Spatiotemporal profiles of ultrafine particles differ from other traffic-related air pollutants: lessons from long-term measurements at fixed sites and mobile monitoring

We use long term fixed site measurements along with extensive mobile monitoring data to evaluate the spatiotemporal correlation of UFP and NOx.


Introduction
Recent studies show that air pollution may be damaging to almost every organ in the human body. 1,2 Ultrane particles (UFP), aerosol particles with aerodynamic diameter <100 nm, are known to penetrate deep within the lungs and may enter the bloodstream and reach sensitive internal organs. 3,4 Furthermore, unlike larger particles, UFP can be deposited into the brain, causing adverse cognitive effects. [5][6][7] While consensus on the health effects of UFP, separate from ne particulate matter (PM 2.5 ), is yet to be reached, 8 there has been an increase in studies linking UFP exposure to cell damage, adverse cardiovascular health effects, and the formation of certain cancers (e.g., brain cancers). 5,[9][10][11][12][13] Due to these health risks, population exposure to UFP is the subject of current investigation in air pollution epidemiology. 14 However, UFP exposure is difficult to characterize due to the high variability in UFP concentrations over short temporal and spatial scales. As with many air pollutants emitted in urban areas, patterns of UFP are spatiotemporally complex: spatial patterns vary over time, and temporal patterns vary in space. However, there are few networks around the world that routinely measure UFP concentrations, typically with a small number of continuous monitors, [15][16][17] which are not capable of resolving the spatiotemporal dynamics of UFP across the urban landscape. Accordingly, health studies employ a range of alternative strategies for estimating spatial patterns of UFP exposure, including short-term distributed sampling, regional scale-air quality modeling, and land-use regression models based on short-term measurements. 18,19 In urban areas the principal sources of UFP include vehicular traffic (tailpipe emissions, brake wear, and tire wear), other combustion of fossil fuels, cooking, and nucleation events. [20][21][22][23] Because traffic is oen assumed to be a dominant source of UFP, exposure to UFP is sometimes approximated based on more commonly observed traffic-related air pollutants (commonly NO x or NO 2 ) as indicator values. [24][25][26][27][28][29] However, the strength of correlation between UFP and other traffic-related air pollutants (TRAPs, including NO, NO 2 , CO, and BC) varies among sites. [30][31][32][33][34][35] In this study we examine conditions under which the spatiotemporal signatures of UFP may differ meaningfully from those of other TRAPs.
Spatiotemporal variation of UFP is inuenced by a complex interplay between local sources, long-range sources, meteorological conditions, and aerosol dynamic processes. 36,37 Unlike other TRAPs, UFP concentrations are strongly affected by regional new particle formation (NPF) from atmospheric vapors. NPF events have been observed in urban, regional, and background environments. [37][38][39][40][41][42][43][44] In recent years, studies from various cities and background sites across the world show particle number (PN) concentrations, which are dominated by the ultrane particle number count, peak during periods with increased solar radiation. 16,[45][46][47][48][49][50][51][52][53][54] Short term studies that have investigated both particle formation and growth have found NPF to be an important contributor to overall UFP concentrations. [55][56][57][58] Brines et al. 21 studied multiple cities in the Mediterranean climatic regions (Barcelona, Madrid, Rome and Los Angeles) and found that although traffic remains the main source of UFP in urban areas, during high insolation (sunny) periods, NPF can become the main source of UFP. Under these conditions UFP concentrations can become decoupled from TRAP concentrations, which are driven primarily by emissions activity. However, most observational comparisons between UFP and other TRAPs are generally based on short-term mobile or distributed-sampler studies, 18,19,[59][60][61][62][63] usually not capable of comprehensively characterizing seasonal patterns.
Here, we combine two unique long-term observational datasets of particle number (PN) concentrations from the San Francisco Bay Area (USA) to investigate the spatial and temporal variation of ultrane particulate matter. We consider multiple spatial scalesfrom ne-scale variation within neighborhoods to a broad rural-to-urban gradientand investigate temporal variation at the diurnal and seasonal scales. Through these observations, we highlight conditions where UFP patterns show substantial deviation from those of other traffic-related air pollutants, likely resulting in a health-relevant divergence in patterns of exposure.

