CV R2 Research Articles

Impact of Urban Neighborhood Morphology on PM2.5 Concentration Distribution at Different Scale Buffers

PM2.5 air pollution is a critical global health issue. This paper introduces an innovative framework to explore the multi-scale relationship between urban morphology and PM2.5 concentrations. An enhanced Land Use Regression (LUR) model integrates geographic, architectural, and visual factors, enabling analysis from neighborhood to regional scales. A stratified sampling strategy, combined with standardized mobile monitoring and fixed-site data, establishes a robust and verifiable data collection methodology. Cross-validation (CV R2 > 0.70) further confirms the model’s reliability and robustness. The nested buffer analysis reveals scale-dependent effects of urban morphology on PM2.5 concentrations, providing quantitative evidence for planning interventions. Quantitative analysis shows land use (β = 0.42, p < 0.01), visual factors (β = 0.38, p < 0.01), and building density (β = 0.35, p < 0.01) in descending order of influence. Geographic factors are significant at the regional scale (2000–3000 m) while architectural parameters dominate at the neighborhood scale (50–500 m), informing both macro-scale spatial optimization and micro-scale design. This framework, through standardized parameters and reproducible procedures, supports cross-regional and cross-scale air quality assessments, providing quantitative metrics for urban planning, neighborhood optimization, and public space design.

Integrating Doppler LiDAR and machine learning into land-use regression model for assessing contribution of vertical atmospheric processes to urban PM2.5 pollution

Air pollution has been recognized as a global issue, through adverse effects on environment and health. While vertical atmospheric processes substantially affect urban air pollution, traditional epidemiological research using Land-use regression (LUR) modeling usually focused on ground-level attributes without considering upper-level atmospheric conditions. This study aimed to integrate Doppler LiDAR and machine learning techniques into LUR models (LURF-LiDAR) to comprehensively evaluate urban air pollution in Hong Kong, and to assess complex interactions between vertical atmospheric processes and urban air pollution from long-term (i.e., annual) and short-term (i.e., two air pollution episodes) views in 2021. The results demonstrated significant improvements in model performance, achieving CV R2 values of 0.81 (95 % CI: 0.75–0.86) for the long-term PM2.5 prediction model and 0.90 (95 % CI: 0.87–0.91) for the short-term models. Approximately 69 % of ground-level air pollution arose from the mixing of ground- and lower-level (105 m–225 m) particles, while 21 % was associated with upper-level (825 m–945 m) atmospheric processes. The identified transboundary air pollution (TAP) layer was located at ~900 m above the ground. The identified Episode one (E1: 7 Jan–22 Jan) was induced by the accumulation of local emissions under stable atmospheric conditions, whereas Episode two (E2: 13 Dec–24 Dec) was regulated by TAP under instable and turbulent conditions. Our improved air quality prediction model is accurate and comprehensive with high interpretability for supporting urban planning and air quality policies.

Estimation of Daily Seamless PM2.5 Concentrations with Climate Feature in Hubei Province, China

The urgent necessity for precise and uninterrupted PM2.5 datasets of high spatial–temporal resolution is underscored by the significant influence of PM2.5 on weather, climate, and human health. This study leverages the AOD reconstruction method to compensate for missing values in the MAIAC AOD throughout Hubei Province. The reconstructed AOD dataset, exhibiting an R2/RMSE of 0.76/0.18, compared to AERONET AOD, was subsequently used for PM2.5 estimation. Our research breaks from traditional methodologies that solely depend on latitude and longitude information. Instead, it emphasizes the use of climate feature as an input for estimating PM2.5 concentrations. This strategic approach prevents potential spatial discontinuities triggered by geolocation information (latitude and longitude), thus ensuring the precision of the PM2.5 estimation (sample/spatial CV R2 = 0.91/0.88). Moreover, we proposed a method for identifying the absolute feature importance of machine-learning models. Contrasted with the relative feature-importance property typical of machine-learning models (a minor difference in the order of top three between geolocation-based and climate-feature-based models, and the slight difference in the top three: 0.08%/0.17%), our method provides a more comprehensive explanation of the absolute significance of features to the model (maintaining the same order and a larger difference in the top three: 0.99%/0.72%). Crucially, our findings demonstrated that AOD reconstruction can mitigate the overestimation of annual mean PM2.5 concentrations (ranging from 0.52 to 9.28 µg/m3). In addition, the seamless PM2.5 dataset contributes to reducing the bias in exposure risk assessment (ranging from −0.11 to 9.81 µg/m3).

