estimates contain substantial uncertainty among populations living far away from these monitoring sites.Other studies have used the U.S. EPA National-Scale Air Toxics Assessment (NATA) study, which uses the National emissions inventory and modeling techniques to better capture spatial variations in air toxics levels [5,6,8].While this modeling approach provides the spatial coverage, some comparison studies have evaluated the agreement between NATA and monitoring data and found substantial underestimation of several carcinogenic air pollutants [3,7].A study that uses both modeling and monitoring methods would help leverage the strengths of each of these approaches.
The Multiple Air Toxics Exposure Study (MATES) is a recurring study that evaluates the cancer risks associated with ambient air toxics in the greater Los Angeles region in California, using both modeling and measurement approaches, including a detailed emissions inventory [9].The study is conducted in the South Coast Air Quality Management District (South Coast AQMD) jurisdiction, which includes 17 million residents across an approximately 10 000 square mile area.The MATES analysis is repeated every several years to provide an updated look at the state of air toxics in the region and allows for an examination of long-term trends.
Previous MATES programs have identified diesel particulate matter (diesel PM) as the main risk driver for air toxics cancer risk, with benzene, 1,3-butadiene, and formaldehyde among the next-highest contributors [10][11][12].Given the multitude of industrial, commercial, and residential air pollution sources, and the large population in the region, tracking air toxics trends in the greater Los Angeles area serves as an important case study for other metropolitan areas.Local and state agencies have implemented policies over the past several decades to reduce air toxics emissions, which have resulted in overall decreased ambient concentrations and reduced air toxics cancer risks, and the MATES programs help quantify these reductions.
Here, we examine the long-term trends in air toxics cancer risks in the greater Los Angeles area from 1998 through 2018, using both air toxics modeling and measurement methodologies.This approach also allows for a more detailed examination of the trends in the key risk drivers in the region.An examination of temporal trends comparing the air toxics cancer risks in the most densely populated areas with the most sparsely populated areas provides quantitative information about the relationship between population density and air toxics levels, and changes in this relationship over time.Together, these data can help quantify the overall impacts of the air toxics emission reduction programs in the Southern California region.

Methods
This analysis includes data from four iterations of MATES, including air toxics measurements and modeling results from April 1998-March 1999 (MATES II), April 2004-March 2006 (MATES III, calendar year 2005 for modeling), July 2012-June 2013 (MATES IV) and May 2018-April 2019 (MATES V) [13].Detailed methods for each MATES program are described elsewhere [9][10][11][12], so only a brief summary is included here.

Air toxics emissions inventory and regional modeling
Details of the emission inventory and modeling are available in the MATES V final report [9].The MATES emissions inventories enumerated the major contributors to air toxics in the region, and we used regional chemical transport modeling to quantify the annual average spatial concentration distribution of those pollutants.These inventories included four source categories: point sources (generally larger, higher-emitting facilities), area sources (smaller sources that are relatively common, such as gas stations and dry cleaners), and both on-road and off-road mobile sources.
Each MATES emissions inventory was based on the criteria pollutant inventories from the most recently approved South Coast AQMD Air Quality Management Plan (AQMP), which correspond to the 1997, 2003, 2012 and 2016 AQMPs for MATES II, III, IV and V, respectively, with updates to reflect the MATES study year.The modeling domain in MATES II-V encompassed the South Coast Air Basin (SoCAB) and the coastal shipping lanes located off the shore of Los Angeles, Orange, and Ventura counties using a grid size of 2 km × 2 km.The modeling domains for MATES IV and V were expanded to include portions of the Coachella Valley (CV).To estimate air toxics emissions, we applied the speciation profiles developed by the California air resources board to the total organic gas and PM emissions [14][15][16][17][18][19].The resulting air toxics emissions inventories included a regional emission inventory to examine long-term air toxics trends and a spatially (2 km × 2 km grid) and temporally (hourly) resolved inventory as inputs to a regional chemical transport model.
In MATES IV, the on-road mobile emissions were estimated using the Southern California Association of Governments (SCAGs) traffic activity data, allocating emissions to each grid cell by hour of day and day of week; the estimates used aggregated average diurnal profiles for each grid, but these profiles did not account for month to month or seasonal differences.For MATES V, the temporal profiles were based on Caltrans Performance Measurement System and weigh-in-motion data, which provided unique traffic volume data for each hour of a year at a specified grid cell.Similarly, MATES V utilized the Automated Information System data, a satellite-based location tracking system for ships and boats, to improve the temporal and spatial allocation of emissions from ocean going vessels and commercial harbor craft.In contrast, MATES IV utilized pre-defined shipping channels to allocate emissions to the modeling domain and monthly container throughputs at the ports of Los Angeles and Long Beach to allocate data temporally.
Regional chemical transport modeling was performed using the Comprehensive Air Quality Model with Extensions (CAMx) software, enhanced with a reactive tracer modeling capability (RTRAC) [20], to estimate ambient levels of air toxics due to known toxic emissions sources.The regional modeling method provided a mechanism to predict the transport of emissions from a variety of source categories as well as individual sources to estimate cancer risk throughout the modeling domain.This analysis complemented the techniques used to assess concentrations and risks from the measurement data acquired at the fixed monitoring sites.Table S-1 shows the list of carcinogenic air toxics included in the MATES models and measurements.Model simulations was conducted for the MATES study periods, 1 July 2012-30 June 2013 for MATES IV and 1 May 2018-30 April 2019 for MATES V.
Following the publication of the MATES V final report, an error in the allocation of the on-road mobile source emissions was identified and subsequently corrected, resulting in the modeled risk in the SoCAB [21] being less than 1 per million (or 0.3%) lower than initially reported.The current analysis uses the updated MATES V modeling results.

