Abstract
Growing evidence links traffic-related air pollution (TRAP) to adverse health effects. Previous air pollution studies focused on a few commonly measured pollutants with poor spatial resolution. Well-designed mobile monitoring studies may address this limitation. We designed an extensive mobile monitoring campaign to characterize TRAP exposure levels for a Seattle-based cohort, the Adult Changes in Thought (ACT) study. The campaign measured particulate matter ≤ 2.5 µm in diameter (PM2.5), ultrafine particulates (UFP), black carbon (BC), nitrogen dioxide (NO2) and carbon dioxide (CO2) at 309 stop locations representative of the cohort. We collected about 29 two-minute measures at each location during all seasons, days of the week, and most times of day for a one-year period. Annual average concentrations of UFPs had similar degrees of spatial variability as BC and NO2 but higher degrees of variability as CO2 and PM2.5. UFPs had less temporal variability than other pollutants. Validation showed good agreement between our BC, NO2, and PM2.5 measurements and regulatory monitoring sites. The results from this campaign will be used to assess TRAP exposure in the ACT cohort.
Synopsis We assessed annual-average traffic-pollutant levels with high spatial resolution at locations representative of participant residences using a temporally balanced short-term mobile monitoring campaign.
1 Introduction
An extensive body of evidence has linked air pollution to adverse health effects including respiratory, cardiovascular and mortality outcomes.1 Recent evidence has begun to link traffic-related air pollution (TRAP) exposure to cognitive function among various populations, including the elderly.2–4 While TRAP is a complex mixture that varies over time and space, pollutants include ultrafine particles (UFP; typically defined as aerodynamic diameter ≤ 100 nm), black carbon (BC), oxides of nitrogen including nitrogen dioxide (NO2), and carbon dioxide (CO2).5 UFPs are increasingly associated with important health outcomes including more neurotoxicity and systemic inflammation than larger particles.6–12
To date, however, much of the epidemiology air pollution research has been limited to the federally defined criteria air pollutants, monitored nationwide through the EPA’s regulatory Air Quality System (AQS) monitoring network. This network has monitored criteria pollutant levels throughout the US since the 1990s, and none specifically include UFPs.13 Furthermore, this network is spatially sparse and thus fails to capture the spatial variability of more quickly decaying pollutants, including many TRAPs.14 The Seattle Census Urbanized Area, for example, averages about 1 monitor every 174 km2 (∼14 active monitors within a land area of about 2,440 km2), most of which measure fine particulate matter with diameter of less than 2.5 µm (PM2.5) and BC.15,16
Mobile monitoring campaigns for assessing air pollution exposure have been used since at least the 1970s and have become increasingly common in recent years in an effort to address the limitations of traditional fixed site monitoring approaches.43,17–23 Typically, a vehicle is equipped with air monitors capable of measuring pollutants with high temporal resolution. Short-term sampling repeatedly occurs with this platform at predefined locations. Past work has shown that repeated short-term air pollution samples can be used to calculate unbiased long-term averages, thus reducing the need for continuous fixed-site monitoring.17,18 Because the sampling duration at individual locations can be quite short, campaigns can increase their spatial coverage with a single platform, thus making this approach more time- and cost-efficient than traditional fixed-site monitoring.
