Land Use Regression Models of On-Road Particulate Air Pollution (Particle Number, Black Carbon, PM2.5, Particle Size) Using Mobile Monitoring

High-resolution urban air pollution mapping

Variation in urban air pollution arises because of complex spatial, temporal, and chemical processes, which profoundly affect population exposure, human health, and environmental justice. This Review highlights insights from two popular in situ measurement methods-mobile monitoring and dense sensor networks-that have distinct but complementary strengths in characterizing the dynamics and impacts of the multidimensional urban air quality system. Mobile monitoring can measure many pollutants at fine spatial scales, thereby informing about processes and control strategies. Sensor networks excel at providing temporal resolution at many locations. Increasingly sophisticated studies leveraging both methods can vividly identify spatial and temporal patterns that affect exposures and disparities and offer mechanistic insight toward effective interventions. This Review summarizes the strengths and limitations of these methods and discusses their implications for understanding fine-scale processes and impacts.

Integrating Mobile and Fixed-Site Black Carbon Measurements to Bridge Spatiotemporal Gaps in Urban Air Quality

Urban air pollution can vary sharply in space and time. However, few monitoring strategies can concurrently resolve spatial and temporal variation at fine scales. Here, we present a new measurement-driven spatiotemporal modeling approach that transcends the individual limitations of two complementary sampling paradigms: mobile monitoring and fixed-site sensor networks. We develop, validate, and apply this model to predict black carbon (BC) using data from an intensive, 100-day field study in West Oakland, CA. Our spatiotemporal model exploits coherent spatial patterns derived from a multipollutant mobile monitoring campaign to fill spatial gaps in time-complete BC data from a low-cost sensor network. Our model performs well in reconstructing patterns at fine spatial and temporal resolution (30 m, 15 min), demonstrating strong out-of-sample correlations for both mobile (Pearson's R ∼ 0.77) and fixed-site measurements (R ∼ 0.95) while revealing features that are not effectively captured by a single monitoring approach in isolation. The model reveals sharp concentration gradients near major emission sources while capturing their temporal variability, offering valuable insights into pollution sources and dynamics.

High-resolution spatial and spatiotemporal modelling of air pollution using fixed site and mobile monitoring in a Canadian city

The development of high-resolution spatial and spatiotemporal models of air pollutants is essential for exposure science and epidemiological applications. While fixed-site sampling has conventionally provided input data for statistical predictive models, the evolving mobile monitoring method offers improved spatial resolution, ideal for measuring pollutants with high spatial variability such as ultrafine particles (UFP). The Quebec Air Pollution Exposure and Epidemiology (QAPEE) study measured and modelled the spatial and spatiotemporal distributions of understudied pollutants, such as UFPs, black carbon (BC), and brown carbon (BrC), along with fine particulate matter (PM2.5), nitrogen dioxide (NO2), and ozone (O3) in Quebec City, Canada. We conducted a combined fixed-site (NO2 and O3) and mobile monitoring (PM2.5, BC, BrC, and UFPs) campaign over 10-months. Mobile monitoring routes were monitored on a weekly basis between 8am-10am and designed using location/allocation modelling. Seasonal fixed-site sampling campaigns captured continuous 24-h measurements over two-week periods. Generalized Additive Models (GAMs), which combined data on pollution concentrations with spatial, temporal, and spatiotemporal predictor variables were used to model and predict concentration surfaces. Annual models for PM2.5, NO2, O3 as well as seven of the smallest size fractions in the UFP range, had high out of sample predictive accuracy (range r2: 0.54-0.86). Varying spatial patterns were observed across UFP size ranges measured as Particle Number Counts (PNC). The monthly spatiotemporal models for PM2.5 (r2 = 0.49), BC (r2 = 0.27), BrC (r2 = 0.29), and PNC (r2 = 0.49) had moderate or moderate-low out of sample predictive accuracy. We conducted a sensitivity analysis and found that the minimum number of 'n visits' (mobile monitoring sessions) required to model annually representative air pollution concentrations was between 24 and 32 visits dependent on the pollutant. This study provides a single source of exposure models for a comprehensive set of air pollutants in Quebec City, Canada. These exposure models will feed into epidemiological research on the health impacts of ambient UFPs and other pollutants.

Weather and the City: Machine Learning for Predicting and Attributing Fine Scale Air Quality to Meteorological and Urban Determinants

Urban air quality persists as a global concern, with critical health implications. This study employs a combination of machine learning (gradient boosting regression, GBR) and spatial analysis to better understand the key drivers behind air pollution and its prediction and mitigation strategies. Focusing on New York City as a representative urban area, we investigate the interplay between urban characteristics and weather factors, showing that urban features, including traffic-related parameters and urban morphology, emerge as crucial predictors for pollutants closely associated with vehicular emissions, such as elemental carbon (EC) and nitrogen oxides (NOx). Conversely, pollutants with secondary formation pathways (e.g., PM2.5) or stemming from nontraffic sources (e.g., sulfur dioxide, SO2) are predominantly influenced by meteorological conditions, particularly wind speed and maximum daily temperature. Urban characteristics are shown to act over spatial scales of 500 × 500 m2, which is thus the footprint needed to effectively capture the impact of urban form, fabric, and function. Our spatial predictive model, needing only meteorological and urban inputs, achieves promising results with mean absolute errors ranging from 8 to 32% when using full-year data. Our approach also yields good performance when applied to the temporal mapping of spatial pollutant variability. Our findings highlight the interacting roles of urban characteristics and weather conditions and can inform urban planning, design, and policy.

A comprehensive review of the development of land use regression approaches for modeling spatiotemporal variations of ambient air pollution: A perspective from 2011 to 2023

Land use regression (LUR) models are widely used in epidemiological and environmental studies to estimate humans' exposure to air pollution within urban areas. However, the early models, developed using linear regressions and data from fixed monitoring stations and passive sampling, were primarily designed to model traditional and criteria air pollutants and had limitations in capturing high-resolution spatiotemporal variations of air pollution. Over the past decade, there has been a notable development of multi-source observations from low-cost monitors, mobile monitoring, and satellites, in conjunction with the integration of advanced statistical methods and spatially and temporally dynamic predictors, which have facilitated significant expansion and advancement of LUR approaches. This paper reviews and synthesizes the recent advances in LUR approaches from the perspectives of the changes in air quality data acquisition, novel predictor variables, advances in model-developing approaches, improvements in validation methods, model transferability, and modeling software as reported in 155 LUR studies published between 2011 and 2023. We demonstrate that these developments have enabled LUR models to be developed for larger study areas and encompass a wider range of criteria and unregulated air pollutants. LUR models in the conventional spatial structure have been complemented by more complex spatiotemporal structures. Compared with linear models, advanced statistical methods yield better predictions when handling data with complex relationships and interactions. Finally, this study explores new developments, identifies potential pathways for further breakthroughs in LUR methodologies, and proposes future research directions. In this context, LUR approaches have the potential to make a significant contribution to future efforts to model the patterns of long- and short-term exposure of urban populations to air pollution.

Enhancing the reliability of particulate matter sensing by multivariate Tobit model using weather and air quality data

Low-cost particulate matter (PM) sensors have been widely used following recent sensor-technology advancements; however, inherent limitations of low-cost monitors (LCMs), which operate based on light scattering without an air-conditioning function, still restrict their applicability. We propose a regional calibration of LCMs using a multivariate Tobit model with historical weather and air quality data to improve the accuracy of ambient air monitoring, which is highly dependent on meteorological conditions, local climate, and regional PM properties. Weather observations and PM2.5 (fine inhalable particles with diameters ≤ 2.5 μm) concentrations from two regions in Korea, Incheon and Jeju, and one in Singapore were used as training data to build a visibility-based calibration model. To validate the model, field measurements were conducted by an LCM in Jeju and Singapore, where R2 and the error after applying the model in Jeju improved (from 0.85 to 0.88) and reduced by 44% (from 8.4 to 4.7 μg m-3), respectively. The results demonstrated that regional calibration involving air temperature, relative humidity, and other local climate parameters can efficiently correct the bias of the sensor. Our findings suggest that the proposed post-processing using the Tobit model with regional weather and air quality data enhances the applicability of LCMs.

Predicting spatial variations in annual average outdoor ultrafine particle concentrations in Montreal and Toronto, Canada: Integrating land use regression and deep learning models

BACKGROUND

Concentrations of outdoor ultrafine particles (UFP; <0.1 µm) and black carbon (BC) can vary greatly within cities and long-term exposures to these pollutants have been associated with a variety of adverse health outcomes.

OBJECTIVE

This study integrated multiple approaches to develop new models to estimate within-city spatial variations in annual median (i.e. average) outdoor UFP and BC concentrations as well as mean UFP size in Canada's two largest cities, Montreal and Toronto.

METHODS

We conducted year-long mobile monitoring campaigns in each city that included evenings and weekends. We developed generalized additive models trained on land use parameters and deep Convolutional Neural Network (CNN) models trained on satellite-view images. Using predictions from these models, we developed final combined models.