Air toxics measurements
Each MATES program included a measurement study conducted over the one-to two-year time period, with seven air monitoring stations in Los Angeles county, and one station each in Orange, Riverside, and San Bernardino counties.The station locations were relatively consistent across MATES iterations, with some stations moved due to changes in available sites.The locations were generally selected to be near population centers but also with the intent of reflecting some geographic diversity.Table S-1 shows a listing of carcinogenic air toxics and indicates whether data for each species is available for each of the various MATES iterations.While MATES III used a 1-in-3 day sampling schedule, the other MATES iterations followed a 1-in-6 day, 24 hour integrated sampling schedule.This type of sampling schedule is designed to capture data that is representative of longer-term ambient levels.
Since diesel PM concentrations cannot be measured directly, the measurement-based analysis relies on estimates derived from components of diesel exhaust.Briefly, for MATES III-V, we calculated the ratio of the model-based estimates of elemental carbon with aerodynamic diameter ⩽2.5 µm (EC2.5) to diesel PM in the grid cells where the monitoring sites are located and applied this ratio to the measured EC2.5 levels to estimate diesel PM concentrations at each monitoring site; MATES II used a similar method but based on elemental carbon with aerodynamic diameter ⩽10 µm.While this necessary conversion does introduce some estimation uncertainties, the diesel PM risk estimates are similar between the monitoring-based and modeling-based methods.Additional detail is provided in previously-published reports [9].

Data handling methods for measurement data
To compare total air toxics cancer risks across MATES iterations, any analytes that were not measured in a particular iteration needed to be imputed.For a given MATES iteration, if data for a particular analyte were available for only some monitoring stations, then the average of the available stations was used to impute data for the missing station(s).Detailed information about data availability by station and analyte are published elsewhere [22].If an analyte was not measured during one of the MATES iterations, then the value was imputed based on the basin averages and trends from the other MATES iterations [22].Several different imputation methods were used and compared; each method produced similar total air toxics cancer risk and overall conclusions owing to the limited impacts of the unmeasured analytes [22].This imputation method only applied to calculations of total air toxics cancer risk, and was not applied to data presented for individual analytes.
Across the four iterations of MATES, both the concentrations and method detection limits (MDLs) varied, and for most pollutants, both the concentrations and MDLs declined over time.Table IV-2 of appendix IV of the MATES V Final Report shows all the MDLs for MATES II through MATES V; the same appendix also contains tables with the average concentration, sample size, and percent below the [22,23].Statistical methods were applied to provide a best estimate of the pollutant concentration averaged over the study period.In general, mean concentrations were calculated using the Kaplan-Meier Mean method with Efron's bias correction [24][25][26].However, where more than 80% of the observations were nondetects, the mean concentration was calculated with a lower and upper bound, using zero and MDL substitution, respectively.Additional detail is provided in appendix XI of the MATES V final report [23].