Still, the designs of past mobile monitoring campaigns have arguably limited their epidemiologic application. Importantly, most campaigns have sampled during limited time periods, for example, weekday business hours during one to three seasons.19,24–26 We previously showed that these limited sampling campaigns likely result in biased long-term human exposure estimates because they do not capture the high temporal variability of many TRAPs, and that the exact degree of bias varies (is not consistent) across site.27 Additionally, many campaigns have sampled along non-residential areas such as highways and industrial areas where air pollution levels may be much higher than the levels that most people are exposed to. Furthermore, most have collected non-stationary (mobile) on-road samples rather than stationary samples along the side of the road closer to participant residences. While non-stationary designs increase spatial coverage, further work is needed to demonstrate whether these are representative of residential human exposure levels.19,28
To address the limitations of past campaigns, we designed an extensive, multi-pollutant mobile monitoring campaign to characterize TRAP exposure levels for the Adult Changes in Thought (ACT) study cohort. ACT is a long-standing, prospective cohort study that has been investigating aging and brain health in the greater Seattle area since 1995.29 The campaign measured TRAP at locations representative of the cohort including 309 stops and thousands of miles worth of on-road measurements in a temporally balanced approach throughout the course of a year. The goal of this paper is to describe the mobile monitoring design’s sampling methodology and to summarize the TRAP measures and spatial covariates associated with TRAP in our study region. To the best of our knowledge, this is one of the most extensive mobile monitoring campaigns conducted in terms of the pollutants measured, the spatial coverage and resolution, and the campaign duration and sampling frequency. This campaign will support future epidemiologic investigations of TRAP exposure in the ACT cohort.
2 Methods
Briefly, multiple pollutants including UFP, BC, PM2.5, NO2 and CO2 were simultaneously measured with high quality instrumentation at 309 stationary (stop) locations along the side of the road. Locations were representative of the cohort’s large spatial and geographical distribution throughout the greater Seattle area. Fixed driving routes were developed to additionally collect thousands of miles of non-stationary, on-road measurements between stop locations. A temporally balanced, year-long driving schedule that measured TRAP during all seasons, days of the week, and most times of the day enabled us to estimate unbiased annual average estimates at the site level. Details are described below.
2.1 Spatial Compatibility of the Selected Stop Locations and the ACT Cohort
We selected a mobile monitoring region in the greater Seattle, WA area that was roughly 1,200 land km2 (463 mi2; Figure 1). The monitoring region was composed of Census Tracts where the majority of the ACT cohort had historically resided between 1989-2018 (87% = 10,330/11,904 locations). This large region fell in western King County and southwest Snohomish County, and it included a variety of urban and rural areas with various land uses including residential, industrial, commercial, and downtown areas. We used the Location-Allocation tool in ArcMap (ArcGIS v. 10.5.1)30 to select 304 sampling locations within the monitoring region that were representative of the ACT cohort (approximately one monitoring location per 33 participant locations; see Supplementary Information [SI] Note S1 for details). Stops were selected to maximize the spatial coverage within the monitoring region while minimizing the distance between sampled stops and cohort locations. Five additional stops were collocations at nearby regulatory air quality monitoring sites measuring pollutants similar to our platform (see below). In total, there were 309 stops. The average (SD) distance between a cohort location and the nearest monitoring stop was 611 (397) m. To ensure that the monitoring stops and cohort locations were spatially compatible, we compared the distribution of various TRAP-related covariates at these locations (Figure 2).31
2.2 Fixed Routes
We used ArcMap’s Network Analyst New Route tool30 and Google Maps32 to develop nine fixed routes based on the 309 stop monitoring locations (SI Note S1). Each route ranged from 75-168 km (47-104 miles) in length and had 28-40 stops (SI Table S1). All routes started and ended at the University of Washington and were intended to maximize residential driving coverage (i.e., reduce highway driving and driving on the same roads). Routes were downloaded from Google Maps to a smart phone and Garmin GPS Navigation System and navigation was set to replicate the same route each time regardless of traffic conditions. SI Figure S1 shows the distribution of various TRAP-related covariates on 100 m on-road segments and cohort locations. As before with the stop locations (Figure 2), road segments had similar geocovariate distributions as ACT cohort locations, indicating that these were spatially compatible.
2.3 Sampling Schedule
Sampling was planned to cover an entire year and was conducted from March 2019 through March 2020 during all seasons and days of the week between the hours of 4 AM and 11 PM. Our previous work has shown that this balanced but slightly reduced sampling schedule taking driver safety and operational logistics into consideration should still generally produce unbiased annual averages.27 To increase temporal coverage, routes were started at different times of the day and driven in both clockwise and counterclockwise directions. A single route was driven each day (∼4-8 drive hours). Make-up routes were conducted throughout the study to resample sites with missing readings (i.e., due to instrumentation or driver errors). Make-up routes occurred during similar times as the originally scheduled sampling time (i.e., season, day of the week, general time of day).