RESULTS

In Toronto, the median observed UFP concentration, UFP size, and BC concentration values were 16,172pt/cm3, 33.7 nm, and 1225 ng/m3, respectively. In Montreal, the median observed UFP concentration, UFP size, and BC concentration values were 14,702pt/cm3, 29.7 nm, and 1060 ng/m3, respectively. For all pollutants in both cities, the proportion of spatial variation explained (i.e., R2) was slightly greater (1-2 percentage points) for the combined models than the generalized additive models and a greater (approximately 10 percentage points) than the deep CNN models. The Toronto combined model R2 values in the test set were 0.73, 0.55, and 0.61 for UFP concentrations, UFP size, and BC concentration, respectively. The Montreal combined model R2 values were 0.60, 0.49, and 0.60 for UFP concentration, UFP size, and BC concentration models respectively. For each pollutant, predictions from the combined, deep CNN, and generalized additive models were highly correlated with each other and differences between models were explored in sensitivity analyses.

CONCLUSION

Predictions from these models are available to support future epidemiological research examining long-term health impacts of outdoor UFPs and BC.

Characterizing Ultrafine Particle Mobile Monitoring Data for Epidemiology

Mobile monitoring is increasingly used to assess exposure to traffic-related air pollutants (TRAPs), including ultrafine particles (UFPs). Due to the rapid spatial decrease in the concentration of UFPs and other TRAPs with distance from roadways, mobile measurements may be non-representative of residential exposures, which are commonly used for epidemiologic studies. Our goal was to develop, apply, and test one possible approach for using mobile measurements in exposure assessment for epidemiology. We used an absolute principal component score model to adjust the contribution of on-road sources in mobile measurements to provide exposure predictions representative of cohort locations. We then compared UFP predictions at residential locations from mobile on-road plume-adjusted versus stationary measurements to understand the contribution of mobile measurements and characterize their differences. We found that predictions from mobile measurements are more representative of cohort locations after down-weighting the contribution of localized on-road plumes. Further, predictions at cohort locations derived from mobile measurements incorporate more spatial variation compared to those from short-term stationary data. Sensitivity analyses suggest that this additional spatial information captures features in the exposure surface not identified from the stationary data alone. We recommend the correction of mobile measurements to create exposure predictions representative of residential exposure for epidemiology.

Exposure Disparities by Income, Race and Ethnicity, and Historic Redlining Grade in the Greater Seattle Area for Ultrafine Particles and Other Air Pollutants

BACKGROUND

Growing evidence shows ultrafine particles (UFPs) are detrimental to cardiovascular, cerebrovascular, and respiratory health. Historically, racialized and low-income communities are exposed to higher concentrations of air pollution.

OBJECTIVES

Our aim was to conduct a descriptive analysis of present-day air pollution exposure disparities in the greater Seattle, Washington, area by income, race, ethnicity, and historical redlining grade. We focused on UFPs (particle number count) and compared with black carbon, nitrogen dioxide, and fine particulate matter (PM 2.5 ) levels.

METHODS

We obtained race and ethnicity data from the 2010 U.S. Census, median household income data from the 2006-2010 American Community Survey, and Home Owners' Loan Corporation (HOLC) redlining data from the University of Richmond's Mapping Inequality. We predicted pollutant concentrations at block centroids from 2019 mobile monitoring data. The study region encompassed much of urban Seattle, with redlining analyses restricted to a smaller region. To analyze disparities, we calculated population-weighted mean exposures and regression analyses using a generalized estimating equation model to account for spatial correlation.

RESULTS

Pollutant concentrations and disparities were largest for blocks with median household income of < $ 20,000 , Black residents, HOLC Grade D, and ungraded industrial areas. UFP concentrations were 4% lower than average for non-Hispanic White residents and higher than average for racialized groups (Asian, 3%; Black, 15%; Hispanic, 6%; Native American, 8%; Pacific Islander, 11%). For blocks with median household incomes of < $ 20,000 , UFP concentrations were 40% higher than average, whereas blocks with incomes of > $ 110,000 had UFP concentrations 16% lower than average. UFP concentrations were 28% higher for Grade D and 49% higher for ungraded industrial areas compared with Grade A. Disparities were highest for UFPs and lowest for PM 2.5 exposure levels.

DISCUSSION

Our study is one of the first to highlight large disparities with UFP exposures compared with multiple pollutants. Higher exposures to multiple air pollutants and their cumulative effects disproportionately impact historically marginalized groups. https://doi.org/10.1289/EHP11662.

A hyperlocal hybrid data fusion near-road PM2.5 and NO2 annual risk and environmental justice assessment across the United States

Exposure to traffic-related air pollutants (TRAPs) has been associated with numerous adverse health effects. TRAP concentrations are highest meters away from major roads, and disproportionately affect minority (i.e., non-white) populations often considered the most vulnerable to TRAP exposure. To demonstrate an improved assessment of on-road emissions and to quantify exposure inequity in this population, we develop and apply a hybrid data fusion approach that utilizes the combined strength of air quality observations and regional/local scale models to estimate air pollution exposures at census block resolution for the entire U.S. We use the regional photochemical grid model CMAQ (Community Multiscale Air Quality) to predict the spatiotemporal impacts at local/regional scales, and the local scale dispersion model, R-LINE (Research LINE source) to estimate concentrations that capture the sharp TRAP gradients from roads. We further apply the Regionalized Air quality Model Performance (RAMP) Hybrid data fusion technique to consider the model's nonhomogeneous, nonlinear performance to not only improve exposure estimates, but also achieve significant model performance improvement. With a R2 of 0.51 for PM2.5 and 0.81 for NO2, the RAMP hybrid method improved R2 by ~0.2 for both pollutants (an increase of up to ~70% for PM2.5 and ~31% NO2). Using the RAMP Hybrid method, we estimate 264,516 [95% confidence interval [CI], 223,506-307,577] premature deaths attributable to PM2.5 from all sources, a ~1% overall decrease in CMAQ-estimated premature mortality compared to RAMP Hybrid, despite increases and decreases in some locations. For NO2, RAMP Hybrid estimates 138,550 [69,275-207,826] premature deaths, a ~19% increase (22,576 [11,288 - 33,864]) compared to CMAQ. Finally, using our RAMP hybrid method to estimate exposure inequity across the U.S., we estimate that Minorities within 100 m from major roads are exposed to up to 15% more PM2.5 and up to 35% more NO2 than their White counterparts.

Design and evaluation of short-term monitoring campaigns for long-term air pollution exposure assessment

BACKGROUND

Short-term mobile monitoring campaigns to estimate long-term air pollution levels are becoming increasingly common. Still, many campaigns have not conducted temporally-balanced sampling, and few have looked at the implications of such study designs for epidemiologic exposure assessment.

OBJECTIVE

We carried out a simulation study using fixed-site air quality monitors to better understand how different short-term monitoring designs impact the resulting exposure surfaces.

METHODS

We used Monte Carlo resampling to simulate three archetypal short-term monitoring sampling designs using oxides of nitrogen (NOx) monitoring data from 69 regulatory sites in California: a year-around Balanced Design that sampled during all seasons of the year, days of the week, and all or various hours of the day; a temporally reduced Rush Hours Design; and a temporally reduced Business Hours Design. We evaluated the performance of each design's land use regression prediction model.

RESULTS

The Balanced Design consistently yielded the most accurate annual averages; while the reduced Rush Hours and Business Hours Designs generally produced more biased results.

SIGNIFICANCE

A temporally-balanced sampling design is crucial for short-term campaigns such as mobile monitoring aiming to assess long-term exposure in epidemiologic cohorts.

IMPACT STATEMENT

Short-term monitoring campaigns to assess long-term air pollution trends are increasingly common, though they rarely conduct temporally balanced sampling. We show that this approach produces biased annual average exposure estimates that can be improved by collecting temporally-balanced samples.

Exposure assessment for air pollution epidemiology: A scoping review of emerging monitoring platforms and designs

BACKGROUND

Both exposure monitoring and exposure prediction have played key roles in assessing individual-level long-term exposure to air pollutants and their associations with human health. While there have been notable advances in exposure prediction methods, improvements in monitoring designs are also necessary, particularly given new monitoring paradigms leveraging low-cost sensors and mobile platforms.

OBJECTIVES

We aim to provide a conceptual summary of novel monitoring designs for air pollution cohort studies that leverage new paradigms and technologies, to investigate their characteristics in real-world examples, and to offer practical guidance to future studies.

METHODS

We propose a conceptual summary that focuses on two overarching types of monitoring designs, mobile and non-mobile, as well as their subtypes. We define mobile designs as monitoring from a moving platform, and non-mobile designs as stationary monitoring from permanent or temporary locations. We only consider non-mobile studies with cost-effective sampling devices. Then we discuss similarities and differences across previous studies with respect to spatial and temporal representation, data comparability between design classes, and the data leveraged for model development. Finally, we provide specific suggestions for future monitoring designs.

RESULTS

Most mobile and non-mobile monitoring studies selected monitoring sites based on land use instead of residential locations, and deployed monitors over limited time periods. Some studies applied multiple design and/or sub-design classes to the same area, time period, or instrumentation, to allow comparison. Even fewer studies leveraged monitoring data from different designs to improve exposure assessment by capitalizing on different strengths. In order to maximize the benefit of new monitoring technologies, future studies should adopt monitoring designs that prioritize residence-based site selection with comprehensive temporal coverage and leverage data from different designs for model development in the presence of good data compatibility.

DISCUSSION

Our conceptual overview provides practical guidance on novel exposure assessment monitoring for epidemiological applications.