Estimation of air toxics cancer risks
Both the measured and modeled concentrations of air toxics averaged over a 12 month period (for MATES II, IV, and V) or 24 month period (for MATES III) were used to calculate the incremental increase in lifetime cancer risk from exposure to these concentrations using methodology consistent with procedures recommended by the California Office of Environmental Health Hazard Assessment (OEHHA) and detailed elsewhere [9,27].Briefly, the cancer potency factors reflect the dose-response estimates for each carcinogenic air pollutant based on toxicological and epidemiological data (table S-1).To calculate the cancer risk assuming a continuous 30 year exposure period at the same average concentration estimated from the modeled or measured data, the cancer potency is multiplied by the combined exposure factor and multiple-pathway adjustment factor.The combined exposure factor accounts for the increased sensitivity of younger children to the negative health effects of air pollution, and the multiple-pathway adjustment factor accounts for oral and dermal exposure routes, such as ingestion of soil, in addition to inhalation exposures.Additionally, for carcinogenic metal air pollutants, a molecular weight adjustment factor is also included in the product to account for the fraction of the overall weight of the particulates that contribute to the health effects of that metal.The resulting air toxics cancer risk is expressed as the incremental number of potential cancer cases among a population of 1 million people over a 70 year lifetime.
For the modeled data, the air toxics cancer risk is calculated for each grid cell.Additionally, population-weighted regional average risks were calculated by weighting the grid cell risk by the estimated fraction of the population in that grid cell and summing weighted grid cell risks in the inclusion area.The estimated population in each grid cell was calculated for each MATES model year based on data from the SCAG regional travel demand model, which provides population data within each transportation analysis zone (TAZ) based on census data and other datasets [28].We estimated the grid cell population by summing the fraction of the population of each TAZ that was wholly or partially contained in the grid cell.For this study, the inclusion area is defined as the mainland portions of the SoCAB.For the measurement data, we calculated the annual average concentration for each measured air toxic, which is used to calculate the air toxics cancer risk for each station separately.We also calculated the unweighted Basin average as the average of all measurements of each air toxic within a MATES iteration.All 10 monitoring stations are located within the SoCAB.In order to examine temporal trends, the air toxics cancer risks for all four MATES periods were calculated using the same methodology with cancer potency values that were updated in the year 2020 [29].

Analysis of trends
The statistical analysis of temporal trends focused on the overall cancer risk and the individual air toxics contributing the greatest amounts to the overall cancer risk and considered statistical significance at an alpha = 0.05 level.Analyses of individual air toxics used data without imputation.To test whether cancer risks based on measured concentrations in 2018 were statistically lower compared to 1998 levels, we used the annual average concentrations at each station and applied a one-sided paired Wilcoxon signed ranks test [30], excluding stations where data were unavailable for that pollutant for either time point.
We additionally examined the relationship between population density and changes in cancer risk over time from 2012 to 2018.The population density was calculated as the population within a grid divided by the area within the grid that is classified as residential land use based on the 2016 land use data from SCAG.The total population of approximately 16.8 million people in 2018 includes people in the 3025 grids that had (1) populations greater than zero in both 2012 and 2018, and (2) had residential land use areas greater than zero; these criteria excluded 2254 of the original 5279 grids.The 756 grids in the top quartile (i.e.most densely populated) represent approximately 53% of the total population while the 756 grids in the bottom quartile (i.e. more sparsely populated) represent approximately 1.5% of the population of the SoCAB in 2018.The top and bottom quartiles were compared using the two-sided Wilcoxon ranksum test [30].For each grid cell, we calculated the percent change from 2012 to 2018 based on modeling results.

Temporal trends in air toxics cancer risk-measurement results
Table 1 shows the basin average and range of station-average concentrations of elemental carbon, benzene, 1,3 butadiene, and formaldehyde for each MATES study.For MATES II, PM10 was used to derive elemental carbon, and PM2.  1) [22].Diesel PM contributes the largest share of the cancer risk, contributing to 50%, 63%, 73%, and 58% of the total cancer risk in MATES V, IV, III and II, respectively.Diesel PM also showed the largest decreases, from a peak in 2004 (MATES III) of 2656 per million to 356 per million in 2018 (MATES V), or an 87% (range among stations 82%-91%) decline.Between 1998-2018, ambient levels of benzene, 1,3-butadiene, perchloroethylene, hexavalent chromium, and arsenic declined between 75%-87%.Ambient formaldehyde and acetaldehyde showed an overall decrease by 25% and 13%, respectively, which includes a 31% increase in formaldehyde and 69% increase in acetaldehyde observed between 2012 and 2018 data.The most recent MATES data identified diesel PM, benzene, 1,3-butadiene, and formaldehyde as the highest contributors to the total cancer risk (figure 1).These four air toxics contributed to approximately 73%, 80%, 83%, and 71% of the total cancer risk in MATES V, IV, III and II, respectively, based on measurement results.Comparing data from 1998 to 2018, the difference in overall average cancer risk and the risk from each of these top four contributors reached statistical significance at the alpha = 0.05 level using a one-sided Wilcoxon signed ranks test.