Twenty-eight two-minute samples were scheduled to be collected at each location while the vehicle was parked along the side of the road. This design choice was justified by our additional analyses of one-minute data from a near-road and a background regulatory site in Seattle. These analyses showed that at least 25 two-minute samples were sufficient to produce annual average estimates with a low average percent error (See SI Figure S2). Furthermore, there was only a negligible improvement in annual average estimates when the sampling duration was extended from 2 to 60 minutes.
2.4 Data Collection
We equipped a Toyota Prius hybrid vehicle with fast-response (1-60 sec), high-quality instrumentation that measured various particle and gas pollutants. Pollutants included UFPs of various size ranges (see Table 1 for size details; instruments were: TSI NanoScan 3910, TSI P-TRAK 8525 with and without a diffusion screen, and Testo DiSCmini), BC (AethLabs MA200), PM2.5 (Radiance Research M903 nephelometer), NO2 (Aerodyne Research Inc. CAPS) and CO2 (Li-Cor LI-850). SI Table S2 for instrumentation details. Note that we refer to “UFP” counts even though instruments measured particles larger than 100 nm in diameter since most particles by count are smaller than 100 nm. Carbon monoxide (CO) was also collected, but measurements did not meet our quality standards and were not subsequently used. The platform additionally collected temperature, relative humidity, and global positioning with real-time tracking. We had duplicates (back-ups) of every instrument type that were periodically collocated for quality assurance purposes (see Quality Assurance and Quality Control section below). Instruments were in the back of the vehicle where they were powered by two rechargeable batteries and connected to gas- or particle-specific manifolds (SI Figure S3-S4). These were connected to a rooftop inlet facing the front of the car used to collect ambient air samples. Instrument clocks were all synchronized at the beginning of each drive within ∼2 seconds. Instruments were started within an hour before the start of each drive and continuously run until the end of the route.
Drivers were instructed to follow the specific Google Map route directions and to take notes of any field anomalies (e.g., sampling behind a school bus or next to a construction site).
Instrument data files were uploaded to a secure data management system (MySQL) after each drive using standardized naming conventions to automate data uploads and the generation of daily data reports (see SI Note S2 for further details).
2.5 Summary Measures
We calculated the median pollutant concentrations for each two-minute site visit. While means can be highly influenced by large concentration deviations (which may be important in some settings), medians are more robust to outliers and may better capture the typical values of skewed data.
We estimated UFP concentrations for particles 20-36 nm in diameter by calculating the difference between the unscreened (20-1,000 nm) and screened (36-1,000 nm) P-TRAK readings.
We estimated PM2.5 concentrations from nephelometer readings using a calibration curve fit to regulatory monitoring data between 1998-2017 (SI Equation S1). Nephelometer light scattering is strongly correlated with PM2.5 and has been used in the Puget Sound region to monitor air quality since 1967.33 We fit the model using daily average measurements from nine non-industrial regulatory air monitoring sites in the region where both PM2.5 (using federal reference methods) and nephelometer light scattering data were collected. We excluded the years 2008-2009 due to nephelometer instrumentation issues noted by the local regulatory agency. The model’s leave-one-site-out cross-validated R2 and root mean square error (RMSE) were 0.92 and 1.97 µg/m3, respectively.
To gain an overall assessment of the relative importance of the three types of particles and two gases, we scaled pollutant concentrations based on each pollutant’s distribution (score = [value-median]/interquartile range) and aggregated at each site (the sum of scores) to characterize overall spatial air pollution trends. In addition to the PM2.5 and BC particle measures, with multiple UFP instrument data available in our study, we focused our analysis on total UFP as measured by one of the instruments, the P-TRAK (20-1,000 nm particles).