Nearly five-year continuous atmospheric measurements of black carbon over a suburban area in central France

Atmospheric black carbon (BC) concentration over a nearly 5 year period (mid-2017-2021) was continuously monitored over a suburban area of Orléans city (France). Annual mean atmospheric BC concentration were 0.75 ± 0.65, 0.58 ± 0.44, 0.54 ± 0.64, 0.48 ± 0.46 and 0.50 ± 0.72 μg m^-3, respectively, for the year of 2017, 2018, 2019, 2020 and 2021. Seasonal pattern was also observed with maximum concentration (0.70 ± 0.18 μg m^-3) in winter and minimum concentration (0.38 ± 0.04 μg m^-3) in summer. We found a different diurnal pattern between cold (winter and fall) and warm (spring and summer) seasons. Further, fossil fuel burning contributed >90 % of atmospheric BC in the summer and biomass burning had a contribution equivalent to that of the fossil fuel in the winter. Significant week days effect on BC concentrations was observed, indicating the important role of local emissions such as car exhaust in BC level at this site. The behavior of atmospheric BC level with COVID-19 lockdown was also analyzed. We found that during the lockdown in warm season (first lockdown: 27 March-10 May 2020 and third lockdown 17 March-3 May 2021) BC concentration were lower than in cold season (second lockdown: 29 October-15 December 2020), which could be mainly related to the BC emission from biomass burning for heating. This study provides a long-term BC measurement database input for air quality and climate models. The analysis of especially weekend and lockdown effect showed implications on future policymaking toward improving local and regional air quality as well.

Design, calibration, and testing of a mobile sensor system for air pollution and built environment data collection: The urban scanner platform

This paper describes a mobile air pollution sampling system, the Urban Scanner, which aims at gathering dense spatiotemporal air quality data to support urban air quality and exposure science. Urban Scanner comprises custom vehicle-mounted sensors for air pollution, meteorology, and built environment data collection (low-cost sensors, wind anemometer, 360 deg camera, LIDAR, GPS) as well as a server to store, process, and map all gathered geo-referenced sensory information. Two levels of sensor calibration were implemented, both in a chamber and in the field, against reference instrumentation. Chamber tests and a set of mathematical tools were developed to correct for sensor noise (wavelet denoising), misalignment (linear and nonlinear), and hysteresis removal. Models based on chamber testing were further refined based on field co-location. While field co-location captures natural changes in air pollution and meteorology, chamber tests allow for simulating fast transitions in these variables, like the transitions experienced by a mobile sensor in an urban environment. The best suite of models achieved an R² higher than 0.9 between sensor output and reference station observations and an RMSE of 2.88 ppb for nitrogen dioxide and 4.03 ppb for ozone. A mobile sampling campaign was conducted in the city of Toronto, Canada, to further test Urban Scanner. We observe that the platform adequately captures spatial and temporal variability in urban air pollution, leading to the development of land-use regression models with high explanatory power.

Impact of Mobile Monitoring Network Design on Air Pollution Exposure Assessment Models

Short-term mobile monitoring campaigns are increasingly used to assess long-term air pollution exposure in epidemiology. Little is known about how monitoring network design features, including the number of stops and sampling temporality, impacts exposure assessment models. We address this gap by leveraging an extensive mobile monitoring campaign conducted in the greater Seattle area over the course of a year during all days of the week and most hours. The campaign measured total particle number concentration (PNC; sheds light on ultrafine particulate (UFP) number concentration), black carbon (BC), nitrogen dioxide (NO2), fine particulate matter (PM2.5), and carbon dioxide (CO2). In Monte Carlo sampling of 7327 total stops (278 sites × 26 visits each), we restricted the number of sites and visits used to estimate annual averages. Predictions from the all-data campaign performed well, with cross-validated R2s of 0.51-0.77. We found similar model performances (85% of the all-data campaign R2) with ∼1000 to 3000 randomly selected stops for NO2, PNC, and BC, and ∼4000 to 5000 stops for PM2.5 and CO2. Campaigns with additional temporal restrictions (e.g., business hours, rush hours, weekdays, or fewer seasons) had reduced model performances and different spatial surfaces. Mobile monitoring campaigns wanting to assess long-term exposure should carefully consider their monitoring designs.

Hyperlocal variation of nitrogen dioxide, black carbon, and ultrafine particles measured with Google Street View cars in Amsterdam and Copenhagen

Hyperlocal air quality maps are becoming increasingly common, as they provide useful insights into the spatial variation and sources of air pollutants. In this study, we produced several high-resolution concentration maps to assess the spatial differences of three traffic-related pollutants, Nitrogen dioxide (NO₂), Black Carbon (BC) and Ultrafine Particles (UFP), in Amsterdam, the Netherlands, and Copenhagen, Denmark. All maps were based on a mixed-effect model approach by using state-of-the-art mobile measurements conducted by Google Street View (GSV) cars, during October 2018 - March 2020, and Land-use Regression (LUR) models based on several land-use and traffic predictor variables. We then explored the concentration ratio between the different normalised pollutants to understand possible contributing sources to the observed hyperlocal variations. The maps developed in this work reflect, (i) expected elevated pollution concentrations along busy roads, and (ii) similar concentration patterns on specific road types, e.g., motorways, for both cities. In the ratio maps, we observed a clear pattern of elevated concentrations of UFP near the airport in both cities, compared to BC and NO₂. This is the first study to produce hyperlocal maps for BC and UFP using high-quality mobile measurements. These maps are important for policymakers and health-effect studies, trying to disentangle individual effects of key air pollutants of interest (e.g., UFP).

A Knowledge Transfer Approach to Map Long-Term Concentrations of Hyperlocal Air Pollution from Short-Term Mobile Measurements

Mobile measurements are increasingly used to develop spatially explicit (hyperlocal) air quality maps using land-use regression (LUR) models. The prevailing design of mobile monitoring campaigns results in the collection of short-term, on-road air pollution measurements during daytime on weekdays. We hypothesize that LUR models trained with such mobile measurements are not optimized for estimating long-term average residential air pollution concentrations. To bridge the knowledge gaps in space (on-road versus near-road) and time (short- versus long-term), we propose transfer-learning techniques to adapt LUR models by transferring the mobile knowledge into long-term near-road knowledge in an end-to-end manner. We trained two transfer-learning LUR models by incorporating mobile measurements of nitrogen dioxide (NO₂) and ultrafine particles (UFP) collected by Google Street View cars with long-term near-road measurements from regular monitoring networks in Amsterdam. We found that transfer-learning LUR models performed 55.2% better in predicting long-term near-road concentrations than the LUR model trained only with mobile measurements for NO₂ and 26.9% for UFP, evaluated by normalized mean absolute errors. This improvement in model accuracy suggests that transfer-learning models provide a solution for narrowing the knowledge gaps and can improve the accuracy of mapping long-term near-road air pollution concentrations using short-term on-road mobile monitoring data.

National Land Use Regression Model for NO2 Using Street View Imagery and Satellite Observations

Land use regression (LUR) models are widely applied to estimate intra-urban air pollution concentrations. National-scale LURs typically employ predictors from multiple curated geodatabases at neighborhood scales. In this study, we instead developed national NO₂ models relying on innovative street-level predictors extracted from Google Street View [GSV] imagery. Using machine learning (random forest), we developed two types of models: (1) GSV-only models, which use only GSV features, and (2) GSV + OMI models, which also include satellite observations of NO₂. Our results suggest that street view imagery alone may provide sufficient information to explain NO₂ variation. Satellite observations can improve model performance, but the contribution decreases as more images are available. Random 10-fold cross-validation R² of our best models were 0.88 (GSV-only) and 0.91 (GSV + OMI)─a performance that is comparable to traditional LUR approaches. Importantly, our models show that street-level features might have the potential to better capture intra-urban variation of NO₂ pollution than traditional LUR. Collectively, our findings indicate that street view image-based modeling has great potential for building large-scale air quality models under a unified framework. Toward that goal, we describe a cost-effective image sampling strategy for future studies based on a systematic evaluation of image availability and model performance.

Prediction of Short-Term Ultrafine Particle Exposures Using Real-Time Street-Level Images Paired with Air Quality Measurements

Within-city ultrafine particle (UFP) concentrations vary sharply since they are influenced by various factors. We developed prediction models for short-term UFP exposures using street-level images collected by a camera installed on a vehicle rooftop, paired with air quality measurements conducted during a large-scale mobile monitoring campaign in Toronto, Canada. Convolutional neural network models were trained to extract traffic and built environment features from images. These features, along with regional air quality and meteorology data were used to predict short-term UFP concentration as a continuous and categorical variable. A gradient boost model for UFP as a continuous variable achieved R² = 0.66 and RMSE = 9391.8#/cm³ (mean values for 10-fold cross-validation). The model predicting categorical UFP achieved accuracies for "Low" and "High" UFP of 77 and 70%, respectively. The presence of trucks and other traffic parameters were associated with higher UFPs, and the spatial distribution of elevated short-term UFP followed the distribution of single-unit trucks. This study demonstrates that pictures captured on urban streets, associated with regional air quality and meteorology, can adequately predict short-term UFP exposure. Capturing the spatial distribution of high-frequency short-term UFP spikes in urban areas provides crucial information for the management of near-road air pollution hot spots.