High-resolution modeling for criteria air pollutants and the associated air quality index in a metropolitan city

The Air Quality Index (AQI), which jointly accounts for levels of criteria air pollutants relative to their guidelines, is largely reported at the city level. Little is known about the spatial patterns of the AQI in terms of the magnitude, temporal variability, and predominant air pollutant contributions at the hyperlocal scale within a city. To fill this research gap, we developed spatiotemporal models for each criteria air pollutant based on an advanced geostatistical framework and estimated daily AQI levels at 100-meter resolution in a metropolitan city in 2019. The model prediction ability (cross-validation, CV, Coefficient of determination, R2, and root mean square error, RMSE) ranged from 0.43 and 1.86 µg/m3 for sulfur dioxide (SO2) to 0.92 and 6.25 µg/m3 for fine particulate matter (PM2.5) across the six air pollutants, leading to good performance in the subsequent AQI estimations (CV R2 = 0.86, RMSE = 10.05). The AQI varies substantially over space at a fine scale and differs from the distributions of individual air pollutants. The unhealthy air quality (AQI > 100 over 75 days) spatial pattern was dominated by excessive ground-level ozone exposure in a large area. Our research provides a useful tool for accurately estimating AQI spatiotemporal variations for population health studies.

Estimating the CO2 emissions of Chinese cities from 2011 to 2020 based on SPNN-GNNWR

Global warming is a serious threat to human survival and health. Facing increasing global warming, the issue of CO2 emissions has attracted more attention. China is a major contributor of anthropogenic CO2 emissions and so it is essential to accurately estimate China's CO2 emissions and analyze their changing characteristics. This study recalculates CO2 emissions from Chinese cities from 2011 to 2020 using the SPNN-GNNWR model and multiple factors to reduce the uncertainty in emission estimates. The SPNN-GNNWR model has excellent predictions (R2: 0.925, 10-fold CV R2: 0.822) when cross-validation is used. The results indicate that the total CO2 emissions in China calculated by the model are close to those accounted for by other authorities in the world, with the total CO2 emissions increasing from 9.122 billion tonnes in 2011 to 9.912 billion tonnes in 2020. The city with the largest increase in CO2 emissions is Tianjin, and the city with the largest decrease is Beijing. The study also reveals the regional differences in CO2 emissions in Chinese mainland, including emissions, emission intensity and per capita emissions. Capturing and understanding the emissions and the related socioeconomic characteristics of different cities can help to develop effective emission mitigation strategies.

Adaptabilities of Water Production Function Models for Rice in Cold and Black Soil Region of China

Crop water production function models (WPFMs) are required methods to study the relationships between yield and water consumption under regulated deficit irrigation (RDI). In this study, a pot experiment was established to study the effect of water deficit during both individual growth stages and across two consecutive growth stages of rice on yield, water consumption, and water use efficiency (WUE) in 2017 and 2018. Light, medium, and severe water deficits were set as 80~90%, 70~80%, and 60~70% of saturated soil moisture content, respectively. The accuracies of five WPFMs were tested based on the experimental results. The results showed that yields and WUE of a light water deficit were higher than those of medium and severe water deficits at each growth stage. The yields and WUE of light drought stress treatments in the flowering and milky stages were higher than the saturated soil moisture control by 4~7.4% and 5.3~20.6%, respectively. Water consumption decreased with increasing water deficit across two consecutive growth stages. The Minhas model had the highest simulation accuracy of the five WPFMs, with relatively lower AE, RMSE, Cv, CRM, and higher R2, which were 0.0002, 0.0634, 6.9965, 0.0002, and 0.9951 in 2017 and 0.0110, 0.0760, 8.9882, 0.0131, and 0.9923 in 2018, respectively. The sensitivity indices for the Minhas model more accurately reflected the sensitivity of rice yield to water deficit at different growth stages in 2017 and 2018, compared with the Jensen model, Stewart model, Blank model, and Singh model. Rice yield was most sensitive to water deficit at the jointing and booting stage. The results indicate that the Minhas model is the most suitable WPFM for guiding rice irrigation practices in cold and black soil regions of China.