We used Lasso regression to characterize the geographic covariates most strongly associated with pollutant concentrations. Cross-validation was used to select the tuning parameter, lambda. Covariates included various indicators of land use, population density, green space, built environment and others (see SI Table S10 for details). There were 835 initial geographic covariates (geocovariates) that were reduced to 183 as follows.34 First, we excluded variables if they lacked variability (less than 20% of the data were different from the most common value) since these were not likely to improve, and could even worsen, the model fit. Next, we excluded all land use proportion variables where the maximum proportion observed in the data was less than 10%. Low values for these variables indicated that these land use types made up a small fraction of the land relative to other land use variables and would not likely have a meaningful impact on observed pollutant concentrations. We dropped variables if their distribution at monitoring sites was very different than their distribution at ACT cohort locations in an effort to reduce model extrapolation later on. This included variables where the standard deviation (SD) in the ACT cohort was more than five times the SD of the mobile monitoring data. We dropped variables if too many outliers were observed in the data (>2% of the total data). Finally, we log-transformed proximity variables to better model pollutant exponential decay with increasing source distance.
2.6 Quality Assurance and Quality Control
We conducted various quality assurance and quality control (QAQC) activities throughout the study period to ensure the reliability and integrity of our data. Activities included calibrating gas instruments; checking particle instruments for zero concentration responses; assessing collocated instruments for agreement; inspecting time series data for concentration pattern anomalies; and dropping readings associated with instrument error codes or those outside the instrument measurement range. SI Note S2 has additional details. Site visit medians (∼2 min) and annual averages for BC, NO2 and PM2.5 estimated from these data were additionally compared against estimates from the five regulatory air monitoring collocation sites.
Computing details are in SI Note S3.
3 Results
3.1 Data Collected
In total, the campaign collected about 1,367 hours of data (driving and stopped) along approximately 34,335 km (21,347 mi), or 6,786 100 m road segments along non-A1 roads (SI Table S1). Each route was driven on 30-35 different occasions. After dropping stop concentrations that did not meet the quality assurance standards (0.61%), the final analyses included over 70,000 two-minute median stop samples (almost 9,000 samples per instrument) collected over the course of 288 drive days from 309 monitoring sites (Table S7).
Sites were sampled an average of 29 times, ranging from 26 to 35 times. Due to the logistical constraints of sampling 309 sites with one platform, some locations were visited fewer times of the day than other locations, though sampling times were still well distributed throughout the day (e.g., morning [e.g., 7 AM], afternoon [e.g., 3 PM] and evening [e.g., 8 PM]; see SI Figure S7).
3.2 Stop Visits
SI Table S6 shows the distribution of median site visit pollutant concentrations for all of the two-minute visits (N = 309 sites x ∼29 visits/site). As a result of UFP particle size distributions, instruments that captured smaller particles more efficiently typically reported higher total concentrations. Ratios of the 95th to 5th quantile concentrations (“Q95/Q05”) show that BC, NO2 and UFPs saw the greatest concentration variability across sites and visits, while CO2 and PM2.5 had relatively lower variability. All UFP measures had relatively high variability, though there were some differences associated with different instruments.
The SI shows the pollutant measurements collected during each visit by site (∼29 visits/site; Table S8, Figure S8-S9). There was a high degree of variability within most sites, between sites, and for different pollutants. UFP instruments measuring smaller particles were more likely to report higher readings and thus higher levels of within-site variability. Once again, UFPs, BC and NO2 had more within-site variability (see “Min/Max” IQR ratio in SI Table S8).
As expected, pollutant concentrations varied temporally throughout the study period (SI Figure S10-S13). Fall, for example, generally had relatively higher pollutant concentrations than summer and winter. Weekends generally had lower concentrations than weekdays. Midday hours generally had lower concentrations than surrounding hours.