Characterization of Annual Average Traffic-Related Air Pollution Concentrations in the Greater Seattle Area from a Year-Long Mobile Monitoring Campaign

Growing evidence links traffic-related air pollution (TRAP) to adverse health effects. We designed an innovative and extensive mobile monitoring campaign to characterize TRAP exposure levels for the Adult Changes in Thought (ACT) study, a Seattle-based cohort. The campaign measured particle number concentration (PNC) to capture ultrafine particles (UFP), black carbon (BC), nitrogen dioxide (NO2), fine particulate matter (PM2.5), and carbon dioxide (CO2) at 309 roadside sites within a large, 1200 land km2 (463 mi2) area representative of the cohort. We collected about 29 two-minute measurements at each site during all seasons, days of the week, and most times of the day over a 1-year period. Validation showed good agreement between our BC, NO2, and PM2.5 measurements and monitoring agency sites (R2 = 0.68-0.73). Universal kriging-partial least squares models of annual average pollutant concentrations had cross-validated mean square error-based R2 (and root mean square error) values of 0.77 (1177 pt/cm3) for PNC, 0.60 (102 ng/m3) for BC, 0.77 (1.3 ppb) for NO2, 0.70 (0.3 μg/m3) for PM2.5, and 0.51 (4.2 ppm) for CO2. Overall, we found that the design of this extensive campaign captured the spatial pollutant variations well and these were explained by sensible land use features, including those related to traffic.

Consistency of the relationship between air pollution and the urban form: Evidence from the COVID-19 natural experiment

The lockdown measures enacted to control the COVID-19 pandemic in Wuhan, China, resulted in a suspension of nearly all non-essential human activities on January 23, 2020. Nevertheless, the lockdown provided a natural experiment to understand the consistency of the relationship between the urban form and air pollution with different compositions of locally or regionally transported sources. This study investigated the variations in six air pollutants (PM_2.5, PM₁₀, NO₂, CO, O₃, and SO₂) in Wuhan before and during the lockdown and in the two same time spans in 2021. Moreover, a hierarchical agglomerative cluster analysis was conducted to differentiate the relative levels of pollutants and to detect the relationships between the air pollutants and the urban form during these four periods. Several features depicting the urban physical structures delivered consistent impacts. A lower building density and plot ratio, and a higher porosity always mitigated the concentrations of NO₂ and PM_2.5. However, they had inverse effects on O₃ during the non-lockdown periods. PM₁₀, CO, and SO₂ concentrations have little correlation with the urban form. This study improves the comprehensive understanding of the effect of the urban form on ambient air pollution and suggests practical strategies for mitigating air pollution in Wuhan.

Leveraging Citizen Science and Low-Cost Sensors to Characterize Air Pollution Exposure of Disadvantaged Communities in Southern California

Assessing exposure to fine particulate matter (PM_2.5) across disadvantaged communities is understudied, and the air monitoring network is inadequate. We leveraged emerging low-cost sensors (PurpleAir) and engaged community residents to develop a community-based monitoring program across disadvantaged communities (high proportions of low-income and minority populations) in Southern California. We recruited 22 households from 8 communities to measure residential outdoor PM_2.5 concentrations from June 2021 to December 2021. We identified the spatial and temporal patterns of PM_2.5 measurements as well as the relationship between the total PM_2.5 measurements and diesel PM emissions. We found that communities with a higher percentage of Hispanic and African American population and higher rates of unemployment, poverty, and housing burden were exposed to higher PM_2.5 concentrations. The average PM_2.5 concentrations in winter (25.8 µg/m³) were much higher compared with the summer concentrations (12.4 µg/m³). We also identified valuable hour-of-day and day-of-week patterns among disadvantaged communities. Our results suggest that the built environment can be targeted to reduce the exposure disparity. Integrating low-cost sensors into a citizen-science-based air monitoring program has promising applications to resolve monitoring disparity and capture “hotspots” to inform emission control and urban planning policies, thus improving exposure assessment and promoting environmental justice.

Modeling spatial variation of gaseous air pollutants and particulate matters in a Metropolitan area using mobile monitoring data

Geo-statistical models have been applied to assess fine-scale air pollution exposures in epidemiological studies. Many of the models were developed for criteria air pollutants rather than others that have not been regulated (e.g., ultrafine particles, black carbon, and benzene) which may also be harmful to human health. We aim to develop spatial models for regulated and non-regulated air pollutants using 6 algorithms and compare their prediction performances. A mobile platform with fast-response monitors was used to measure gaseous air pollutants (nitrogen dioxides, carbon monoxide, sulfur dioxides, ozone, benzene, toluene, methanol) and particulate matters (black carbon, surface area, count- and volume-concentrations of ultrafine particles) in Beijing, China for 30 days from July to October 2008. Mobile monitoring data for model building were spatially aggregated into 130 road segments of approximately 600-m interval on the sampling routes after temporal adjustment of background concentrations. The best models for the air pollutants were dominated by traffic variables, which explained more than 60% of the spatial variations (range: 0.61 for methanol to 0.88 for ozone) based on the highest cross-validation R² and the lowest root mean square error among different algorithms. Amongst the 6 algorithms, the spatial models using partial least squares regression (PLS, a dimension reduction algorithm) and random forest (RF, a machine learning algorithm) algorithms outperformed the models with other algorithms. Exposure predictions from the best models varied substantially with distinct spatial patterns between the air pollutants. Predictions with multiple modeling algorithms were moderately correlated with each other for the same pollutant at the fine-scale grids across the city. Exposure models, especially based on PLS and RF algorithms, captured the spatial variation of short-term average concentrations, had adequate predictive validity, and could be applied to assess toxic air pollutant exposures in human health studies.

Three-dimensional building morphology impacts on PM2.5 distribution in urban landscape settings in Zhejiang, China

Three-dimensional (3D) urban landscape patterns and building morphology are crucial for urban planning and essential for urban landscape functions. In this study, fixed and mobile monitoring sites were used to determine the spatial distribution of PM_2.5 concentrations in Hangzhou. Six 3D metrics were selected to analyze the response of PM_2.5 pollution to landscape patterns and building morphology, while their two-dimensional (2D) counterparts' metrics were also analyzed to contrast the differences. A variance partitioning analysis (VPA) was performed to measure the combined and relative contribution of 3D and 2D metrics to the changes in PM_2.5 concentrations. The results showed that: (1) on the 3D scale, forming a building pattern with a combination of different building heights can eliminate the accumulation of PM_2.5; (2) on the 2D scale, fragmentation and decentralization of landscapes and building patches alleviate PM_2.5 pollution; and (3) 3D building morphology indicators have the highest explanatory power (40.94%) for the changes of PM_2.5 concentrations. It turns out that the explanatory power of 3D metrics for PM_2.5 concentrations changes is much greater than that of 2D metrics. In addition, when compared to building morphology indicators from a single dimension, the combination of 2D and 3D metrics is better able to reflect urban PM_2.5 pollution. The results of this study expand our understanding of how PM_2.5 pollution responds to 2D and 3D metrics and provide useful information for urban planning.

Evaluating the Meteorological Effects on the Urban Form-Air Quality Relationship Using Mobile Monitoring

Predictive models based on mobile measurements have been increasingly used to understand the spatiotemporal variations of intraurban air quality. However, the effects of meteorological factors, which significantly affect the dispersion of air pollution, on the urban-form-air-quality relationship have not been understood on a granular level. We attempt to fill this gap by developing predictive models of particulate matter (PM) in the Bronx (New York City) using meteorological and urban form parameters. The granular PM data was collected by mobile low-cost sensors as the ground truth. To evaluate the effects of meteorological factors, we compared the performance of models using the urban form within fixed and wind-sensitive buffers, respectively. We find better predictive power in the wind-sensitive group (R = 0.85) for NC₁₀ (number concentration for particles with diameters of 1 μm-10 μm) than the control group (R = 0.01), and modest improvements for PM_2.5 (R = 0.84 for the wind sensitive group, R = 0.77 for the control group), indicating that incorporating meteorological factors improved the predictive power of our models. We also found that urban form factors account for 62.95% of feature importance for NC₁₀ and 14.90% for PM_2.5 (9.99% and 4.91% for 3-D and 2-D urban form factors, respectively) in our Random Forest models. It suggests the importance of incorporating urban form factors, especially for the uncommonly used 3-D characteristics, in estimating intraurban PM. Our method can be applied in other cities to better capture the influence of urban context on PM levels.

Mixed-Effects Modeling Framework for Amsterdam and Copenhagen for Outdoor NO2 Concentrations Using Measurements Sampled with Google Street View Cars

High-resolution air quality (AQ) maps based on street-by-street measurements have become possible through large-scale mobile measurement campaigns. Such campaigns have produced data-only maps and have been used to produce empirical models [i.e., land use regression (LUR) models]. Assuming that all road segments are measured, we developed a mixed model framework that predicts concentrations by an LUR model, while allowing road segments to deviate from the LUR prediction based on between-segment variation as a random effect. We used Google Street View cars, equipped with high-quality AQ instruments, and measured the concentration of NO₂ on every street in Amsterdam (n = 46.664) and Copenhagen (n = 28.499) on average seven times over the course of 9 and 16 months, respectively. We compared the data-only mapping, LUR, and mixed model estimates with measurements from passive samplers (n = 82) and predictions from dispersion models in the same time window as mobile monitoring. In Amsterdam, mixed model estimates correlated r_s (Spearman correlation) = 0.85 with external measurements, whereas the data-only approach and LUR model estimates correlated r_s = 0.74 and 0.75, respectively. Mixed model estimates also correlated higher r_s = 0.65 with the deterministic model predictions compared to the data-only (r_s = 0.50) and LUR model (r_s = 0.61). In Copenhagen, mixed model estimates correlated r_s = 0.51 with external model predictions compared to r_s = 0.45 and r_s = 0.50 for data-only and LUR model, respectively. Correlation increased for 97 locations (r_s = 0.65) with more detailed traffic information. This means that the mixed model approach is able to combine the strength of data-only mapping (to show hyperlocal variation) and LUR models by shrinking uncertain concentrations toward the model output.

