Air pollution is a major environmental threat for urban populations, affecting the health of 9 out of 10 urban residents (World Health Organization, 2018). Within urban environments, locally high concentrations of air pollutants are common (Strosnider et al., 2017). As populations continue to migrate to urban areas (United Nations, 2019), air pollution will prove a persistent and growing threat to urban populations; this threat is pronounced for subsets of the population who are of lower socioeconomic status or are physiologically vulnerable to air pollutants due to age or pre-existing conditions, as these groups are disproportionately impacted by the negative health impacts of air pollution (Perlin et al., 1999; Zhou et al., 2011; Gray et al., 2013). In order to attenuate negative health impacts of air pollution in the future, it is imperative that we accurately assess the spatiotemporal distribution of air pollution in urban environments. Comprehensive air pollution monitoring is crucial to understanding where, and how, to focus efforts to attenuate air pollution and its associated health risks in cities equitably.

PM consists of heterogeneous mixtures of organic (Tsapakis et al., 2002) and inorganic components (Kelly and Fussell, 2012) that vary in size, shape, composition, and origin within the urban environment (City of Philadelphia, 2019; Shakya et al., 2019). Coarse size fractions (PM_2.5–PM₁₀) of PM largely originate from crustal sources, whereas fine PM (PM_0.1–PM_2.5) derive mainly from industrial emissions, non-renewable power generation, and vehicle exhausts (United States Environmental Protection Agency, 2015). BC is a major component of PM that results from the incomplete combustion of fossil fuels and other organic matter. As such, the presence of BC is often used as an indicator of urban traffic pollution (Targino et al., 2016). Quantifying the abundance and distribution of various PM sizes in urban environments is of particular interest to public health (Dominici et al., 2006), as prolonged exposure to PM is associated with increased rates of mortality (Dockery et al., 1993); smaller particles, especially those within finer size fractions, can easily deposit in the lungs (Miller et al., 1979), leading to a number of negative health outcomes including reduced lung function (Shakya et al., 2016), asthma (Rabinovitch et al., 2006), cardiovascular and respiratory disease (Paul et al., 2019), and pathogen exposure (Cao et al., 2014; Stewart et al., 2021).

Many studies have investigated urban air quality, but such studies typically rely on a small number of stationary points of measurement (Vallius et al., 2005; Zhang and Cao, 2015) and interpolation (Burke et al., 2001; Zhang et al., 2013) to characterize air pollution across an entire city. While these methods are effective at capturing temporal trends in local pollutant concentrations, they are less able to capture fine-scale spatial variation in air pollution throughout urban environments. This is crucial in urban areas, where emissions sources are not uniformly distributed and pollutant dispersal patterns change over short distances due to differences in vegetation (Brantley et al., 2014; Gromke & Ruck, 2012; Salmond et al., 2013; Xing & Brimblecombe 2019) and urban structure (Abhijith and Gokhale, 2015; Gallagher et al., 2015; Hagler et al., 2012). In recent years, mobile monitoring has emerged as a novel method with which to study the spatial and temporal distribution of air pollutants (Van Poppel et al., 2013; Deville Cavellin et al., 2016; Targino et al., 2016; Apte et al., 2017; Shakya et al., 2019; Sm et al., 2019; deSouza et al., 2020). As mobile monitoring methods are capable of collecting data at finer spatial scales than is feasible with stationary monitoring (Van den Bossche et al., 2015; Shakya et al., 2019), mobile monitoring is capable of providing more accurate and meaningful information about air quality in Philadelphia and can identify areas where pollutants with detrimental health impacts pose the greatest risk to human health.

In this study, we employ vehicular mobile monitoring of PM across 24 different size fractions between 0.25–10 μm and BC throughout the urban landscape of Philadelphia, Pennsylvania during the summer of 2019. Mobile monitoring allows us to observe the spatiotemporal distribution of air pollutants and discern patterns in variation at a fine spatial scale (Gozzi et al., 2016), and in doing so, identify times and locations in an urban environment such as Philadelphia where high concentrations of air pollutants are prevalent. Using Getis-Ord Gi* hot spot analysis, we identify where and when hotspots of air pollution occur to determine where statistically significant high air pollutant concentrations appear consistently. With a more fine-scale, holistic assessment of urban air pollution, cities can determine where air quality improvements are most needed, and better develop strategies intended to reduce ambient air pollution through future urban planning and development.