Temporal trends in air toxics cancer risk-modeling results
The cancer risk reduction trends derived from air toxics emission inventories and regional air toxics modeling are consistent with the measurement-based trends observed.Between 1998 and 2018, population-weighted average air toxics cancer risk for the SoCAB decreased by 82% from 2564 to 454 per million (figure 2).Risk from DPM decreased from 1759 to 305 per million, while risk from other air toxics decreased from 805 to 149 per million, both reflecting a reduction of approximately 82%.Between 2012 and 2018 (the most recent two data points), the modeling results showed a 54% decrease in population-weighted air toxics cancer risk, including a 56% decrease in risk from DPM and a 51% decrease in risk from other air toxics.

Spatial variation in air toxics cancer risk-modeling results
Figure 3 shows the spatial distribution of the 2018 (MATES V) modeled cancer risk in each grid cell, which ranged from 74 to 1066 per million.The grid cells having the highest cancer risk were located near the Ports of Los Angeles and Long Beach, the Los Angeles International Airport, and an area around a railyard southeast of downtown Los Angeles.Similar to previous MATES results, the higher-risk areas tend to be along transportation and goods movement corridors, consistent with diesel PM being the main risk driver.Concentrations of benzene and 1,3-butadiene are more uniformly distributed across the region with somewhat higher levels in the coastal area.With light-duty vehicles being the main contributor to these emissions, along with some stationary sources such as refineries and gas stations, this geographic pattern is reasonable to expect.Table 2 shows summary statistics of the 2012 and 2018 data.The median absolute deviation (MAD), a measure of variation, across grid cells comparing 2012 and 2018 data, with 2018 data showing less variation for diesel PM, 1,3-butadiene and formaldehyde compared to 2012, although variation across grid cells was similar for benzene.Figures S1-S4 show the change in modeled cancer risk from MATES IV to MATES V for DPM, benzene, formaldehyde, and 1,3-butadiene, where negatives correspond to lower risk in MATES V. We also analyzed spatial patterns of emission changes from MATES IV to MATES V for each of these pollutants (not shown).Modeled risk trends reflect the combined effect of emission trends and specific meteorological conditions during the one-year MATES IV and MATES V studies.Because each iteration of MATES used the meteorological data corresponding to the study time frame, and because there were changes in the spatial surrogates used to allocate emissions to modeling grids and to assign mobile source emissions, it is important not to overinterpret changes at the individual grid cell level.Instead, these data are used to reveal larger-scale patterns across the region from 2012 to 2018.
DPM cancer risk decreased throughout the South Coast Air Basin and along I-10 in the CV, with the largest decrease at the port area in Long Beach and in downtown Los Angeles.This is expected due to the large decrease of DPM emissions from trucks and other sources involved in goods movement.Benzene emissions and cancer risk tended to increase off the coast of Los Angeles county and Orange county, in Santa Clarita, along I-15 near Lake Elsinore, and along some other freeways.Benzene cancer risk decreased slightly in most of the South Coast Air Basin, likely due to the specific meteorological data used for the modeling and the decrease of total benzene emissions.Although 1,3-butadiene emissions increased by 17% from MATES IV to MATES V, the cancer risk decreased slightly in much of Los Angeles county and northern Orange county, which may be attributed to the specific meteorological data used for MATES IV and MATES V.Primary formaldehyde emissions slightly in most grid cells in the South Coast Air Basin.The largest emission decreases occurred near LAX airport and the port area in Long Beach and the largest increases were in the Angeles National Forest.Cancer risk due to total formaldehyde decreased slightly throughout the South Coast Air Basin and CV.Cancer risk due to acetaldehyde also decreased slightly, with a maximum decrease of 8 per million in a grid cell, and with larger decreases occurring in inland areas and mountains.