Materials and methods
This study combines long-term xed site measurements and onroad mobile monitoring measurements in the San Francisco (SF) Bay Area, California, USA. The SF Bay Area climate is temperate, with moderate winters and summers in the coastal areas and warm summer days in the inland valleys. Representative seasonal and diurnal proles of key meteorological parameters are presented in the ESI (Fig. S1 †). The xed sites and mobile monitoring measurements were spread across environments representing the varying levels of urbanization, traffic activity and composition, and other urban emissions activity.

Fixed sites
We incorporate hourly pollution monitoring data collected at four xed sites in the Bay Area Air Quality Management District's monitoring network over a period of 4-6 years (Jul-2011-Jan-2018). 64 The chosen sites represent different land uses and emissions intensity levels for the San Francisco Bay Area, including near-highway (Laney College), urban (Redwood City), suburban (Livermore), and rural (Sebastopol) sites. PN and NO x were measured at each of these four sites, CO at three sites (near-highway, urban, and rural), and BC at two sites (nearhighway and suburban). Details of xed sites (measurement period, pollutants measured, distance from nearest highway, and their location on a map) are presented in the ESI (Table S1 and Fig. S2 †). For the xed sites, PN concentrations were measured using condensation particle counters (CPC, TSI, model 3783, D p > 7 nm). NO x was measured using chemiluminescence analyzers (Thermo Scientic, model 42i). Black Carbon (BC) and Carbon Monoxide (CO) were measured using aethalometers (Teledyne, model 633, equivalent to a Magee Scientic model AE33) and gas lter correlation CO analyzers (Thermo Scientic, model 48i) respectively. We use data from 2015 (year with almost full coverage for measured pollutants at all sites) for calculating the annual average (Section 3.1). However, unless stated otherwise, we use all available hourly data (4-6 years) from the xed sites for the analyses presented in this study. For our analysis we dene daytime as 8 am-8 pm and nighttime as 8 pm-8 am. We dene summer as the rst day of June through the nal day of August and winter as the rst day of December through the nal day of February.

Mobile monitoring
To investigate time-stable trends with higher spatial resolution, we incorporate observations from a 32 month mobile monitoring campaign in the SF Bay Area 65 using two Google Street View cars equipped with the Aclima mobile measurement and data acquisition platform outtted with research grade equipment (Aclima Inc., San Francisco CA). To allow for a comparison of patterns by season, this analysis focuses on the portions of the driving domain most extensively sampled during the campaign: Downtown Oakland and West Oakland neighborhoods. West Oakland is comprised of low-and mid-rise residential/commercial neighborhoods, numerous small and mid-sized industries and warehouses, is adjacent to a port, and is surrounded by major highways. Downtown Oakland has mixed residential and commercial zoning with mid-and highrise construction. Overall, we collected $1125 hours of mobile monitoring data on 50 kilometers of roads within these two neighborhoods, with a range of 1-133 repeated visits to each road segment on 1-54 unique days over this 32 month period (May-2015-May-2017). The mobile monitoring campaign employed fast response lab-grade instruments. Ultrane particles were measured using CPCs (TSI, model 3788, D p > 2.5 nm), NO using chemiluminescence (Model CLD64, Eco Physics AG, Switzerland), and NO 2 was measured using cavity-attenuation phase-shi spectroscopy (Model T500U, Teledyne Inc., San Diego, CA). The CPCs used for mobile monitoring campaign had a cut point of 2.5 nm, compared to the 7 nm cut point of the xed sites CPCs (we do not compare concentrations between xed sites and mobile monitoring in this study). The mobile monitoring platforms had separate inlets for particle and gas measurements, with particle inlets designed to minimize diffusional sampling losses. To minimize the inuence of selfemissions on the measured pollutant concentrations, these collocated inlets were positioned in a forward-facing orientation several inches above the roof line at the rear edge of the front window of the cars. Details of the mobile monitoring setup have been presented in Apte et al. 65 and its ESI. † Data processing for mobile monitoring followed the steps described in Messier et al.: 66 road line geometry data for the San Francisco Bay Area were obtained from OpenStreetMaps (OSM) and converted to point geometry at 30 m spacing, corresponding to the midpoint of individual 30 m road segments. Each measured 1 Hz data point was 'snapped' to the coordinates of the nearest road segment. Data were collected at >1600 total road segments in West and Downtown Oakland, California. The road segments were designated as 'highway', 'arterial', or 'residential' road data based on OSM classication codes. For seasonal or long-term spatial patterns, an additional data reduction technique was applied to ensure that each repeated drive through a given road segment (drive pass) was represented equally. 