An Ensemble Model-Based Estimation of Nitrogen Dioxide in a Southeastern Coastal Region of China

NO2 (nitrogen dioxide) is a common pollutant in the atmosphere that can have serious adverse effects on the health of residents. However, the existing satellite and ground observation methods are not enough to effectively monitor the spatiotemporal heterogeneity of near-surface NO2 concentrations, which limits the development of pollutant remediation work and medical health research. Based on TROPOMI-NO2 tropospheric column concentration data, supplemented by meteorological data, atmospheric condition reanalysis data and other geographic parameters, combined with classic machine learning models and deep learning networks, we constructed an ensemble model that achieved a daily average near-surface NO2 of 0.03° exposure. In this article, a meteorological hysteretic effects term and a spatiotemporal term were designed, which considerably improved the performance of the model. Overall, our ensemble model performed better, with a 10-fold CV R2 of 0.89, an RMSE of 5.62 µg/m3, and an MAE of 4.04 µg/m3. The model also had good temporal and spatial generalization capability, with a temporal prediction R2 and a spatial prediction R2 of 0.71 and 0.81, respectively, which can be applied to a wider range of time and space. Finally, we used an ensemble model to estimate the spatiotemporal distribution of NO2 in a coastal region of southeastern China from May 2018 to December 2020. Compared with satellite observations, the model output results showed richer details of the spatiotemporal heterogeneity of NO2 concentrations. Due to the advantages of using multi-source data, this model framework has the potential to output products with a higher spatial resolution and can provide a reference for downscaling work on other pollutants.

Historically understanding the spatial distributions of particle surface area concentrations over China estimated using a non-parametric machine learning method

A non-parametric ensemble model was proposed to estimate the long-term (2015–2019) particle surface area concentrations (SA) over China for the first time on basis of a vilification dataset of measured particle number size distribution. This ensemble model showed excellent cross-validation R2 value (CV R2 = 0.83) as well as a relatively low root-mean-square error (RMSE = 195.0 μm2/cm3). No matter in which year, considerable spatial heterogeneity of SA was found over China with higher SA in Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Middle Lower Reaches of Yangtze River (MLYR). From 2015 to 2019, SA significantly decreased in representative city clusters. The reduction rates were 140.1 μm2·cm−3·a−1 in BTH, 110.7 μm2·cm−3·a−1 in Pearl River Delta (PRD), 105.2 μm2·cm−3·a−1 in YRD, and 92.4 μm2·cm−3·a−1 in Sichuan Basin (SCB), respectively. Even though such quick reduction, high SA (ranged from ~800 μm2/cm3 to ~1750 μm2/cm3) during the heavy pollution period (PM2.5 > 75 μg/m3) still existed in the above-mentioned city clusters and may provide rich reaction vessels for multiphase chemistry. A dichotomy of enhanced annual 4th maximum daily 8-h average O3 concentrations (4MDA8 O3) and decreased SA during summertime was found in Shanghai, a representative city of YRD. In Chengdu (SCB), increased 4MDA8 O3 concentration was associated with a synchronous increase of SA from 2017 to 2019. Differently, 4MDA8 O3 concentrations enhanced in Beijing (BTH) and Guangzhou (PRD), while not significant for SA before 2018. This work will greatly deepen our understanding of the historical variation and spatial distributions of SA over China.