Comparison of cancer risk and population density
A comparison of the modeling results from the most recent two iterations can provide information about recent trends across this region on a pollutant-by-pollutant basis.Figure 4 shows boxplots of the percent change in cancer risk from 2012 to 2018, comparing the top and bottom quartiles of grid cells based on population density.Negative values indicate a decrease in cancer risk from 2012 to 2018.The total cancer risk from all modeled pollutants, diesel PM, and benzene show larger declines in more densely populated areas compared to more sparsely populated areas.These differences are found to be statistically significant using the two-sided Wilcoxon rank sum test (p < 0.01 for each test).Formaldehyde shows the opposite pattern, with larger declines from 2012 to 2018 in the more sparsely populated regions (p < 0.01).

Discussion
Air toxics cancer risks in the greater Los Angeles region have decreased by more than 80% between 1998 and 2018, including a decrease of about 50% in the period between 2012 to 2018, based on population-weighted model-derived estimates.These decreases are largely driven by decreased concentrations of diesel PM, which remains as the single most important contributor to air toxics cancer risk in the region.The next-highest pollutants contributing to air toxics cancer risk were benzene and 1,3-butadiene, which also showed decreasing trends during these time periods.Geographic areas with higher air toxics cancer risks include the major goods movement corridors, with generally higher risks near major truck routes, and the highest risks observed near the marine ports, a major international airport, and a railyard southeast of downtown Los Angeles.Temporal trends indicate that as air toxics levels decreased overall, there have been larger decreases near stations that originally had higher air toxics levels, resulting in a narrowing of the differences across stations.Larger declines in air toxics levels were observed in areas with higher population density.
These findings provide an updated assessment of air toxics health impacts in this region, and they are consistent with prior findings in California [31].Propper et al used monitoring data from sites across California and examined ambient concentrations of air toxics from 1992 to 2012.The authors reported a 68% decrease in diesel PM levels, a 90% decrease in benzene, 1,3-butadiene, perchloroethylene, and hexavalent chromium levels, and an approximate 20% decrease in formaldehyde and acetaldehyde levels.and the bottom quartile of grids ('Sparse' population), with cancer risks estimated from modeling data.Negative values indicate a decrease in cancer risk from 2012 to 2018.The inner line within the boxplots denotes the median value, the boxes denote the 25th and 75th percentiles, and the whiskers denote the 10th and 90th percentiles.The two-sided Wilcoxon ranksum test was used to compare the modeled cancer risk ratio of MATES V over MATES IV for the 'Dense' and 'Sparse' population grids for all modeled pollutants; the difference was significant with p < 0.01.Similar tests were done for Diesel PM, benzene, and formaldehyde, all with p < 0.01.
Cancer risk from these seven air toxics combined decreased by 76% in this time period.There is some expected overlap in the results with the current study, since several of the monitoring sites included in this study were also used in the prior MATES programs, and the study notes consistency in the estimated decreases in diesel PM levels for the corresponding time period.Because this study used similar methods to estimate cancer risk (i.e.applying the OEHHA risk assessment guidance but for inhalation-only exposures), it is reasonable to compare these results quantitatively.The percent of the air toxics cancer risk attributable to diesel PM inhalation exposure was estimated to be 71% in Propper et al, based on their assessment of seven pollutants.These results are generally consistent with the findings based on MATES data, including the generally smaller overall decrease in formaldehyde and acetaldehyde levels.Some differences in risk estimates across the two studies are expected since the MATES study included more recent years of data and accounted for multiple exposure pathways (rather than inhalation-only risks).Both the Propper study and the current MATES analysis show that ambient air toxics cancer risks are far higher than the U.S. EPA or OEHHA threshold for unacceptable cancer risk (100 in a million), although this threshold is intended to be applied only to individual sources, such as a clean-up site or a facility, rather than to ambient concentrations [32][33][34].
The decreases in air toxics cancer risk from diesel PM between 2012 and 2018 are consistent with expectations based on the regulations that were implemented in California during this time period.These policies include the California Truck and Bus Regulation ( §13 CCR 2025), which required PM filters or upgrades on heavy-duty engines beginning in 2012; the Airborne Toxic Control Measure (ATCM) for transport refrigeration units that set emissions standards beginning in 2009 ( §13 CCR 2477); and the ATCM for mobile cargo handling equipment that reduced emissions with compliance deadlines beginning in 2006 ( §13 CCR 2479).Concurrent programs incentivizing the turnover to lower emission trucks, such as the Carl Moyer Memorial Air Quality Standards Attainment Program and the Proposition 1B Goods Movement Emission Reduction Program, also helped to accelerate adoption of cleaner heavy-duty engines in advance of regulatory deadlines [35,36].Our results showing generally larger improvements in air toxics levels in densely populated areas compared to sparsely populated areas, suggest that as air toxics levels have declined, the differences in air toxics levels between more urbanized and more rural areas has also decreased.
Two recent studies examined U.S. national trends using monitoring data [6,7] or modeling data [6].Both of these studies focused on HAPs with unit risk estimates established by the U.S. EPA Office of Air Quality and Planning Standards, so the studies do not explicitly provide estimates of diesel PM-related risk, but do report trends and patterns for benzene, 1,3-butadiene, formaldehyde, and other toxics.Additionally, because both studies used different methods for cancer risk estimation, the risk estimates are not directly comparable to the MATES results provided here, although it is still valid to compare overall trends.Notably, only the MATES study combined both monitoring and modeling data in the same study to provide a more robust assessment of ambient air toxics exposures.
Weitekamp et al analyzed monitoring data from 2013-2017, including two stations within the South Coast AQMD region.For these two stations, the authors identified carbonyls (mostly formaldehyde) as the main risk drivers and the overall temporal trend in cancer risk across the five-year time period was generally flat, decreasing by 5% and 12% at Los Angeles and Rubidoux, respectively.In contrast, the MATES results for the equivalent time period showed an approximately 40% decrease in air toxics cancer risk based on measurement data.These differences are not surprising, given that the Weitekamp study did not explicitly account for cancer risk from diesel PM and the decreases in risk observed in MATES were driven by reductions in diesel PM levels.The MATES measurement data showed an increase in formaldehyde and acetaldehyde levels between 2012 to 2018, but the risk increases associated with the increased carbonyl levels were more than offset by decreases in benzene, 1,3-butadiene, and other risk contributors that declined substantially during this time period.The MATES V modeling data identified that the majority of these carbonyls were from secondary formation rather than direct emissions, so the recent increase in ambient levels cannot be ascribed to any one particular source category.Notably, the modeling results showed an overall decrease in formaldehyde from 2012 to 2018, suggesting that there may be some other contributors that resulted in these differences between the measurement and modeling data for this pollutant.The oxidation of anthropogenic and biogenic volatile organic compounds (VOCs) is a significant source of formaldehyde in the atmosphere [31] along with direct emissions from mobile sources and biomass burning.While anthropogenic VOC emissions in the region have declined over the past several decades [37,38], biomass burning from wildfires and biogenic VOC emissions are generally not influenced by air quality regulations and thus may exhibit random year-to-year variability.