66 First, measurements for each 'drive pass' were reduced into a single drive pass mean concentration value. The median of drive pass means at each road segment was used as the core metric for mobile monitoring spatial analyses. These analyses exclude road segments with data from fewer than 5 sampling days as the small sample size limits statistical condence in concentration estimates at those locations. For logistical reasons, mobile measurements were restricted to weekday and daytime conditions. Given the aforementioned sample-size considerations, the mobile measurements are capable of resolving seasonal-average spatial patterns for typical daytime conditions, but lack the granular temporal resolution of continuous xed-site measurements. Accordingly, our mobile measurements capture the seasonal aspects of spatiotemporal variability, but do not reect the marked difference in TRAP spatial patterns that occur between day and night that have been identied by prior mobile transect studies. 67

Supplementary datasets
We use mesoscale (regional) meteorological data for wind speed (10 m from ground) and planetary boundary layer height (PBLH) to calculate the regional-scale ventilation coefficient (VC ¼ wind speed Â PBLH). These data for the SF Bay area were obtained from NASA's meteorological reanalysis dataset, MERRA2. 68 MERRA2 has a spatial resolution of 0.5 Â 0.625 (55 km Â 60 km) and an hourly temporal resolution. Given the large area covered in a single MERRA2 grid, the grid we chose covered most of the sites including the near-highway, urban, and suburban site.
For the near-highway site, we obtained traffic data (vehicular and truck ow) from the Freeway Performance Measurement System maintained by the California Department of Transportation. 69 We chose the traffic sensor closest to our nearhighway xed site (Mainline VDS 400218). Both the traffic sensor and the xed site were adjacent to highway I880-N. For our analysis, we used hourly traffic data for 2015 and monthly data for 2014-2018.

Temporal variation of TRAP concentrations
Among the xed sites, TRAP concentrations were highest for the near-highway site, followed by the urban, suburban, and rural sites, consistent with patterns observed in other areas. 70 For 2015 (year with almost full coverage for measured pollutants at all sites), from near-highway to rural, the annual average (5th-95th percentile) PN concentrations were 29 900 cm À3 (7960-65 500 cm À3 ), 11 900 cm À3 (1850-31 100 cm À3 ), 10 100 cm À3 (1990-21 100 cm À3 ), and 3500 cm À3 (430-10 500 cm À3 ) respectively. Annual average NO x concentrations followed the same order -34.7 ppb (7.1-87.7 ppb) for the near-highway, 18.8 ppb (2.3-61.1 ppb) for the urban, 17.4 ppb (1.9-63.4 ppb) for the suburban, and 8.4 ppb (1.2-28.6 ppb) for the rural site. Among these sites, BC was only monitored at the near-highway and the suburban sites for which the annual average concentrations were 1.43 mg m À3 (0.28-3.65 mg m À3 ) and 0.78 mg m À3 (0.09-2.61 mg m À3 ) respectively. CO annual average concentrations were 0.47 ppm (0.25-0.86 ppm) at the near-highway site, 0.44 ppm (0.23-0.92 ppm) at the urban site, and 0.38 ppm (0.20-0.66 ppm) at the rural site. The low CO concentrations and the small differences among sites are consistent with the large reductions in vehicular CO emissions ($80-90%) over the last few decades in urban areas in the US. 71 Pollutant concentrations exhibited high seasonal and diurnal variability. In Fig. 1, we present the diurnal and seasonal concentration proles from each xed site. We also separate weekdays and weekends since vehicular traffic (especially truck traffic) is generally lower on the weekends. For example, at the near-highway monitoring site in Oakland, CA, total traffic volumes on Interstate 880 were 13% higher on weekdays than weekends (ESI Fig. S3 †). For the same time period and site, heavy duty truck traffic volumes were 150% higher on weekdays than weekends. The lower weekend heavy duty truck traffic is consistent with the decreased activity at the nearby port of Oakland during the weekends (the port is closed on the weekends).