Estimating long-term PM10-2.5 concentrations in six US cities using satellite-based aerosol optical depth data

A major challenge in assessing the health risks of PM10-2.5 is the limited ground-level measurement data from which to estimate exposure. This is especially problematic for studying long-term PM10-2.5 health effects since PM10-2.5 is more spatially variable than PM2.5 or PM10, particularly in urban areas. Fortunately, Aerosol Optical Depth (AOD) data from satellites offer opportunities to assess PM10-2.5 more broadly. Our project leverages measurements from NASA's Terra satellite to estimate long-term PM10-2.5 concentrations in six US urban areas (Los Angeles, CA; Chicago, IL; St. Paul, MN; Baltimore, MD; New York, NY; Winston-Salem, NC) for 2000–2012. We calibrated AOD (1 km2 resolution) with EPA monitored PM10 and PM2.5 levels daily using an area-specific mixed-modeling approach with land-use regression. We then used spatial smoothing in generalized additive mixed-models to predict daily PM10 and PM2.5 when AOD was missing. PM10-2.5 was estimated after taking the difference of spatially matched PM10 and PM2.5 daily predictions. Model performance for our long-term average predictions was evaluated using leave-one-station-out cross-validation and compared to alternative, nearest-monitor and inverse distance weighting (IDW) approaches. Final long-term PM10-2.5 predictions were well correlated with measured levels estimated from collocated PM2.5 and PM10 sites in five of the six areas, with spatial CV R2 ranging from 0.50 to 0.97. Only in Winston-Salem did the model have very little predictive ability (R2: 0.34). All spatial predictions performed better than the nearest-monitor and IDW alternatives. In contrast, our final PM10-2.5 predictions had poor temporal performance, with mean monitor-level CV R2 ranging from 0.15 to 0.42. Given the superior performance of our spatial predictions compared to nearest-monitor and IDW alternatives and the high costs of field sampling, our results show the potential for combining AOD data with land-use regression to estimate long-term PM10-2.5 concentrations in localized areas.

Estimating 2013–2019 NO2 exposure with high spatiotemporal resolution in China using an ensemble model

Air pollution has become a major issue in China, especially for traffic-related pollutants such as nitrogen dioxide (NO2). Current studies in China at the national scale were less focused on NO2 exposure and consequent health effects than fine particulate exposure, mainly due to a lack of high-quality exposure models for accurate NO2 predictions over a long period. We developed an advanced modeling framework that incorporated multisource, high-quality predictor data (e.g., satellite observations [Ozone Monitoring Instrument NO2, TROPOspheric Monitoring Instrument NO2, and Multi-Angle Implementation of Atmospheric Correction aerosol optical depth], chemical transport model simulations, high-resolution geographical variables) and three independent machine learning algorithms into an ensemble model. The model contains three stages: (1) filling missing satellite data; (2) building an ensemble model and predicting daily NO2 concentrations from 2013 to 2019 across China at 1×1 km2 resolution; (3) downscaling the predictions to finer resolution (100 m) at the urban scale. Our model achieves a high performance in terms of cross-validation to assess the agreement of the overall (R2 = 0.72) and the spatial (R2 = 0.85) variations of the NO2 predictions over the observations. The model performance remains moderately good when the predictions are extrapolated to the previous years without any monitoring data (CV R2 > 0.68) or regions far away from monitors (CV R2 > 0.63). We identified a clear decreasing trend of NO2 exposure from 2013 to 2019 across the country with the largest reduction in suburban and rural areas. Our downscaled model further improved the prediction ability by 4%–14% in some megacities and captured substantial NO2 variations within 1-km grids in the urban areas, especially near major roads. Our model provides flexibility at both temporal and spatial scales and can be applied to exposure assessment and epidemiological studies with various study domains (e.g., national or citywide) and settings (e.g., long-term and short-term).