Using the 2014 NATA emissions inventory, Scheffe et al employed a hybrid modeling approach using an AERMOD, Gaussian dispersion model and community multiscale air quality chemical transport model to estimate cancer risk from HAPs along with an exposure model to account for typical time-activity patterns [6].Similar to Weitekamp et al, the authors found that approximately half of the air toxics cancer risk is attributed to ambient formaldehyde, mostly from secondary formation.Other key cancer risk contributors included 1,3-butadiene, acetaldehyde, benzene, carbon tetrachloride, and naphthalene.In the MATES dataset, aside from diesel PM, the top contributors to inhalation cancer risk based on the modeled exposures were benzene, formaldehyde, 1,3-butadiene, hexavalent chromium, and acetaldehyde.Carbon tetrachloride was not modeled in MATES (because there are no local sources of emissions and the measured concentrations were uniform across monitoring stations), but was included in the measurement study, and is estimated to contribute about 7% of the overall multi-pathway cancer risk.These findings regarding the top contributors to cancer risk are generally consistent, with the exception of the NATA study not having explicit information about diesel PM impacts.
The Weitekamp study included a comparison of the results from the measurement and modeling approaches, using the census tract to derive the NATA estimate.This comparison found that the measured concentrations of formaldehyde were higher than the NATA estimates in most locations, and the average cancer risks estimated using monitoring-based data were 40% higher than modeled estimates based on the NATA study.For the two Southern California monitors in this study, the measurement-based cancer risks were 104% and 78% higher than the risks estimated based on NATA [7].
Within the MATES dataset, while the overall risk estimates from both measured and modeled air toxics are similar, there are some key differences that limit the ability to compare the measurement results and the modeling results directly.First, the modeling analysis focuses on the pollutants that have previously been identified to be the major risk drivers, and included 15 carcinogenic air toxics including diesel PM, while the monitoring-based risk calculations included 30 carcinogenic air toxics (but no direct measure of diesel PM).Second, the calculation of cancer risks from diesel PM for the monitoring-based analysis relies on a conversion factor that uses measured concentrations of EC.Finally, the summary results from the modeling analysis are reported as population-weighted estimates, while the measurement-based summary results are presented as an unweighted average across the 10 monitoring stations.However, the monitoring stations were generally located in more populated areas, so they largely represent areas with higher population density.When examining the cancer risks based on the same set of 15 air toxics included in the model, the model-based estimate ranged from 21% lower (Long Beach) to 61% higher (Rubidoux) compared to the measurement-based estimates, with an average difference of 15%.
There are some inherent challenges and uncertainties in estimating air toxics cancer risks, including the necessity of making assumptions about population exposures.The MATES program focuses on outdoor air pollution exposures and does not account for time spent indoors, which generally reduces exposures to outdoor pollutants.Workplace exposures (both indoor and outdoor) and residential indoor air pollution, such as VOCs from building materials, are not included in the study and can be significant sources of exposures [39].Therefore, these exclusions could either over-or underestimate total air toxics exposures.Limitations pertaining to the measurement, modeling, and risk estimation methods have been discussed previously [9].Briefly, the methods applied in the present study make a conservative assumption that individuals are exposed to those estimated levels continuously for a 30 year period, so there may be some overestimation of cancer risks based on outdoor air toxics and a potential underestimation of cancer risk from air toxics that tend to be elevated indoors.While the suite of compounds with significant contributions to cancer risk measured during each MATES iteration is intended to be reasonably comprehensive for priority air toxics, advances in measurement techniques, changes in risk factors, and the discovery of new sources that contribute to risk may occur.For example, ethylene oxide was not measured during the MATES V campaign due to the absence of practical analytical methods for measuring low background levels in 2018.However, since the MATES V measurements were conducted, OEHHA has indicated that the cancer risk factor may be revised significantly upward and analytical methods have improved such that detection limits are low enough to quantify typical background levels.The periodic nature of the MATES campaigns allows for inclusion of additional species as science evolves.For example, ethylene oxide will be measured as part of the MATES VI campaign.
Risk estimates rely on having unit risk factors for each carcinogen.Although the unit risks are based on the state of the science, there are uncertainties in estimating these factors based on published data [27], which can either over-or under-estimate the risk; however, these assessments are finalized only after extensive scientific and public review.Agencies may choose to establish different risk factors for use in risk assessments, most notably, OEHHA has cancer risk factors for diesel PM but U.S. EPA has not established an equivalent risk factor (but concludes that diesel exhaust contributes substantially to air toxics cancer risk) [40].By including diesel PM in the study and using the OEHHA risk factor consistently across MATES iterations, our study is able to highlight this important contributor to air toxics cancer risk and track trends.This use of a conversion factor to calculate diesel PM cancer risks based on measured elemental carbon introduces some uncertainties, but the overall consistency of the trends observed across the measurementbased and modeling-based results, and the observed decreasing trend of measured elemental carbon, provide additional confidence that the shape of these trends is correct.
Multiple factors affect the spatial patterns of modeled risk, including meteorological conditions that change from year to year.Additionally, updates to the WRF and CAMx models and improvements in the methods of allocating county-total emissions to modeling grids are examples of methodological improvements that limit direct grid cell level comparisons across MATES iterations, although comparisons of larger spatial patterns are more reliable.Importantly, trends identified through modeled risk should be interpreted as overall, long-term trends across the study years, incorporating changes in emissions as well as meteorological factors.Although the MATES V modeled data corrected for the erroneous geographic allocation of mobile source emissions, the MATES IV data used in these analyses did not include the error correction.To assess potential bias caused by this error, we performed sensitivity tests comparing the results of the trend analysis by population density using the MATES V corrected and uncorrected results and found that the overall difference in estimated air toxics cancer risk was less than 1%.Therefore, although the MATES IV modeled data still contains this same error, the magnitude of the error is small relative to the magnitude of the overall temporal trends in risk, so it is unlikely to affect the conclusions of our analysis.Given these uncertainties, the risk estimates can be interpreted as estimates of relative impact based on a set of consistently-applied assumptions, and the main conclusions to draw from these results are the overall temporal trends and spatial patterns of air toxics cancer risk.Finally, the Wilcoxon rank sum test assumes independence within and between samples; the validity of this assumption is limited when applying it to the gridded model data because air toxics concentrations are not spatially independent.