Average diurnal proles indicate the combined inuence of the traffic activity and ventilation patterns. In the winter, all sites showed a strong peak for all pollutants (PN, NO x , BC, and CO) during morning and evening rush hours, with concentrations 2-7Â greater than the mid-day trough in concentrations ( Fig. 1). In the ESI (Fig. S1 †) we present the diurnal and seasonal averages for the ventilation coefficient (and other meteorological parameters) in the SF Bay Area. The lower wind speeds and mixing height during mornings and evenings of winter months resulted in lower ventilation. Conversely, owing to higher wind speeds and mixing height, warmer periods were generally more ventilated with the summer mid-days having the highest ventilation coefficient. Furthermore, some of the highest solar radiation was also observed during the summer daytime. The midday trough observed in all diurnal pollution proles except those of summertime PN reects the strong effect of increased atmospheric dilution during the middle of the day coupled with a reduction in traffic volumes on many urban roads outside of rush hour. Substantial weekday-weekend differences in these peaks demonstrate their dependence on weekday traffic. Winter NO x concentrations generally exceeded summer concentrations by 80-300%, reecting increased dilution in summer months coupled with potential reduction in combustion sources. Rush hour concentrations were greatly reduced in summer months, especially during early evening hours, possibly because the higher boundary layer during the summer evenings caused greater mixing and dilution of traffic emissions, while winter commutes oen take place when the boundary layer height is shallower.
Summertime PN patterns signicantly diverged from those of other TRAPs. Unlike in winter conditions where NO x and PN peaks aligned during both mornings and evenings, summer patterns at all sites showed a mid-day PN peak that does not correspond to a peak in any other TRAP. This increase in PN without concomitant increases in other products of primary combustion, occurring during high-insolation midday hours (10 am to noon) and independent of weekend/weekday traffic differences, strongly suggests new particle formation (NPF). The summer daytime peak PN concentrations for non-near-highway sites were $3Â greater than the concentrations observed during the rest of the day. At the near-highway site, high midday concentrations resulted in high concentrations throughout the day, without distinct peaks corresponding to morning and evening traffic-related sources. For winters, all sites except the near-highway site had early morning ($6 am) and evening peaks ($6 pm) for all pollutants (PN, NO x , BC, and CO). For example, the winter weekday morning PN concentration peaks were 1.4-1.6Â higher higher than the midday troughs. For these non-nearhighway sites, morning PN peaks were 1.5-2.0Â higher on weekdays than on weekends on average; evening winter peaks differed only slightly between weekdays and weekends (within 10%).