The average daily mean concentrations observed throughout the measurement period were 11.55 ± 5.34 μg/m³ for PM₁, 13.48 ± 5.59 μg/m³ for PM_2.5, 16.13 ± 5.80 for µg/m³ PM₁₀, and 1.56 ± 0.39 μg/m³ for BC. The annual recommended mean for PM_2.5 was exceeded on 8 of 12 sampling days, while the annual recommended mean for PM₁₀ was exceeded on 3 of 12 sampling days; throughout the data collection period, the average daily mean for PM_2.5 exceeds the recommended mean, while the average daily mean for PM₁₀ is within one standard deviation of exceeding the recommended mean (Figure 2). Observed PM and BC concentrations had roughly Poisson distributions (Figures 3A–D); lower concentrations were observed much more frequently than higher concentrations. The lowest concentrations of PM and BC in Philadelphia are found in Philadelphia’s Lower North (LNO), West Park (WP), and West (W) planning zones. The highest concentrations of PM across all size fractions are found in Philadelphia’s North Delaware (NDEL), River Wards (RW), and North (NOR) planning zones (Figures 3A–C), while BC concentrations are highest in the RW, Lower Far Northeast (LFNR), and Upper Far Northeast (UFNE) planning zones (Figure 3D). We found that fine PM_2.5 comprises approximately 83.6% of the observed PM₁₀ in Philadelphia, while BC accounts for about 11.6% of fine PM_2.5 in Philadelphia (Figure 4). BC was strongly correlated with PM_2.5 concentrations at the multivariate level when considering their relationships across all days (Figure 5B, Procrustes, m² = 0.9249, p = 0.043), and significant (p < 0.05) positive correlations between BC and PM_2.5 were observed on 10 of the 12 days of data collection (Supplementary Figure S2). Among days where we found a significant correlation between PM_2.5 and BC, weak to moderate relationships were observed (Supplementary Figure S2, Spearman’s ρ: 0.215-0.616).

Statistically significant hotspots were found on all days across all measured size fractions of PM and BC throughout Philadelphia. The average concentrations of hotspots (Figure 6) within a given cell ranged from 8.7 ± 4.6 μg/m³ for BC; 18.7 ± 7.1 μg/m³ for PM₁; 28.0 ± 8.8 μg/m³ for PM_2.5; and 46.0 ± 17.3 μg/m³ for PM_10. All pollutants measured had hotspots that exhibited a tendency to recur in the same locations across multiple days; hotspots for PM₁ appeared in the same cell on as many as six separate days throughout the data collection period, while hotspots for PM_2.5, PM₁₀, and BC appeared in the same cell on up to five different days.

Concentrations for all pollutants exhibited variation from day-to-day, with variation resulting from differences in weather conditions, traffic conditions, and slight deviations from the planned routes resulting from road closures (Figures 2, 7; Supplementary Tables S3–S6). All observations indicate the presence of PM 5 µm in diameter or smaller. PM exceeding 5 µm in diameter is not as ubiquitous throughout the data collection period, with larger particles not being detected at times throughout each day. Mass values observed for particles 1.6 µm diameter and larger generally demonstrated the greatest variation throughout each day, with particles with a diameter 0.5 µm and smaller also showing less within-day variation (Figure 7). The number, duration, and timing of PM_2.5 hotspots (Figure 7) varied from day to day; however, they were most consistently seen from 8:00–9:00 AM (30.19% of all hotspots) in complement with other studies where air pollutant concentrations peak during the morning hours (Zhao et al., 2009; Tunno et al., 2012).