A National Land Use Regression Model for NO2 using Street View Imagery and Satellite-based Air Quality Estimates

BACKGROUND AND AIM: Characterizing air pollution with high spatiotemporal resolution is important for assessing health impacts and crafting control strategies. Land Use Regression (LUR) is often used to estimate fine-scale ambient concentrations; national-scale LURs typically employ hundreds of geographic features which offer information at neighborhood or regional scales. In this study, we develop national NO₂ models with (1) built environment features derived from street view imagery and (2) satellite estimates of air quality. Both data sources are publicly available and consistent across large geographies. METHODS: We collected NO₂ concentrations at EPA monitors and satellite-based NO₂ tropospheric column abundance during 2007-2015. In a previous study, we successfully developed single city LUR models for measures of particulates using variables extracted from Google Street View (GSV) images. Following the same method, we extracted features from GSV images (n=242836) within 500m of NO₂ monitors using a deep learning model (PSPNet). We used only GSV images collected during the same year of the NO₂ observations. For comparison, we developed two random forest models: 1) a full model with both GSV features and satellite-based estimates of NO₂ and 2) a model with only GSV features. RESULTS:Our full model (GSV + satellite) had good performance (10-fold cross validation [CV] R²: 0.69; mean absolute error [MAE]: 2.04 ppb). When only GSV features within 500m of NO₂ monitors were included in the model, performance decreased (10-fold CV R²: 0.50; MAE: 2.85 ppb), consistent with previous national LUR studies. We plan to improve these models by exploring the impact of GSV image availability and density for monitor-year pairs, expanding the buffer area for collecting images, and adding kriging into our models. CONCLUSIONS:Our findings suggest that street view imagery and satellite estimates of air quality have great potential for building large scale air quality models (national or global) under a unified framework. KEYWORDS: air pollution, modeling, oxides of nitrogen

Daily PM2.5 concentration estimates by county, ZIP code, and census tract in 11 western states 2008\u20132018

We created daily concentration estimates for fine particulate matter (PM2.5) at the centroids of each county, ZIP code, and census tract across the western US, from 2008–2018. These estimates are predictions from ensemble machine learning models trained on 24-hour PM2.5 measurements from monitoring station data across 11 states in the western US. Predictor variables were derived from satellite, land cover, chemical transport model (just for the 2008–2016 model), and meteorological data. Ten-fold spatial and random CV R2 were 0.66 and 0.73, respectively, for the 2008–2016 model and 0.58 and 0.72, respectively for the 2008–2018 model. Comparing areal predictions to nearby monitored observations demonstrated overall R2 of 0.70 for the 2008–2016 model and 0.58 for the 2008–2018 model, but we observed higher R2 (>0.80) in many urban areas. These data can be used to understand spatiotemporal patterns of, exposures to, and health impacts of PM2.5 in the western US, where PM2.5 levels have been heavily impacted by wildfire smoke over this time period.

Using machine learning to link spatiotemporal information to biological processes in the ocean: a case study for North Sea cod recruitment

Marine organisms are subject to environmental variability on various temporal and spatial scales, which affect processes related to growth and mortality of different life stages. Marine scientists are often faced with the challenge of identifying environmental variables that best explain these processes, which, given the complexity of the interactions, can be like searching for a needle in the proverbial haystack. Even after initial hypothesis-based variable selection, a large number of potential candidate variables can remain if different lagged and seasonal influences are considered. To tackle this problem, we propose a machine learning framework that incorporates important steps in model building, ranging from environmental signal extraction to automated variable selection and model validation. Its modular structure allows for the inclusion of both parametric and machine learning models, like random forest. Unsupervised feature extractions via empirical orthogonal functions (EOFs) or self-organising maps (SOMs) are demonstrated as a way to summarize spatiotemporal fields for inclusion in predictive models. The proposed framework offers a robust way to reduce model complexity through a multi-objective genetic algorithm (NSGA-II) combined with rigorous cross-validation. We applied the framework to recruitment of the North Sea cod stock and investigated the effects of sea surface temperature (SST), salinity and currents on the stock via a modified version of random forest. The best model (5-fold CV r2 = 0.69) incorporated spawning stock biomass and EOF-derived time series of SST and salinity anomalies acting through different seasons, likely relating to differing environmental effects on specific life-history stages during the recruitment year.

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.