Conclusions
The recurring assessment of air toxics health impacts provided by the MATES program helps to quantify the impact of air toxics reduction programs in the South Coast AQMD region, including the impact of both regulatory programs and incentive-based programs [35,36,41,42].The long history of the MATES program allows for an evaluation of trends over a 20 year time period, and the application of statistical methods to account for missing data, improvements in measurement techniques, and slight changes in substances measured allows for appropriate comparisons over time.Additionally, the detailed modeling techniques allow for an examination of the spatial trends across this large, highly populated region.Together, these data help provide a unique, detailed assessment of air toxics exposures to guide future efforts, pointing especially to the importance of reducing diesel pollution to improve region-wide air toxics impacts.The measurement and modeling methodologies employed throughout MATES can be adapted in other regions around the world to quantify the impact of air toxics, draw conclusions about the contributing sources, and track trends in air toxics.
The MATES emissions inventories have previously identified the major sources of diesel PM in the South Coast AQMD jurisdiction to be both on-road and off-road engines, including heavy-duty trucks, locomotives, and industrial off-road equipment.There are existing efforts to reduce diesel PM emissions in this region, including the California Truck and Bus Regulation ( §13 CCR 2025) that has a key implementation deadline in 2023, the South Coast AQMD Rule 2305 (Warehouse Indirect Source Rule) with implementation deadlines between 2022 and 2026 [43], and the California Advanced Clean Trucks Regulation ( §13 CCR 1963) which requires manufacturers to sell an increasing percentage of zero-emission trucks beginning in 2024.These efforts, along with additional ongoing toxics rule development and program implementation efforts at local and state air agencies, will help reduce the health impacts of diesel PM and other air toxics.While the progress of toxics reduction programs over the past several decades are encouraging, with the data suggesting that the spread in air toxics cancer risk is decreasing between the most and least polluted communities, continued work is needed to further reduce the carcinogenic risks of air toxics to protect public health.