Air pollution in Philadelphia is dominated by fine PM (PM_2.5). The PM_2.5/PM₁₀ ratio observed throughout the data collection period (83.6%) marks an increase relative to a stationary PM monitoring study that found that PM_2.5 only comprised 75% of PM in Philadelphia during the summers of 1992–1993 (Burton et al., 1996). The dominance of fine PM is reflected in the number of days that pollutant concentrations exceed their maximum annual recommended concentrations; while PM₁₀ exceeds recommended concentrations (20 μg/m³) on just 3 of the 12 sampling days, PM_2.5 exceeds its recommended concentration (10 μg/m³) on 8 sampling days. The prevalence of fine PM in Philadelphia suggests that anthropogenic sources are becoming increasingly large contributors to urban air pollution. Part of this increase may be due to the nature of on-road sampling itself; areas closer in proximity to roads tend to have higher concentrations of fine PM and lower concentrations of coarse PM relative to background sites due to the prevalence of traffic-related emissions (Barzyk et al., 2009; Karner et al., 2010; Yu et al., 2016). The increased proportion of finer PM may also result from changes within Philadelphia over time, such as urban development or increased traffic, that attenuate and/or remove potential sources of coarse PM while increasing potential sources of fine PM.

As BC and PM_2.5 are significantly correlated, the proportion of BC to overall fine PM in the air is helpful in determining the role of combustion processes in local pollution trends (Ni et al., 2014; Kim et al., 2017; Xu et al., 2017), which can in turn offer valuable information about potential sources of air pollutants in specific areas of urban environments. While the relationship between BC and fine PM in Philadelphia varies spatially (Figure 4) and temporally (Supplementary Figure S2), the overall BC/PM_2.5 ratio (11.6%) is comparable to BC/PM_2.5 ratios observed in other major cities, which range from 5–20% (Yu et al., 2015) and references therein). BC makes up a greater proportion of the measured PM_2.5 (Figure 4) in the North Delaware (NDEL) planning district, where heavy traffic conditions, proximity to major roads such as Interstate 95 (I-95), and local industry likely contribute to elevated BC concentrations. BC also makes up a significant fraction of PM_2.5 in the areas surrounding the Northeast Philadelphia Airport in RFNE and The Philadelphia International Airport in LSW, as landings and takeoffs by aircrafts at airports have been shown to increase local BC concentrations in the atmosphere (Agarwal et al., 2019). Pairwise comparison of BC and PM_2.5 concentrations revealed that the relationship between the two pollutants was generally variable from day to day (Supplementary Figure S2); variation in this relationship from day to day is due in part to the heterogeneity of emission sources and the urban landscape (Van den Bossche et al., 2015).

PM₁, PM_2.5, and PM₁₀ are spatially heterogenous and have similar spatial distributions throughout Philadelphia (Figure 3). The similarities that the PM_2.5 and PM₁ distributions share with the PM₁₀ distributions further demonstrate that most PM variation in Philadelphia can be characterized by changes in fine PM. Likewise, while there is slight variation in the location of hotspots among the different pollutants, the overall spatial distribution of hotspots throughout Philadelphia is similar across the three PM size fractions and BC (Figure 5). The areas with the highest frequency and concentration of hotspots are highly trafficked areas that are home to large public utility properties and heavy industry, which are likely significant sources of PM in this area (Philadelphia City Planning Commission, 2021). Relatively few hotspots were found in northern and western Philadelphia, which are located well outside of Philadelphia’s urban core. More open spaces have been shown to have a positive impact on air quality (Merbitz et al., 2012), and a previous study on air pollution in Philadelphia revealed associations between open spaces and low PM concentrations (Shakya et al., 2019). Traffic and industrial activity are less prominent in these areas relative to Philadelphia’s urban core, and a greater abundance of vegetation may curb air pollution primarily by uptake via leaf stomata and particle deposition (Beckett et al., 2000; Nowak et al., 2006).

The recurrence of hot spots in specific locations suggests that there are areas in Philadelphia where pollutant concentrations are consistently elevated relative to the surrounding area. A notable cluster of cells in the NDEL, RW, and NOR planning zones contain high concentration PM hotspots across multiple days. Other clusters of recurring hotspots are found within the University Southwest (USW) and LFNE planning zones. These hotspots are likely attributed to primary particles emitted from morning rush-hour traffic, where the number and density of vehicles on the road is higher relative to the rest of the day. The presence of hot spots outside the morning trend may be attributed to areas closer to industrial sites as in Chow et al., 1994, such as those located along the Interstate 676 (I-676) and I-95 corridors. While a substantial fraction of PM in urban areas originates from combustion engines, increased solar radiation during the summer months likely enhances the contribution of secondary particles formed from photooxidation of precursor molecules (Claeys et al., 2004; Shakya and Griffin, 2010) during the afternoon.