Predicting bicycling and walking traffic using street view imagery and destination data

Few studies predict spatial patterns of bicycling and walking across multiple cities using street-level data. This study aims to model bicycle and pedestrian traffic at 4145 count locations across 20 U.S. cities using new micro-scale variables: (1) destinations from Google Point of Interest data (e.g., restaurants, schools) and (2) pixel classification from Google Street View imagery (e.g., sidewalks, trees, streetlights). We applied machine learning algorithms to assess how well street-level variables predict bicycling and walking rates. Adding street-level variables improved out-of-sample prediction accuracy of bicycling and walking activities. We also found that street-level variables (10-fold CV R2: 0.82–0.88) may be a useful alternative to Census data (0.85–0.88). Macro-scale factors (e.g., zoning) captured by Census data and micro-scale factors (e.g., streetscapes) captured in our street-level data are both useful for predicting active travel. Our models provide a new tool for estimating and understanding the spatial patterns of active travel.

Integration of Low-Cost Sensor Measurements into High-Resolution PM2.5 Exposure Modeling

Background: Low-cost air quality sensors are promising supplements to regulatory monitors for PM2.5 exposure assessment. However, little has been done to incorporate the low-cost sensor measurements in large-scale PM2.5 exposure modeling. Objectives: We conducted spatially varying calibration and developed a down-weighting strategy to optimize the use of low-cost sensor data in PM2.5 estimation. Methods: In California, PurpleAir low-cost sensors were paired with Air Quality System (AQS) regulatory stations and calibration of the sensors was performed by Geographically Weighted Regression. The calibrated PurpleAir measurements were then given lower weights according to their residual errors and fused with AQS measurements into a Random Forest model to generate 1-km daily PM2.5 estimates. Results: The calibration reduced PurpleAir’s systematic bias to ~0 μg/m³ and residual errors by 36%. Increased sensor bias was found to be associated with higher temperature and humidity as well as a longer operating time. The weighted prediction model outperformed the AQS-based prediction model with an improved random CV R2 of 0.86, an improved spatial CV R2 of 0.81, and a lower prediction error. The temporal CV R2 did not improve due to the temporal discontinuity of PurpleAir. Conclusions: The inclusion of PurpleAir data allowed the predictions to better reflect PM2.5 spatial details and hotspots.

Using Google Street View Imagery in Land Use Regression to Predict Street Level Particulate Air Pollution

Background/aim: Land-use regression (LUR) is frequently applied to estimate spatial patterns of air pollution, which is necessary for exposure assessment and epidemiological studies. Traditional LUR typically relies on data that is jurisdiction-specific or has lower spatial resolution. In this study, we investigate use of a promising new set of variables for LUR - features extracted from Google Street View (GSV) imagery - and compare GSV-only LUR to traditional LUR.Methods: We build models on previously collected mobile monitoring data in Blacksburg, VA (Particle Number [PN] and Black Carbon [BC]). We collected ~14,000 GSV images around the mobile monitoring routes. Images included 2 categories: (1) images within 50 meters of measurements to capture central features and (2) images aggregated within 8 buffers (250-2,000m) to capture background features. A deep learning model was used to classify features (e.g., vegetation, buildings, cars, water). The percentage of features within each image was used in stepwise linear regression to develop GSV-only LUR models.Results: We found that the GSV-only models had comparable performance to traditional LUR. Adjusted R2 (10-fold CV R2) was 0.76 (0.65) for PN and 0.69 (0.58) for BC, which were 5.6% (-5.8%) higher for PN and 4.5% (3.6%) higher for BC than traditional LUR models. Furthermore, adjusted R2 for PN (BC) models that used only central features were 0.45 (0.41) suggesting that street-level images at the measurement location can describe a significant amount of variability for these pollutants. Collectively, our findings suggest that GSV imagery is a promising data source for predicting street-level patterns of air pollution.Conclusions: Our results suggest that GSV imagery may be an effective data source for LUR. GSV offers potentially two major advantages: (1) finer spatial (i.e., street-level) resolution and (2) the ability to apply consistent data collection and processing protocols across large geographies and political boundaries.