A three-dimensional LUR framework for PM2.5 exposure assessment based on mobile unmanned aerial vehicle monitoring

Land use regression (LUR) models have been widely used in epidemiological studies and risk assessments related to air pollution. Although efforts have been made to improve the performance of LUR models so that they capture the spatial heterogeneity of fine particulate matter (PM_2.5) in high-density cities, few studies have revealed the vertical differences in PM_2.5 exposure. This study proposes a three-dimensional LUR (3-D LUR) assessment framework for PM_2.5 exposure that combines a high-resolution LUR model with a vertical PM_2.5 variation model to investigate the results of horizontal and vertical mobile PM_2.5 monitoring campaigns. High-resolution LUR models that were developed independently for daytime and nighttime were found to explain 51% and 60% of the PM_2.5 variation, respectively. Vertical measurements of PM_2.5 from three regions were first parameterized to produce a coefficient of variation for the concentration (CVC) to define the rate at which PM_2.5 changes at a certain height relative to the ground. The vertical variation model for PM_2.5 was developed based on a spline smoothing function in a generalized additive model (GAM) framework with an adjusted R² of 0.91 and explained 92.8% of the variance. PM_2.5 exposure levels for the population in the study area were estimated based on both the LUR models and the 3-D LUR framework. The 3-D LUR framework was found to improve the accuracy of exposure estimation in the vertical direction by avoiding exposure estimation errors of up to 5%. Although the 3-D LUR-based assessment did not indicate significant variation in estimates of premature mortality that could be attributed to PM_2.5, exposure to this pollutant was found to differ in the vertical direction. The 3-D LUR framework has the potential to provide accurate exposure estimates for use in future epidemiological studies and health risk assessments.

Insights from Application of a Hierarchical Spatio-Temporal Model to an Intensive Urban Black Carbon Monitoring Dataset

Existing regulatory pollutant monitoring networks rely on a small number of centrally located measurement sites that are purposefully sited away from major emission sources. While informative of general air quality trends regionally, these networks often do not fully capture the local variability of air pollution exposure within a community. Recent technological advancements have reduced the cost of sensors, allowing air quality monitoring campaigns with high spatial resolution. The 100×100 black carbon (BC) monitoring network deployed 100 low-cost BC sensors across the 15 km² West Oakland, CA community for 100 days in the summer of 2017, producing a nearly continuous site-specific time series of BC concentrations which we aggregated to one-hour averages. Leveraging this dataset, we employed a hierarchical spatio-temporal model to accurately predict local spatio-temporal concentration patterns throughout West Oakland, at locations without monitors (average cross-validated hourly temporal R ²=0.60). Using our model, we identified spatially varying temporal pollution patterns associated with small-scale geographic features and proximity to local sources. In a sub-sampling analysis, we demonstrated that fine scale predictions of nearly comparable accuracy can be obtained with our modeling approach by using ~30% of the 100×100 BC network supplemented by a shorter-term high-density campaign.

Flexible Bayesian Ensemble Machine Learning Framework for Predicting Local Ozone Concentrations

3D-grid-based chemical transport models, such as the Community Multiscale Air Quality (CMAQ) modeling system, have been widely used for predicting concentrations of ambient air pollutants. However, typical horizontal resolutions of nationwide CMAQ simulations (12 × 12 km2) cannot capture local-scale gradients for accurately assessing human exposures and environmental justice disparities. In this study, a Bayesian ensemble machine learning (BEML) framework, which integrates 13 learning algorithms, was developed for downscaling CMAQ estimates of ozone daily maximum 8 h averages to the census tract level, across the contiguous US, and was demonstrated for 2011. Three-stage hyperparameter tuning and targeted validations were designed to ensure the ensemble model's ability to interpolate, extrapolate, and capture concentration peaks. The Shapley value metric from coalitional game theory was applied to interpret the drivers of subgrid gradients. The flexibility (transferability) of the 2011-trained BEML model was further tested by evaluating its ability to estimate fine-scale concentrations for other years (2012-2017) without retraining. To demonstrate the feasibility of using the BEML approach to strictly "data-limited" situations, the model was applied to downscale CMAQ outputs for a future-year scenario-based simulation that considers effects of variations in meteorology associated with climate change.

Development and Performance Evaluation of a Low-Cost Portable PM2.5 Monitor for Mobile Deployment

The concentration of fine particulate matter (PM2.5) is known to vary spatially across a city landscape. Current networks of regulatory air quality monitoring are too sparse to capture these intra-city variations. In this study, we developed a low-cost (60 USD) portable PM2.5 monitor called Smart-P, for use on bicycles, with the goal of mapping street-level variations in PM2.5 concentration. The Smart-P is compact in size (85 × 85 × 42 mm) and light in weight (147 g). Data communication and geolocation are achieved with the cyclist’s smartphone with the help of a user-friendly app. Good agreement was observed between the Smart-P monitors and a regulatory-grade monitor (mean bias error: −3.0 to 1.5 μg m−3 for the four monitors tested) in ambient conditions with relative humidity ranging from 38 to 100%. Monitor performance decreased in humidity > 70% condition. The measurement precision, represented as coefficient of variation, was 6 to 9% in stationary mode and 6% in biking mode across the four tested monitors. Street tests in a city with low background PM2.5 concentrations (8 to 9 μg m−3) and in two cities with high background concentrations (41 to 74 μg m−3) showed that the Smart-P was capable of observing local emission hotspots and that its measurement was not sensitive to bicycle speed. The low-cost and user-friendly nature are two features that make the Smart-P a good choice for empowering citizen scientists to participate in local air quality monitoring.

Fine-scale modeling of the urban heat island: A comparison of multiple linear regression and random forest approaches

Characterizing the spatiotemporal variability of the Urban Heat Island (UHI) and its drivers is a key step in leveraging thermal comfort to create not only healthier cities, but also to enhance urban resilience to climate change. In this study, we developed specific daytime and nighttime multiple linear regression (MLR) and random forest (RF) models to analyze and predict the spatiotemporal evolution of the Urban Heat Island intensity (UHII), using the air temperature (T_air) as the response variable. We profited from the wealth of in situ T_air data and a comprehensive pool of predictors variables - including land cover, population, traffic, urban geometry, weather data and atmospheric vertical indices. Cluster analysis divided the study period into three main groups, each dominated by a combination of weather systems that, in turn, influenced the onset and strength of the UHII. Anticyclonic circulations favored the emergence of the largest UHII (hourly mean of 5.06 °C), while cyclonic circulations dampened its development. The MLR models were only able to explain a modest percentage of variance (64 and 34% for daytime and nighttime, respectively), which we interpret as part of their inability to capture key factors controlling T_air. The RF models, on the other hand, performed considerably better, with explanatory power over 96% of the variance for daytime and nighttime conditions, capturing and mapping the fine-scale T_air spatiotemporal variability in both periods and under each cluster condition. The feature importance analysis showed that the meteorological variables and the land cover were the main predictors of the T_air. Urban planners could benefit from these results, using the high-performing RF models as a robust framework for forecasting and mitigating the effects of the UHI.

Application of land use regression to map environmental noise in Shanghai, China

BACKGROUND

Urban environment noise has been linked with wide adverse effects on health; however, noise epidemiological researches are hindered by the lack of large-scale population-based exposure assessment.

OBJECTIVE

We aimed to measure noise levels over multiple seasons and to establish an LUR model to assess the spatial variability of intra-urban noise and identify its potential sources in Shanghai, China.

METHODS

Forty-minute (L_{Aeq, 40 min}) measurements of environmental noise were collected at 144 fixed sites, and each was visited three times (morning, afternoon, and evening) in winter, spring, and summer in 2019. Noise measurements were then integrated with land-use types, road networks, socioeconomic variables, and geographic information systems to construct LUR models. Ten-fold cross-validation was used to test the model performance.

RESULTS

A total of 1296 measurements and 29 predicting variables were used to estimate the spatial variation in environmental noise. The annual mean (±standard deviation) of L_{Aeq, 40min}, was 62 ± 8 dB (A). Significant variations were observed among monitoring sites but not between seasons or time of day. The LUR model explained 79% of the spatial variability of the noise, and the R² of the ten-fold cross-validation was 0.75. The most contributory predictors of noise level were road-related variables all within the 50-m buffers, followed by urban area within a 50-m buffer, total area of buildings within a 1000-m buffer, and number of restaurant clusters within a 50-m buffer. Farmland area within a 100-m buffer was the only negative variable in the model. A 50-m resolution noise prediction map was produced and suggested high noise level in urban areas and near traffic arteries.