Temporal trends are evident in the concentration and distribution of PM₁, PM_2.5, and PM₁₀ from day to day (p < 0.05, Supplementary Tables S3–S5). Conversely, BC concentrations did not exhibit strong temporal variation (Supplementary Tables S6). The PM_2.5 size fraction (Figure 5D) clustered into two distinct time periods separated at approximately 11:08 AM, which complements our finding of BC clusters at approximately 10:56 AM (Figure 5C). Larger (PM₁₀) and smaller size (PM₁) fractions varied in their separation of peaks by time. PM₁ displayed less discrete temporal clustering (Figure 5E), with a break in clustering at approximately 10:08 AM. A cutoff was not found for PM₁₀. The lack of temporal clustering for PM₁₀, as seen in our results as well as a previous mobile monitoring study (Peters et al., 2013), affirms that coarse PM emission is relatively stochastic throughout the day, and may be attributed to the prevalence of crustal sources (e.g. dust resuspension) in areas that were sampled. These results complement findings in (Hankey and Marshall 2015), where PM size fractions were found to exhibit different concentrations in the morning and afternoon.

In this study, we demonstrate the potential for a mobile monitoring approach to assess the fine-scale spatiotemporal distribution of air pollutants throughout a major city. Our findings demonstrate the heterogeneity of PM and BC concentrations in space and time and show that finer PM comprises the majority of PM pollution in urban environments. Thus, measures that target fine PM emissions are paramount to reducing exposure to air pollution for residents of Philadelphia and other urban environments. Regulations and other efforts geared toward lowering PM emissions should be implemented on a city-wide scale, though hot spot analysis of PM and BC data reveal specific locations in Philadelphia (North, North Delaware, and River Wards planning districts) and times (morning hours) at which air pollution poses the greatest and most persistent threat to human health and wellbeing. These planning districts and the areas surrounding them merit further study throughout the year; as the recommended annual mean PM_2.5 concentration is exceeded on most sampling days, it is in the interest of public health to ascertain whether or not this trend is limited to the summer months. Should fine PM pollutant concentrations be chronically high throughout the year, individuals who are constantly exposed to the polluted air become more susceptible to cardiovascular and pulmonary damage. Additional consideration should be given to areas within these planning districts that are home to higher percentages of physiologically vulnerable (e.g. young, elderly, individuals with pre-existing conditions) and socioeconomically disadvantaged residents. It is crucial to address air quality problems in an environmentally just fashion to promote equitable urban sustainability, and further study in these areas can provide insight that can reduce disparities in the environmental health burden and improve overall human health in urban environments.

Though our analysis reveals spatiotemporal variation in PM and BC, along with possible causes of this variation, it stops short of estimating the contributions of specific sources to this variability; as our air pollution measurements covary in space and time, it is difficult to quantify the extent of variation resulting from spatial influences (locations of point sources, movement of non-point sources) and temporal influences (temporally sensitive atmospheric processes, random events) separately. Future analyses should consider the influence of urban composition and structure on air pollutant concentrations throughout cities. Cities can be quite different from each other compositionally and structurally, and the roles of urban structure (Cárdenas Rodríguez et al., 2016) and land use (Weng and Yang, 2006; Shakya et al., 2019) are important drivers of variation in urban air pollution. Cities can also vary significantly with regard to the regional climate; future studies of air pollution in Philadelphia and other urban environments should ascertain the influences of climate, specifically prevailing winds and temperature, on the spatiotemporal distribution of air pollutants. These analyses can be used to supplement existing knowledge regarding where and when air pollution may have the greatest adverse impacts on human and environmental health, which is essential to sustainable and equitable air quality management in our ever-expanding urban environments.