To highlight the diverging temporal patterns of PN and NO x , we computed the PN/NO x ratio. If a common primary source drives the concentrations of both PN and NO x in the urban environment, and both species have approximately similar lifetimes, then we would not expect the PN/NO x ratio to have strong time-dependence. Were these conditions met, NO x could serve as a proxy for PN. However, contrary to that assumption, we show that there is a strong mid-day enhancement in PN to NO x . In Fig. 2 we present the xed-site diurnal variation in the PN/NO x ratio for winter and summer, and also separated by weekday-weekend. For all xed sites, the PN/NO x ratio is highest during the summer daytime. Depending on time of day, the PN/ NO x ratio for the near-highway site was 1-3Â higher for summer than for winter. The largest difference between winter and summer PN/NO x ratio was observed for the urban and the suburban site with PN/NO x ratio between seasons ranging from 1-5Â for urban and 1-5Â for suburban. For the rural site the PN/NO x ratio was 1-4Â higher during the summer compared to winter. Furthermore, summer PN/NO x were markedly higher on weekends which is consistent with the assumption that reduced weekend traffic results in lower concentrations of both NO x and associated directly-emitted PN, further accentuating the relative contribution of PN associated with NPF. It was notable that the absolute increase in the average weekend daytime PN peak concentrations from winter to summer was the highest for the near-highway site (+29 900 cm À3 ) as compared to the urban (+13 800 cm À3 ), suburban (+9400 cm À3 ), and the rural (+3000 cm À3 ) sites. The larger change in the absolute PN concentrations for more urban and generally polluted sites could be indicative of the role of locally-emitted semivolatile and intermediate precursors as contributors to daytime PN peaks during the summer months. [72][73][74] We present in Table 1 the Spearman correlation (r s ) matrix among the TRAPs for the near-highway, urban, suburban, and the rural sites based on hourly-average concentrations. At all sites, the correlation between PN and any other TRAP was 0.41-0.76. The inter-pollutant correlation among non-PN pollutants were generally higher compared to PN. NO x and BC were highly correlated at the near-highway site (r s ¼ 0.91) and at the suburban site (r s ¼ 0.91). NO x and CO were well correlated at the near-highway site (r s ¼ 0.70) and the urban (r s ¼ 0.81) site. Furthermore, for all the sites, PN concentrations were similarly correlated with NO x (r s,near-highway ¼ 0.72, r s,urban ¼ 0.66, r s,suburban ¼ 0.49, r s, rural ¼ 0.69) than either NO (0.76, 0.59, 0.44, 0.56) or NO 2 (0.57, 0.66, 0.52, 0.68). For the suburban and rural sites, among the NO x species, NO was least correlated with PN and NO 2 and NO x were similarly correlated to PN. The lifetime of NO x exceeds of that of NO and NO 2 : oxidation of NO to NO 2 is one of the most rapid daytime NO sinks, and that photolysis of NO 2 to NO is a rapid sink daytime sink of NO 2 . While interconversion of NO and NO 2 occur at the timescales of a few minutes, NO x has a photochemical lifetime of 2-4 h. 75 Fig. 2 Average diurnal variations for PN/NO x ratio for near-highway, urban, suburban, and rural sites by season (summer and winter). Separated by weekday and weekend to illustrate role of changing traffic volumes. The values presented are the ratio of the averages, and not the average of the ratios. All available hourly data of TRAP concentrations from Jul-2011 to Jan-2018 was used for these calculations. To illuminate whether the low overall hourly correlation of PN and NO x was driven by seasonal factors, we performed an additional set of correlation analyses of hourly data stratied by season (Table 2). Lower pairwise correlation between PN and other TRAPs was driven by differences in summertime patterns -PN and NO x were least correlated during the summer daytime. For the non-near-highway sites, the summer daytime r s values were 0.01-0.53, suggesting that particularly during periods of higher photochemical activity, the dominant inuence on NO x concentrations (traffic and other fuel combustion) differed from the dominant inuence on PN concentrations (new particle formation from nucleation) during this period which had higher photochemical activity. The linear t of weekly averaged NO x and the ventilation coefficient was higher than PN and the ventilation coefficient for all sites (ESI Fig. S4 †). While the R 2 values for NO x and the ventilation coefficient ranged between 0.19-0.51 across the xed sites, the R 2 between PN and the ventilation coefficient ranged between 0.00-0.34.