Pollutant-specific Analysis of Variance (ANOVA) models of visit concentrations (SI Table S9) indicated most of the variability seen occurred within sites, rather than across sites or over time. UFPs had the highest within-site variability (∼86-90% of the total) while NO2 had the least (∼67%). Relatively speaking, NO2 and UFP visit concentrations had the most spatial variability, while PM2.5 had the least. NO2 and BC had the most daily variability, while PM2.5 once again had the least. CO2 and NO2 had the most hourly variability, while UFPs had the least. Finally, CO2 had the most seasonal variability while UFPs had the least.
3.3 Annual Averages
Estimated annual average pollutant concentrations across all monitoring sites are shown in Table 1 and Figure S14. There was roughly a 5- to 9-fold difference between the lowest and highest UFP site concentrations. On the other hand, NO2 and BC concentrations had a 6-fold difference across sites, PM2.5 had a 2-fold difference, while CO2 varied little across sites. The IQR to median ratio similarly indicated that UFPs, BC and NO2 had the highest relative annual average variability across site, while PM2.5 and CO2 had the lowest variability.
Pearson correlation coefficients for annual average concentrations across pollutants at the 309 monitoring sites ranged from as low as 0.27, indicating poor linear association, to 0.95, indicating very high linear association (SI Figure S15). PM2.5 and CO2 were generally poorly correlated with most other pollutants; BC and NO2 had moderate correlations with other pollutants; while individual UFP measures were strongly correlated with other UFP measures, moderately correlated with BC and NO2, and poorly correlated with PM2.5 and CO2.
The highest levels of overall air pollution (sum of the scores) were generally between north Seattle and the Seattle-Tacoma (Sea-Tac) International Airport, while the lowest levels were generally in the northeast (e.g., Woodinville, Bellevue) and southwest (West Seattle to Federal Way coastline) parts of the monitoring region (Figure 3 - Figure 4, SI Figure S16). Sites with high overall air pollution levels near downtown Seattle were generally driven by elevated BC and NO2 levels, while high concentration sites between Seattle and the Sea-Tac International Airport were generally driven by UFPs (SI Figure S17).
Lasso regression models indicated that the covariate most strongly associated with UFPs was proximity to the Sea-Tac International Airport, followed by proximity to major roadways (see SI Figure S18). BC was most strongly associated with rail yard proximity; while PM2.5 was associated with major road, airports and coastline proximity. NO2 was most associated with proximity to rail yards, large ports and coastlines. Finally, CO2 was most strongly associated with proximity to rail yards, medium ports, and major roadways.
3.4 Collocations at Regulatory Monitoring Sites
Median two-minute BC, NO2 and PM2.5 measurements from mobile monitoring stops were generally in agreement with measurements from regulatory sites (SI Figure S19). Annual average estimates from our mobile monitoring campaign measurements were similar to annual average estimates from comparable two-minute samples at regulatory monitoring sites used as collocations (SI Figure S20). These were in moderate agreement with true annual average concentrations at those sites (based on all of the available data during the study period).