CONCLUSION

LUR can be a robust method for reflecting noise variability in megacities such as Shanghai and may provide an efficient solution for noise exposure assessment in areas where noise maps are not available.

Deployment of a mobile platform to characterize spatial and temporal variation of on-road fine particles in an urban area

Traffic-related air pollutants (TRAPs) pose a serious health hazard for residents and commuters in urban areas. In this study, a real-time mobile monitoring system was deployed in Taipei, a typical East Asian city with an overlap of high population density, traffic, and special structures (e.g., viaducts), to capture the on-road TRAPs at different times of the day. In general, black carbon, ultrafine particles (UFPs), CO concentrations, and lung deposition surface area (LDSA) were positively correlated with traffic flow, and for PM_2.5, a more independent fluctuating concentration was observed. During rush-hour periods, the mean concentrations of UFPs, PM_2.5, and LDSA were 6.12 × 10⁴ ± 3.83 × 10⁴ cm^-3, 23 ± 8 μg/m³, and 2.29 × 10² ± 1.20 × 10² μm²/cm³, respectively. Additionally, the UFP number concentration and LDSA were two times higher along the high-traffic commuting route than along the lower traffic route. Pollutants tended to accumulate at sites near viaducts and high buildings and were significantly influenced by vehicle composition. In this study, the ratio of LDSA to total particle surface area concentration was used as an indicator of the degree of particle irregularity, which was directly related to aging during transport.

Identifying low-PM2.5 exposure commuting routes for cyclists through modeling with the random forest algorithm based on low-cost sensor measurements in three Asian cities

Cyclists can be easily exposed to traffic-related pollutants due to riding on or close to the road during commuting in cities. PM_2.5 has been identified as one of the major pollutants emitted by vehicles and associated with cardiopulmonary and respiratory diseases. As routing has been suggested to reduce the exposures for cyclists, in this study, PM_2.5 was monitored with low-cost sensors during commuting periods to develop models for identifying low exposure routes in three Asian cities: Taipei, Osaka, and Seoul. The models for mapping the PM_2.5 in the cities were developed by employing the random forest algorithm in a two-stage modeling approach. The land use features to explain spatial variation of PM_2.5 were obtained from the open-source land use database, OpenStreetMap. The total length of the monitoring routes ranged from 101.36 to 148.22 km and the average PM_2.5 ranged from 13.51 to 15.40 μg/m³ among the cities. The two-stage models had the standard k-fold cross-validation (CV) R² of 0.93, 0.74, and 0.84 in Taipei, Osaka, and Seoul, respectively. To address spatial autocorrelation, a spatial cross-validation approach applying a distance restriction of 100 m between the model training and testing data was employed. The over-optimistic estimates on the predictions were thus prevented, showing model CV-R² of 0.91, 0.67, and 0.78 respectively in Taipei, Osaka, and Seoul. The comparisons between the shortest-distance and lowest-exposure routes showed that the largest percentage of reduced averaged PM_2.5 exposure could reach 32.1% with the distance increases by 37.8%. Given the findings in this study, routing behavior should be encouraged. With the daily commuting trips expanded, the cumulative effect may become significant on the chronic exposures over time. Therefore, a route planning tool for reducing the exposures shall be developed and promoted to the public.

National Empirical Models of Air Pollution Using Microscale Measures of the Urban Environment

National-scale empirical models of air pollution (e.g., Land Use Regression) rely on predictor variables (e.g., population density, land cover) at different geographic scales. These models typically lack microscale variables (e.g., street level), which may improve prediction with fine-spatial gradients. We developed microscale variables of the urban environment including Point of Interest (POI) data, Google Street View (GSV) imagery, and satellite-based measures of urban form. We developed United States national models for six criteria pollutants (NO₂, PM_2.5, O₃, CO, PM₁₀, SO₂) using various modeling approaches: Stepwise Regression + kriging (SW-K), Partial Least Squares + kriging (PLS-K), and Machine Learning + kriging (ML-K). We compared predictor variables (e.g., traditional vs microscale) and emerging modeling approaches (ML-K) to well-established approaches (i.e., traditional variables in a PLS-K or SW-K framework). We found that combined predictor variables (traditional + microscale) in the ML-K models outperformed the well-established approaches (10-fold spatial cross-validation (CV) R² increased 0.02-0.42 [average: 0.19] among six criteria pollutants). Comparing all model types using microscale variables to models with traditional variables, the performance is similar (average difference of 10-fold spatial CV R² = 0.05) suggesting microscale variables are a suitable substitute for traditional variables. ML-K and microscale variables show promise for improving national empirical models.

Development of season-dependent land use regression models to estimate BC and PM1 exposure

Reliable estimation of exposure to black carbon (BC) and sub-micrometer particles (PM₁) within a city is challenging because of limited monitoring data as well as the lack of models suitable for assessing the intra-urban environment. In this study, to estimate exposure levels in the inner-city area, we developed land use regression (LUR) models for BC and PM₁ based on specially designed mobile monitoring surveys conducted in 2019 and 2020 for three seasons. The daytime and nighttime LUR models were developed separately to capture additional details on the variation in pollutants. The results of mobile monitoring indicated similar temporal variation characteristics of BC and PM₁. The mean concentrations of pollutants were higher in winter (BC: 4.72 μg/m³; PM₁: 56.97 μg/m³) than in fall (BC: 3.74 μg/m³; PM₁: 33.29 μg/m³) and summer (BC: 2.77 μg/m³; PM₁: 27.04 μg/m³). For both BC and PM₁, higher nighttime concentrations were found in winter and fall, whereas higher daytime concentrations were observed in the summer. A supervised forward stepwise regression method was used to select the predictors for the LUR models. The adjusted R² of the LUR models for BC and PM₁ ranged from 0.39 to 0.66 and 0.45 to 0.80, respectively. Traffic-related predictors were incorporated into all the models for BC. In contrast, more meteorology-related predictors were incorporated into the PM₁ models. The concentration surface based on the LUR models was mapped at a spatial resolution of 100 m, and significant seasonal and diurnal trends were observed. PM₁ was dominated by seasonal variations, whereas BC showed more spatial variation. In conclusion, the development of season-dependent diurnal LUR models based on mobile monitoring could provide a methodology for the estimation of exposure and screening of influencing factors of BC and PM₁ in typical inner-city environments, and support pollution management.

Improving emissions inputs via mobile measurements to estimate fine-scale Black Carbon monthly concentrations through geostatistical space-time data fusion

Isolating air pollution sources in a complex transportation environment to quantify their contribution is challenging, particularly with sparse stationary measurements. Mobile measurements can add finer spatial resolution to support source apportionment, but they exhibit limitations when characterizing long term concentrations. Dispersion models can help overcome these limitations. However, they are only as reliable as their input emissions inventories. Herein, we developed an innovative method to revise emissions through inverse modeling and improve dispersion modeling predictions using stationary/mobile measurements. One specific revision estimated an adjustment factor of ~306 for warehouse emissions, indicating a significant underestimation of our initial estimates. This revised emission rate scaled up nationally would correspond to ~3.5% of the total Black Carbon emissions in the U.S. Nevertheless, domain-specific revisions only contribute to a 4% increase of area source emissions while improving R² from monthly estimates at fixed sites by 38%. After revising emissions through inverse dispersion modeling, we combine this model with stationary/mobile measurements through Bayesian Maximum Entropy (I-DISP BME) to produce temporally coarse yet spatially fine data fusion. We compare this novel data fusion approach to BME using only measurements (Flat BME). A 10-fold conventional cross-validation (representative of months with mobile measurements) shows that all BME methods have R² values that range from 0.787 to 0.798. A 2-fold cross-validation (representative of months with no mobile measurements) shows that the R² for I-DISP BME increases by a factor 90 when compared to Flat BME. Furthermore, not only is our novel I-DISP BME method more accurate than the classic Flat BME method, but the area it detects as highly exposed can be up to 5 times larger than that detected by the less accurate Flat BME method.

Modelling nationwide spatial variation of ultrafine particles based on mobile monitoring

BACKGROUND

Large nation- and region-wide epidemiological studies have provided important insights into the health effects of long-term exposure to outdoor air pollution. Evidence from these studies for the long-term effects of ultrafine particles (UFP), however is lacking. Reason for this is the shortage of empirical UFP land use regression models spanning large geographical areas including cities with varying topographies, peri-urban and rural areas. The aim of this paper is to combine targeted mobile monitoring and long-term regional background monitoring to develop national UFP models.

METHOD

We used an electric car to monitor UFP concentrations in selected cities and towns across the Netherlands over a 14-month period in 2016-2017. Routes were monitored 3 times and concentrations were averaged per road segment. In addition, we used kriging maps based on regional background monitoring (20 sites; 3 × 2 weeks) over the same period to assess annual average regional background concentrations. All road segments were used to model spatial variation of UFP with three different land-use (regression) approaches: supervised stepwise regression, LASSO and random forest. For each approach, we also tested a deconvolution method, which segregates the average concentration at each road segment into a local and background signal. Model performance was evaluated with short-term (400 sites across the Netherlands; 3 × 30 minutes) and external longer-term measurements (42 sites in two major cities; 3 × 24 hours). We also compared predictions of all six models at 1000 random addresses spread over the country.