To further illustrate the dynamics of PN against other TRAPs at a higher temporal resolution, we developed a set of heatmaps representing the full timeseries of PN, NO x , and BC measurements with each day of the year (x-axis) divided into hourly concentrations (y-axis). Fig. 3 presents heatmaps for the suburban site, with the heatmaps for all sites presented in the ESI (Fig. S5-S8 †). This visualization clearly illustrates how the diurnal prole of PN concentrations tracks the diurnal cycle of other traffic related air pollutants during winter months, and decouples from the TRAPs in other seasons. While the daytime PN concentration peaks are most apparent in the peak summer months (June-August), some daytime PN peaks can also be observed in April, May, and September. Based on this year-long heatmap, the months between October and March can be classied as the "low-NPF" season as compared to the summer which is the "high-NPF" season. To maximize the amount of mobile monitoring data that we can include for analysis (and thus improve our analytical precision and spatial coverage), we therefore use these low-NPF and high-NPF designations in our core analysis of mobile monitoring data. Sensitivity analyses presented in the ESI † demonstrate strong agreement in the spatial and temporal patterns of data between the low-NPF and winter periods, and between the high-NPF and summer periods (Fig. S9-S12 †).

Spatiotemporal variation of PN and NO x : spatial patterns vary by season
In Fig. 4 we present the road-segment daytime mean concentrations of PN and NO x estimated on the basis of mobile monitoring in West Oakland and Downtown Oakland. Consistent with xed-site measurements, these maps show the opposing seasonal patterns for PN and NO x . The on-road measurements were mostly made during the daytime making them more sensitive to photochemically-driven NPF. While onroad concentrations of PN increased from the low-NPF winter months to the high-NPF season, on-road concentrations of NO x decreased from the low-NPF to high-NPF conditions. Average on-road NO x levels decreased by a similar proportion for all road types: 29% for residential roads, 27% for arterials, and 27% for highways. This distributed decrease in NO x concentrations is consistent with the higher ventilation during the high-NPF season (summer). However, PN concentrations increased from low to high-NPF season for all road types. The increase in median PN levels from low-to high-NPF season was relatively lower for highways (+8900 cm À3 , +24%), compared to the dramatic increase observed on arterials (+16 000 cm À3 , +64%), as well as residential roads (+18 800 cm À3 , +84%). While the spatial variation in NO x remained consistent between seasons, we see a decrease in the spatial variability of PN. As shown in Table 3, during the high-NPF season the interquartile range of concentrations on each road type was smaller than during the low-NPF season. We did not nd evidence of such a trend for Table 2 Spearman correlation (r s ) between PN and NO x for nearhighway, urban, suburban, and rural sites calculated by season (summer, winter, and all seasons) and time-of-day (daytime, nighttime, and entire day). All available hourly data of TRAP concentrations from Jul-2011 to Jan-2018 was used for these calculations NO x . This relatively spatially uniform increase in PN levels during the high-NPF season suggests a potentially large contribution of nucleation to PN concentrations even for onroad concentrations during periods with high insolation. We show maps of the PN/NO x ratio to indicate road-types and times when the PN levels deviated sharply from NO x . To make our analysis less sensitive to outliers in instantaneous PN and NO x mobile measurements, we computed the ratio of the road-segment median concentrations. In Fig. 5 we present the spatiotemporal and seasonal variation in the PN/NO x ratio. In general, the PN/NO x ratio for on-road concentrations was higher for the high-NPF season daytime compared to the low-NPF season. The seasonal difference was the highest for residential roads with 1-4 pm PN/NO x ratio $6Â higher for the high-NPF season compared to the low-NPF season. For the arterial roads, the high-NPF season PN/NO x ratio was $4Â higher than low-NPF season between 12-5 pm. The seasonal difference in the high-NPF season and low-NPF season PN/NO x ratio was the least for the highways ($2Â higher than low-NPF season between 11 am-5 pm). The PN/NO x ratio was highest for residential streets since they have the lowest NO x concentrations and the PN concentrations would be less spatially variable during NPF events (a regional phenomenon). Overall, the diurnal proles of the PN/NO x ratio from on-road  measurements corroborate the ndings from the xed sites (Fig. 2), and emphasize how the seasonal divergence in this ratio is most strongly observed in locales that are relatively less inuenced by direct traffic emissions. As with xed site hourly data, time-averaged road segment concentrations show lower correlation summertime on all road types during the summer (Table S4 †), with a particularly notable decrease on residential roads.