4 Discussion
In this paper, we describe the design of an innovative mobile monitoring campaign specifically developed to estimate unbiased, highly spatially resolved, long-term TRAP exposures in an epidemiologic cohort. To date, this is one of the most extensive multi-pollutant mobile monitoring campaigns conducted in terms of spatial coverage (∼1,200 land km2), sampling density (309 monitoring locations along 9 routes, or 1 monitor every 3.9 land km2), and sampling frequency (7 days a week; 288 days over a one-year period) and duration (∼5 driving hours per day between the hours of 4 AM – 11 PM). The spatial resolution achieved by this campaign was significantly greater than would be expected from fixed regulatory monitoring approaches. We had one monitor per 3.9 km2 of land area rather than 183 km2 (6 regulatory sites in the monitoring area), almost a 50-fold increase. The average (SD) monitoring distance from an ACT cohort location was 611 (397) m rather than 5,805 (2,805) m, almost a ten-fold difference. Monitor proximity to prediction (i.e., cohort) locations is an important determinant of accurate exposure assessment. Additionally, we previously showed that the extensive temporal sampling of this campaign across hours, days of the week and seasons is expected to produce more accurate and unbiased annual average estimates as compared to more common campaigns with reduced sampling.27
A unique aspect of this campaign was the collection of stationary samples along the side of the road. While most other campaigns have only collected non-stationary, on-road samples, various studies have shown that mobile samples are generally higher in concentration than stationary samples.19,35–38 The completion of our stationary and non-stationary campaign positions us to conduct future work on how non-stationary data may be used responsibly for epidemiologic applications. Among the relatively few campaigns that have collected stationary rather than mobile samples alone, most have sampled for longer than two minutes (about 15-60 minutes per stop).39Our analyses indicated that shorter sampling periods produce comparably good estimates without adding excessive amounts of time to mobile monitoring campaigns (See SI Figure S2). Our use of a hybrid vehicle meant that the vehicle’s engine was off and running off battery during stop sampling periods, thus reducing the possibility of self-contamination.
ANOVA model results indicate differences across pollutants in terms of their spatial and temporal variability. This finding is particularly relevant for short-term mobile monitoring campaigns, which could design their campaigns to adequately capture the variability of the pollutants of interest. These findings suggest that repeated sampling at any given site is crucial since most of the variability for all measured pollutants was seen within sites. We observed a spatial variability for UFPs that was similar to that of NO2 but higher than more commonly measured pollutants including BC, CO2 and PM2.5. UFPs also had less temporal variability than all other measured pollutants. These findings suggest that UFP monitoring may benefit from higher than normal spatial coverage even if the temporal coverage is reduced as a result. The implementation of these principles for epidemiologic exposure assessment should translate to reduced exposure misclassification. Overall, these results are in line with past literature that has shown differing spatial and temporal contrasts across pollutants,40,41 though our work increases the robustness of these findings using a more spatially resolved, multi-pollutant dataset that includes less commonly measured UFPs.
The findings from this campaign demonstrate the region’s generally low air pollution levels. The range of concentrations across sites for PM2.5 (3.4-7.2 µg/m3) and NO2 (3.8-22 ppb) were well below the National Ambient Air Quality Standards (NAAQS) annual average levels of 15 µg/m3 and 53 ppb, respectively.42 Median UFP (∼3,000-10,000 pt/cm3) and BC (∼600 ng/m3) site concentrations were lower than what others have reported in cities throughout the world where mean study values range from roughly 6,000-64,000 UFP pt/cm3 and 400-14,000 BC ng/m3 (UFP19,36,37,43–57; BC17,19,37,46,47,52,57–69). While CO2 site concentrations (417-455 ppm) were above the 2019 global average of 412 ppb,70 they were in line with past work noting elevated carbon footprint levels in dense, high-income cities and affluent suburbs.71,72 Still, the high concentration variability seen across sites for pollutants like UPF, BC and NO2 (Table 1) suggest that future epidemiological analyses may have more power to observe health effects from these pollutants than those that are less variable, for example PM2.5 and CO2.
The similarity between BC, NO2 and PM2.5 measurements from our campaign and collocated regulatory monitoring sites (see SI Figure S19-S20) confirms that our campaign estimates were accurate. Some of the discrepancies between the two monitoring approaches may be due to differences in the sampling instrumentation, the exact sampling location, and quality assurance and quality control procedures. While we were unable to compare CO2 or UFP measurements to regulatory observations, duplicate instrument collocations generally showed good agreement. Additionally, CO2 instruments were regularly calibrated, and different UFP instruments were well correlated with each other.