Figure 1 .
Figure 1.Temporal trends for (a) sum of cancer risk for all measured air toxics in the South Coast Air Basin and for (b) diesel particulate matter, (c) benzene, (d) 1,3-butadiene, and (e) formaldehyde.These four air toxics were the top contributors to the total cancer risk in MATES V.The p-values shown are from the one-sided Wilcoxon signed ranks test comparing data from MATES II to MATES V.

Figure 2 .
Figure 2. Modeling-based population-weighted air toxics cancer risk (per million) in the SoCAB.

Figure 3 .
Figure 3. Air toxics cancer risk based on 2018 modeling data (MATES V).The MATES V monitoring site locations are also shown on the map.

Figure
Figure Percent change in cancer risk from 2012 to 2018 for the top quartile of grids based on population density ('Dense'and the bottom quartile of grids ('Sparse' population), with cancer risks estimated from modeling data.Negative values indicate a decrease in cancer risk from 2012 to 2018.The inner line within the boxplots denotes the median value, the boxes denote the 25th and 75th percentiles, and the whiskers denote the 10th and 90th percentiles.The two-sided Wilcoxon ranksum test was used to compare the modeled cancer risk ratio of MATES V over MATES IV for the 'Dense' and 'Sparse' population grids for all modeled pollutants; the difference was significant with p < 0.01.Similar tests were done for Diesel PM, benzene, and formaldehyde, all with p < 0.01.

Table 1 .
Average and range of pollutant concentrations across monitoring stations, by MATES iteration.
a For MATES II, PM10 was used to derive Elemental carbon, and PM2.5 was used to derive Elemental carbon for MATES III-V.
5 was used to derive elemental carbon for MATES III-V.The concentrations have generally declined over time, though not always monotonically.Based on measurement results, the total cancer risk from air toxics decreased by 84% from a basin average of 4548 per million in 1998-1999 (MATES II) to 718 per million in 2018-2019 (MATES V), with all stations showing this decreasing trend, see figure 1.While not all species have shown a monotonic decline in cancer risk over time, all species do show an overall decrease from 1998 to 2018 (figure

Table 2 .
Mean, median, min, max, and median absolute deviation (MAD) of modeled cancer risk across 5279 grid cells comparing MATES iterations from 2012 to 2018 (expressed in units of cancer risk per million).