Conclusions
UFP levels are governed by an interplay between proximity to sources such as traffic, diurnal and seasonal ventilation changes, and new particle formation from nucleation. By using long term measurements from xed-sites and highly spatially resolved on-road data, we were able to analyze spatiotemporal variation of PN concentrations. We observed daytime peaks in PN concentrations at multiple sites during the warmer months that were not observed for other primary traffic-related pollutants. In approximate terms, we observed a $2Â increase of PN concentrations during mid-day hours relative to the morning rush hours, while NO x and other TRAP pollutant concentrations typically dropped by $2Â over the same period. We take this nding as further evidence that NPF can complement traffic as a major source of ambient PN. In very rough terms, this nding implies that for the half of the year where NPF is common in the SF Bay Area, approximately half or more of the PN concentrations might be attributed to new particle formation during the peak hours for this photochemical process. Because the spatiotemporal variation in NO x concentrations differs from PN, using NO x (or other traffic-related air pollutants) as a proxy for PN (or UFP) concentrations could result in inaccuracies in estimating UFP exposure. These ndings may have particular relevance for high insolation urban areas where NPF can contribute to a large fraction of UFP concentrations. 21 It should also be noted that while PN concentrations are generally a good proxy for UFP concentrations in urban ambient air in the USA, this assumption does not hold for extremely polluted environments where a large fraction of the particles can be larger than 100 nm. 44,76 Long-term xed site and mobile monitoring measurements can advance understanding of the spatiotemporal patterns of various air pollutants. 65,77 While highly resolved spatio-temporal measurements of particle size distributions may be unlikely, a compromise such as the collection of long term measurements of size distributions along a gradient of xed site locations may help us understand the role of primary emissions vs. NPF in contributing to UFP concentrations. Furthermore, recent advances in understanding semivolatile and intermediate volatility precursors, and the emerging recognition of the pivotal role of volatile chemical emissions 72 in urban reactive chemistry might suggest that this issue merits detailed investigation from a chemically resolved perspective as well.
Our ndings should inform future assessment modeling of urban UFP exposure: while traffic activity and the location of major roadways are governing factors in temporal and spatial patterns of TRAP concentrations (including UFP), the effect of NPF during sunny/warmer seasons substantially alters both temporal and spatial UFP trends. The decoupling of trends between UFP and other TRAPs could result in inaccuracies in exposure modeling. First, if the model assumes or implies a xed ratio between UFP and other TRAPs (i.e., PN/NO x ratio) based on a single season of sampling, it may signicantly misrepresent UFP patterns. The ratio is highly seasonally dependent, as well as dependent on time-of-day. Second, if the model overweights the importance of roadway proximity or traffic activity, the exposure in residential or far-from-road areas would be substantially underestimated for summer months. Activity-based exposure models, which integrate exposure over a range of locations reecting an individual's daily activities, would be particularly impacted by assuming NO x is an appropriate surrogate for UFP. For example, higher UFP concentrations during the daytime can imply higher exposure during this time when people are more likely to be outdoors. NPF can contribute to substantial UFP concentrations and the contribution can be higher for locations with higher concentrations of UFP precursors. There is growing evidence of the adverse health effects of UFP. A nal public health implication of these ndings is that reductions in exposure to UFP cannot rely only on reducing direct emissions, they must also depend on the reduction of UFP precursors.

Conflicts of interest
There are no conicts to declare. ML is employed by Aclima, Inc. Research described in this article was funded by the Health Effects Institute, an organization jointly funded by the United States Environmental Protection Agency (assistance award no. R-82811201) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reect the views of Health Effects Institute, or its sponsors, nor do they necessarily reect the views and policies of the United States Environmental Protection Agency or motor vehicle and engine manufacturers.