As expected, we observed elevated annual average pollutant levels near aircraft, railroad, marine, and automobile sources. These findings are in line with past work.17–19,23,48,55,73–78 Interestingly, major roadway covariates were associated with all of the pollutants, but they were not the most significant predictor for any of the pollutants. This observed correlation in the data is likely because larger roadways have greater vehicle volumes and greater numbers of high emitter vehicles (e.g., medium and heavy-duty diesel trucks) compared to smaller roadways. Still, these results highlight the critical role that non-automobile source emissions play on the region’s air pollution levels. Discrepancies in the covariates most strongly associated with each pollutant (e.g., airport vs. rail yard proximity) reflect important regional sources and unique spatial patterns of these pollutants. UFPs, for example, were distinguished from other pollutants in that levels were strikingly elevated near the Sea-Tac International Airport. While other pollutants like BC, NO2 and CO2 all had elevated levels near rail yards, they each had additional geographic covariates that further characterized their spatial patterns. These findings reinforce the uniqueness of each pollutant and the importance of monitoring for each. This is particularly true for UFPs given the limited monitoring data available and their unique spatial and temporal patterns.
While UFPs are generally characterized as particles under 100 nm in diameter, this definition is not standardly defined and varies from study to study. In fact, much of the existing and accessible instrumentation does not efficiently measure particles below 10-20 nm, and it often includes particles above 100 nm. This concern was echoed by the EPA’s 2019 Integrated Science Assessment for Particulate Matter, where they noted the different size cuts used to define UFPs as an important limitation in their ability to synthesize findings from the literature.79 Since most particles by count are in the smaller size range with few above 100 nm, instruments that effectively measure smaller particles sizes measure total UFP counts more accurately. In our data, for example, instruments that measured particles as small as 10 nm in diameter (i.e., the NanoScan and DiSCmini) showed sharp spikes in UFP concentrations near the airport and downtown Seattle. Note that differences in UFP readings between the NanoScan and DiSCmini is likely related to the NanoScan being an optical and condensation particle counter (CPC) whereas the DiSCmini is an electrical mobility analyzer. CPCs can detect very small particles more efficiently than mobility analyzers.80 UFP measures from P-TRAKS, on the other hand, excluded particles smaller than 20 nm in diameter and reported attenuated spike concentrations. In fact, P-TRAK readings were generally lower than those from NanoScan and DiSCmini instruments, despite having a much higher upper size bound (420 and 700 nm vs 1,000 nm) because they missed a large number of small particles with diameters between 10-20 nm. Nonetheless, UFP measures from different instruments were strongly correlated with each other. This strengthens our confidence in the quality of our measurements and suggests that the use of different UFP instrumentation across studies could still produce useful exposure assessments.
Furthermore, the characterization of size-resolved UFP counts from multiple instruments is unique. We observed, for example, elevated levels of relatively smaller UFPs (approximately 20-36 nm) almost exclusively near the airport. Past work has similarly linked elevated levels of these smaller UFPs, which contribute greatly to total UFP concentrations, to airport activity.74,81 We are well-positioned to further explore UFP sources as well as size-specific exposure effects on epidemiological outcomes in future work.
Notably, other than different UFP measures, most pollutants were moderately or weakly correlated with one another. PM2.5 and CO2, in particular, were poorly correlated with other measures of more spatially variable, traffic-related air pollutants including BC, UFPs and NO2. Though UFPs were more strongly correlated with NO2 and BC than with PM2.5 levels, there were some deviations, for example, near the airport. These results highlight the importance of collecting long-term measurements of traffic-related air pollutants using exposure assessment designs with higher spatial resolutions than the designs typically used for more common pollutants, such as studies that rely on PM2.5 from regulatory monitoring sites. This may be particularly relevant for urban or other areas characterized by major emission sources such as airports or railroad systems, which may be important contributors to local and/or regional air pollution levels within an airshed. The differential spatial variability between these pollutants reinforces our interest in multi-pollutant exposure assessment. This is growing interest in the field of air pollution epidemiology,40,82–85 and something that we are positioned to make a meaningful contribution to in future work.