RESULTS

We found similar predictive performance for the six models, with validation R² values from 0.25 to 0.35 for short-term measurements and 0.52 to 0.60 for longer-term external measurements. Models with and without deconvolution had similar predictive performance. All models based on the deconvolution method included a regional background kriging map as important predictor. Correlations between predictions at random addresses were high with Pearson correlations from 0.84 to 0.99. Models overestimated exposure at the short-term and long-term sites by about 20-30% in all cases, with small differences between regions and road types.

CONCLUSION

We developed robust nation-wide models for long-term UFP exposure combining mobile monitoring with long-term regional background monitoring. Minor differences in predictive performance between different algorithms were found, but the deconvolution approach is considered more physically realistic. The models will be applied in Dutch nation-wide health studies.

High-Spatial-Resolution Estimates of Ultrafine Particle Concentrations across the Continental United States

There is growing evidence that ultrafine particles (UFP; particles smaller than 100 nm) are likely more toxic than larger particles. However, the health effects of UFP remain uncertain due in part to the lack of large-scale population-based exposure assessment. We develop a national-scale empirical model of particle number concentration (PNC; a measure of UFP) using data from mobile monitoring and fixed sites across the United States and a land-use regression (LUR) modeling framework. Traffic, commercial land use, and urbanicity-related variables explain much of the spatial variability of PNC (base model R² = 0.77, RMSE = 2400 cm^-3). Model predictions are robust across a diverse set of evaluations [random 10-fold holdout cross-validation (HCV): R² = 0.72, RMSE = 2700 cm^-3; spatially defined HCV: R² = 0.66, RMSE = 3000 cm^-3; evaluation against an independent data set: R² = 0.54, RMSE = 2600 cm^-3]. We apply our model to predict PNC at ∼6 million residential census blocks in the contiguous United States. Our estimates are annual average concentrations for 2016-2017. The predicted national census-block-level mean PNC ranges between 1800 and 26 600 cm^-3 (population-weighted average: 6500 cm^-3), with hotspots in cities and near highways. Our national PNC model predicts large urban-rural, intra-, and inter-city contrasts. PNC and PM_2.5 are moderately correlated at the city scale, but uncorrelated at the regional/national scale. Our high-spatial-resolution national PNC estimates are useful for analyzing population exposure (socioeconomic disparity, epidemiological health impact) and environmental policy and regulation.

Mapping Urban Air Quality from Mobile Sensors Using Spatio-Temporal Geostatistics

With the advancement of technology and the arrival of miniaturized environmental sensors that offer greater performance, the idea of building mobile network sensing for air quality has quickly emerged to increase our knowledge of air pollution in urban environments. However, with these new techniques, the difficulty of building mathematical models capable of aggregating all these data sources in order to provide precise mapping of air quality arises. In this context, we explore the spatio-temporal geostatistics methods as a solution for such a problem and evaluate three different methods: Simple Kriging (SK) in residuals, Ordinary Kriging (OK), and Kriging with External Drift (KED). On average, geostatistical models showed 26.57% improvement in the Root Mean Squared Error (RMSE) compared to the standard Inverse Distance Weighting (IDW) technique in interpolating scenarios (27.94% for KED, 26.05% for OK, and 25.71% for SK). The results showed less significant scores in extrapolating scenarios (a 12.22% decrease in the RMSE for geostatisical models compared to IDW). We conclude that univariable geostatistics is suitable for interpolating this type of data but is less appropriate for an extrapolation of non-sampled places since it does not create any information.

Assessing the Impact of Wildfires on the Use of Black Carbon as an Indicator of Traffic Exposures in Environmental Epidemiology Studies

Epidemiological studies frequently use black carbon (BC) as a proxy for traffic-related air pollution (TRAP). However, wildfire smoke (WFS) represents an important source of BC not often considered when using BC as a proxy for TRAP. Here, we examined the potential for WFS to bias TRAP exposure assessments based on BC measurements. Weekly integrated BC samples were collected across the Denver, CO region from May to November 2018. We collected 609 filters during our sampling campaigns, 35% of which were WFS-impacted. For each filter we calculated an average BC concentration. We assessed three GIS-based indicators of TRAP for each sampling location: annual average daily traffic within a 300 m buffer, the minimum distance to a highway, and the sum of the lengths of roadways within 300 m. Median BC concentrations were 9% higher for WFS-impacted filters (median = 1.14 μg/m3, IQR = 0.23 μg/m3) than nonimpacted filters (median = 1.04 μg/m3, IQR = 0.48 μg/m3). During WFS events, BC concentrations were elevated and expected spatial gradients in BC were reduced. We conducted a simulation study to estimate TRAP exposure misclassification as the result of regional WFS. Our results suggest that linear health effect estimates were biased away from the null when WFS was present. Thus, exposure assessments relying on BC as a proxy for TRAP may be biased by wildfire events. Alternative metrics that account for the influence of "brown" carbon associated with biomass burning may better isolate the effects of traffic emissions from those of other black carbon sources.

Characterizing long-term NO2 concentration surfaces across a large metropolitan area through spatiotemporal land use regression modelling of mobile measurements

A spatiotemporal land use regression (LUR) model optimized to predict nitrogen dioxide (NO₂) concentrations obtained from on-road, mobile measurements collected in 2015-16 was independently evaluated using concentrations observed at multiple sites across Toronto, Canada, obtained more than ten years earlier. This spatiotemporal LUR modelling approach improves upon estimates of historical NO₂ concentrations derived from the previously used method of back-extrapolation. The optimal spatiotemporal LUR model (R² = 0.71 for prediction of NO₂ data in 2002 and 2004) uses daily average NO₂ concentrations observed at multiple long-term monitoring sites and hourly average wind speed recorded at a single site, along with spatial predictors based on geographical information system data, to estimate NO₂ levels for time periods outside of those used for model development. While the model tended to underestimate samplers located close to the roadway, it showed great accuracy when estimating samplers located beyond 100 m which are probably more relevant for exposure at residences. This study shows that spatiotemporal LUR models developed from strategic, multi-day (30 days in 3 different months) mobile measurements can enhance LUR model's ability to estimate long-term, intra-urban NO₂ patterns. Furthermore, the mobile sampling strategy enabled this new LUR model to cover a larger domain of Toronto and outlying suburban communities, thereby increasing the potential population for future epidemiological studies.

Using Street View Imagery to Predict Street-Level Particulate Air Pollution

Land-use regression (LUR) models are frequently applied to estimate spatial patterns of air pollution. Traditional LUR often relies on fixed-site measurements and GIS-derived variables with limited spatial resolution. We present an approach that leverages Google Street View (GSV) imagery to predict street-level particulate air pollution (i.e., black carbon [BC] and particle number [PN] concentrations). We developed empirical models based on mobile monitoring data and features extracted from ∼52 500 GSV images using a deep learning model. We tested theory- and data-driven feature selection methods as well as models using images within varying buffer sizes (50-2000 m). Compared to LUR models with traditional variables, our models achieved similar model performance using the street-level predictors while also identifying additional potential hotspots. Adjusted R² (10-fold CV R²) with integrated feature selection was 0.57-0.64 (0.50-0.57) and 0.65-0.73 (0.61-0.66) for BC and PN models, respectively. Models using only features near the measurement locations (i.e., GSV images within 250 m) explained ∼50% of air pollution variability, indicating PN and BC are strongly affected by the street-level built environment. Our results suggest that GSV imagery, processed with computer vision techniques, is a promising data source to develop LUR models with high spatial resolution and consistent predictor variables across administrative boundaries.

Spatiotemporal Characteristics and Driving Factors of Black Carbon in Augsburg, Germany: Combination of Mobile Monitoring and Street View Images

The study investigates the spatial pattern of black carbon (BC) at a high spatial resolution in Augsburg, Germany. Sixty two walks were performed to assess the concentrations of equivalent black carbon (eBC), ultraviolet particulate matter (UVPM), and equivalent brown carbon (eBrC) in different seasons and at different times of the day with a mobile platform (i.e., trolley). Along with BC measurements, images of street microenvironments were recorded. Meteorological parameters, including temperature, relative humidity, and wind speed, were monitored. The BC concentrations showed significant spatial heterogeneity and diurnal variations peaking in the morning and at night. The highest BC concentrations were observed near dense traffic. The correlations between BC and street views (buildings, roads, cars, and vegetation) were weak but highly significant. Moreover, meteorological factors also influenced the BC concentration. A model based on street view images and meteorological data was developed to examine the driving factors of the spatial variability of BC concentrations at a higher spatial resolution as different microenvironments based on traffic density. The best results were obtained for UVPM and eBC (71 and 70% explained variability). eBrC (53%), to which other sources besides road traffic can also make significant contributions, is modeled less well.