A feature of mobile monitoring campaigns is their reliance on repeated, short-term samples in order to achieve increased spatial coverage when compared to traditional long-term monitoring approaches. Since we collected about 29 two-minute samples per site (about an hour of data), we recognize that the resulting annual average site estimates are noisy. ANOVA models, for example, showed that most of the variability observed in the two-minute measurements occurred within sites rather than across sites or over time. Still, our work suggests that our annual average estimates should be within a 20-30% error on average (See SI Figure S2). Furthermore, mobile monitoring campaigns may benefit from calculating weighted annual averages (e.g., weights for weekday samples = 5/7, weekend samples = 2/7) given the generally small number of visits achieved per site, and since sampling may not be completely balanced despite efforts to do so.
Future mobile monitoring campaigns aiming to assess long-term estimates may be guided by the design and findings from this study, though we acknowledge that the unique geographical, meteorological and source characteristics of a given airshed may produce somewhat different results.
Finally, campaigns aiming to assess health outcomes with exposure periods of interest longer than a year should consider conducting longer or repeated campaigns to better characterize changing air pollution levels over time. Such monitoring, for example, may better capture natural disasters such as large wildfires, which are becoming increasingly important predictors of air quality during fire season.86–90 These did not meaningfully impact regional air quality during this study’s monitoring period, however.
The multi-pollutant approach of this campaign, which included multiple UFP measures, showed that the air pollution mixture varies throughout the region, with different pollutants dominating air pollution levels in different areas. While automobiles were important predictors of elevated air pollution levels, non-traffic sources such as the Sea-Tac International Airport and rail yards were also important sources in the region. UFP’s relatively low concentration variability by season, hour of day, and day of week with higher relative variability within- and between-sites distinguishes this infrequently characterized pollutant from more commonly measured pollutants such as PM2.5 or BC. Furthermore, while UFP readings from multiple instruments were well correlated, the variability in average levels across instruments suggests that a standardized UFP measurement approach would improve the consistency across studies. For example, a standardized measurement approach would benefit from the inclusion of particle sizes below 20 nm given their meaningful contribution to particle counts.
These results demonstrates that the design of this campaign captured spatial and temporal pollutant variations that can be explained by sensible land use features, including those related to traffic, along with seasonal and diurnal effects. These data will thereby be useful for assessing representative TRAP exposures in the ACT cohort. Next steps for this work include developing exposure prediction models with these data for the cohort and greater Seattle area. The rich dataset from this extensive campaign will additionally allow us to investigate the role of study design in terms of spatial and temporal coverage (e.g., number of sites, number of site visits) on exposure prediction models and subsequent epidemiologic studies.
Data Availability
The data presented in this work may be available upon request.
5 Acknowledgements
We are grateful to our two drivers, Jim Sullivan and Dave Hardie, for all of their efforts collecting these data, and to Brian High for building and supporting the database. This work was funded by the Adult Changes in Thought – Air Pollution (ACT-AP) Study (National Institute of Environmental Health Sciences [NIEHS], National Institute on Aging [NIA], R01ES026187), BEBTEH: Biostatistics, Epidemiologic & Bioinformatic Training in Environmental Health (NIEHS, T32ES015459), and the University of Washington Interdisciplinary Center for Exposure, Disease, Genomics & Environment (NIEHS, 2P30 ES007033-26). Research described in this article was conducted in part under contract to the Health Effects Institute (HEI), an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Award No. CR-83998101) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI, or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers.
6 References
- 1.↵
- 2.↵
- 3.
- 4.↵
- 5.↵
- 6.↵
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.
- 21.
- 22.
- 23.↵
- 24.↵
- 25.
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.
- 45.
- 46.↵
- 47.↵
- 48.↵
- 49.
- 50.
- 51.
- 52.↵
- 53.
- 54.
- 55.↵
- 56.
- 57.↵
- 58.
- 59.
- 60.
- 61.
- 62.
- 63.
- 64.
- 65.
- 66.
- 67.
- 68.
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.
- 76.
- 77.
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.
- 84.
- 85.↵
- 86.↵
- 87.
- 88.
- 89.
- 90.↵