Spatial regression modelling of particulate pollution in Calgary, Canada

The study presents a spatial analysis of particulate pollution, which includes not only particulate matter, but also black carbon, a pollutant of growing concern for human health. We developed land use regression (LUR) models for two particulate matter size fractions, PM_2.5 and PM₁₀, and for δC, an index calculated from black carbon (BC)-a component of PM_2.5-which indicates the portion of organic versus elemental BC. LUR models were estimated over Calgary (Canada) for summer 2015 and winter 2016. As all samples exhibited significant spatial autocorrelation, spatial autoregressive lag (SARlag) and error (SARerr) models were computed. SARlag models were preferred for all pollutants in both seasons, and yielded goodness of fit aligned with or higher than values reported in the literature. LUR models yielded consistent sets of predictors, representing industrial activities, traffic, and elevation. The obtained model coefficients were then combined with local land use variables to compute fine-scale concentration predictions over the entire city. The predicted concentrations were slightly lower and less dispersed than the observed ones. Consistent with observed pollution records, prediction maps exhibited higher concentration over the road network, industrial areas, and the eastern quadrants of the city. Lastly, results of a corresponding study of PM in summer 2010 and winter 2011 were considered. While the small size of the 2010-2011 sample hampered a multi-temporal analysis, we cautiously note comparable seasonal patterns and consistent association with land use variables for both PM fine fractions over the 5-year interval.

Potential for developing independent daytime/nighttime LUR models based on short-term mobile monitoring to improve model performance

Land use regression model (LUR) is a widespread method for predicting air pollution exposure. Few studies have explored the performance of independently developed daytime/nighttime LUR models. In this study, fine particulate matter (PM_2.5), inhalable particulate matter (PM₁₀), and nitrogen dioxide (NO₂) concentrations were measured by mobile monitoring during non-heating and heating seasons in Taiyuan. Pollutant concentrations were higher in the nighttime than the daytime, and higher in the heating season than the non-heating season. Daytime/nighttime and full-day LUR models were developed and validated for each pollutant to examine variations in model performance. Adjusted coefficients of determination (adjusted R²) for the LUR models ranged from 0.53-0.87 (PM_2.5), 0.53-0.85 (PM₁₀), and 0.33-0.67 (NO₂). The performance of the daytime/nighttime LUR models for PM_2.5 and PM₁₀ was better than that of the full-day models according to the results of model adjusted R² and validation R². Consistent results were confirmed in the non-heating and heating seasons. Effectiveness of developing independent daytime/nighttime models for NO₂ to improve performance was limited. Surfaces based on the daytime/nighttime models revealed variations in concentrations and spatial distribution. In conclusion, the independent development of daytime/nighttime LUR models for PM_2.5/PM₁₀ has the potential to replace full-day models for better model performance. The modeling strategy is consistent with the residential activity patterns and contributes to achieving reliable exposure predictions for PM_2.5 and PM₁₀. Nighttime could be a critical exposure period, due to high pollutant concentrations.

Development and transferability of ultrafine particle land use regression models in London

Due to a lack of routine monitoring, bespoke measurements are required to develop ultrafine particle (UFP) land use regression (LUR) models, which is especially challenging in megacities due to their large area. As an alternative, for London, we developed separate models for three urban residential areas, models combining two areas, and models using all three areas. Models were developed against annual mean ultrafine particle count cm^-3 estimated from repeated 30-min fixed-site measurements, in different seasons (2016-2018), at forty sites per area, that were subsequently temporally adjusted using continuous measurements from a single reference site within or close to each area. A single model and 10 models were developed for each individual area and combination of areas. Within each area, sites were split into 10 groups using stratified random sampling. Each of the 10 models were developed using 90% of sites. Hold-out validation was performed by pooling the 10% of sites held-out each time. The transferability of models was tested by applying individual and two-area models to external area(s). In model evaluation, within-area mean squared error (MSE) R² ranged from 14% to 48%. Transferring individual- and combined-area models to external areas without calibration yielded MSE-R² ranging from -18 to 0. MSE-R² was in the range 21% to 41% when using particle number count (PNC) measurements in external areas to calibrate models. Our results suggest that the UFP models could be transferred to other areas without calibration in London to assess relative ranking in exposures but not for estimating absolute values of PNC.

Data fusion for air quality mapping using low-cost sensor observations: Feasibility and added-value

In Nantes, low-cost sensors were installed in the city center and deployed on driving school cars, ambulances and service vehicles to measure PM₁₀ concentrations. This work aims to use the large amount of observations provided by the sensors for air quality mapping at the urban scale in order to show the potential added-value with respect to the dispersion model (ADMS-Urban) calculations. A preprocessing is applied to the raw sensor dataset to remove the unreliable observations based on an outlier detection technique and to compensate for the measurement drift by adjusting the estimated daily variation of the underlying background PM₁₀ concentrations with the reference stations. Then, data fusion is performed by combining the preprocessed fixed and mobile low-cost sensor observations and the 2016 annual average of the ADMS-Urban outputs. The measurement uncertainty related to the low-cost sensors and the dispersion of the data are considered in data fusion as the Variance of Measurement Error (VME). The spatial interpolation is achieved at hourly resolution and results are presented for November 29th, 2018 from 7 am to 7 pm. Hourly fused maps show disparate responses to data fusion mainly depending on the variability of the sensor data and the correlation between the sensor observations and the drift. The data fusion performance has been investigated by comparing the daily average of the estimated concentrations, the reference observations and the hourly model outputs at each station of the Air Pays de la Loire network. Results show that considering the model alone implies 8% bias whereas including the LCS observations reduces the bias to 2.5%. However, the concentration distributions related to the data fusion are characterized by a lower dispersion than the reference observations and the model estimation. Thus, the fusion smooths the PM₁₀ peaks. In addition, the effect of the measurement uncertainty has been investigated by doubling it or reducing it to the reference station measurement uncertainty. The sensitivity study demonstrates that the performance is increasing by reducing the uncertainty. This highlights the importance to estimate accurately the measurement uncertainty of the devices to ensure relevant air quality mapping. The method efficiency is also quite limited by the low correlation between the sensor observations and the model used as external drift in the kriging that may be explained by the remaining bias on LCS data. Efforts on this issue might increase the performance of the spatial interpolation.

A gradient boost approach for predicting near-road ultrafine particle concentrations using detailed traffic characterization

This study investigates the influence of meteorology, land use, built environment, and traffic characteristics on near-road ultrafine particle (UFP) concentrations. To achieve this objective, minute-level UFP concentrations were measured at various locations along a major arterial road in the Greater Toronto Area (GTA) between February and May 2019. Each location was visited five times, at least once in the morning, mid-day, and afternoon. Each visit lasted for 30 min, resulting in 2.5 h of minute-level data collected at each location. Local traffic information, including vehicle class and turning movements, were processed using computer vision techniques. The number of fast-food restaurants, cafes, trees, traffic signals, and building footprint, were found to have positive impacts on the mean UFP, while distance to the closest major road was negatively associated with UFP. We employed the Extreme Gradient Boosting (XGBoost) method to develop prediction models for UFP concentrations. The Shapley additive explanation (SHAP) measures were used to capture the influence of each feature on model output. The model results demonstrated that minute-level counts of local traffic from different directions had significant impacts on near-road UFP concentrations, model performance was robust under random cross-validation as coefficients of determination (R²) ranged from 0.63 to 0.69, but it revealed weaknesses when data at specific locations were eliminated from the training dataset. This result indicates that proper cross-validation techniques should be developed to better evaluate machine learning models for air quality predictions.

Comparison of Machine Learning and Land Use Regression for fine scale spatiotemporal estimation of ambient air pollution: Modeling ozone concentrations across the contiguous United States

BACKGROUND

Spatial linear Land-Use Regression (LUR) is commonly used for long-term modeling of air pollution in support of exposure and epidemiological assessments. Machine Learning (ML) methods in conjunction with spatiotemporal modeling can provide more flexible exposure-relevant metrics and have been studied using different model structures. There is however a lack of comparisons of methods available within these two modeling frameworks, that can guide model/algorithm selection in air quality epidemiology.

OBJECTIVE

The present study compares thirteen algorithms for spatial/spatiotemporal modeling applied for daily maxima of 8-hour running averages of ambient ozone concentrations at spatial resolutions corresponding to census tracts, to support estimation of annual ozone design values across the contiguous US. These algorithms were selected from nine representative categories and trained using predictors that included chemistry-transport model predictions, meteorological factors, land use and land cover, and stationary and mobile emissions.

METHODS

To obtain the best predictive performance, model structures were optimized through a repeated coarse/fine grid search with expert knowledge. Six target-oriented validation strategies were used to prevent overfitting and avoid over-optimistic model evaluation results. In order to take full advantage of the power of different algorithms, we introduced tuning sample weights in spatiotemporal modeling to ensure predictive accuracy of peak concentrations, that is crucial for exposure assessments. In spatial modeling, four interpretation and visualization tools were introduced to explain predictions from different algorithms.

RESULTS

Nonlinear ML methods achieved higher prediction accuracy than linear LUR, and the improvements were more significant for spatiotemporal modeling (nearly 10%-40% decrease of predicted RMSE). By tuning the sample weights, spatiotemporal models can predict concentrations used to calculate ozone design values that are comparable or even better than spatial models (nearly 30% decrease of cross-validated RMSE). We visualized the underlying nonlinear relationships, heterogeneous associations and complex interactions from the two best performing ML algorithms, i.e., Random Forest and Extreme Gradient Boosting, and found that the complex patterns were relatively less significant with respect to model accuracy for spatial modeling.

CONCLUSION

Machine Learning can provide estimates that are actually more interpretable and practical than linear regression to improve accuracy in modeling human exposures. A careful design of hyperparameter tuning and flexible data splitting and validations is crucial to obtain reliable and stable results. Desirable/successful nonlinear models are expected to capture similar nonlinear patterns and interactions using different ML algorithms.