Use of Low-cost Sensors to Develop Land Use Regression Models for PM2.5 of 6 Urban Areas in the US

Background/Aim: Existing national Land Use Regression (LUR) models are mainly developed based on ground-level regulatory monitors to predict and assess ambient exposure of air pollution at unmonitored locations. Emerging Low-cost sensors have improved network density and coverage that may refine the LUR model development.Methods: We retrieved and calibrated annual average PM2.5 concentrations from the low-cost sensors (i.e., PurpleAir sensors) of 6 urban areas in the US. Our independent variables included 11 categories (n = 339) of geographic features (e.g., traffic, population, land use, and satellite air pollution measurements). We developed PurpleAir LUR models (using only the PurpleAir sensors) and hybrid LUR models (using both the regulatory and low-cost monitors) for predicting annual average PM2.5 concentrations; we applied a partial least squares-universal kriging approach. We compared the exposure assessment of different LUR-derived population-weighted predictions.Results: LUR models using only the PurpleAir sensors showed reasonable performance: 10-fold CV R2 = 0.66, mean absolute error [MAE] = 2.01 µg/m3. However, the external evaluation using the EPA monitors suggested that the PurpleAir-only LUR models may consistently over-predict PM2.5 concentrations. We observed that the hybrid LUR models (R2: 0.85, MAE: 1.02 µg/m3) performed better as compared to the PurpleAir-only LUR indicating that regulatory monitors and low-cost sensors could be integrated to refine LUR models. We also noticed that the PurpleAir and hybrid LUR were spatially correlated with the regulatory-based LUR (e.g., Los Angeles: R2: 0.82, MAE: 0.77 µg/m3). The LUR-derived population-weighted predictions suggested that integrating low-cost sensors into LUR may help catch hot spots.Conclusions: LUR model development using low-cost sensor network is feasible to capture spatial variability that maybe missed by the regulatory monitors and obtain promising performance where regulatory monitors are unavailable. Future low-cost sensor-based LUR can be expanded nationally to track exposures more accurately and inform health policies.

Comparison of Different Missing-Imputation Methods for MAIAC (Multiangle Implementation of Atmospheric Correction) AOD in Estimating Daily PM2.5 Levels

The immense problem of missing satellite aerosol retrievals (Aerosol Optical Depth, (AOD)) detrimentally affects the prediction ability of ground-level PM2.5 concentrations and may lead to unavoidable biases. An appropriate missing-imputation method has not been well developed to date. This study developed a two-stage approach (AOD-imputation stage and PM2.5-prediction stage) to predict short-term PM2.5 exposure in mainland China from 2013–2018. At the AOD-imputation stage, geostatistical methods and machine learning (ML) algorithms were examined to interpolate 1 km satellite aerosol retrievals. At the PM2.5-prediction stage, the daily levels of PM2.5 were predicted at a resolution of 1 km, based on interpolated AOD and meteorological data. The statistical performances of the different interpolation methods were comprehensively compared at each stage. The original coverage of retrieved AOD was 15.46% on average. For the AOD-imputation stage, ML methods produced a higher coverage (98.64%) of AOD than geostatistical methods (21.43–87.31%). Among ML algorithms, random forest (RF) or extreme gradient boosted (XG-interpolated) AOD produced better interpolated quality (CV R2 = 0.89 and 0.85) than other algorithms (0.49–0.78), but XGBoost required only 15% of the computing time of RF. For the PM2.5 predicted stage, neither RF-AOD nor XG-AOD could guarantee higher accuracy in PM2.5 estimations (CV R2 = 0.88 (RF or XG-AOD) compared to 0.85 (original)), or more stable spatial and temporal extrapolation (spatial, (temporal) CV R2 = 0.83 (0.83), 0.82 (0.82), and 0.65 (0.61) for RF, XG, and original). For the AOD-imputation stage, the missing-filled efficiency depended more on external information, while the missing-filled accuracy relied more on model structure. For the PM2.5 predicted stage, efficient AOD interpolation (or the ability to eliminate the missing data) was a precondition for the stable spatial and temporal extrapolation, while the quality of interpolated AOD showed less significant improvements. It was found that XG-AOD is a better choice to estimate daily PM2.5 exposure in health assessments.

Estimating PM2.5 concentrations in Yangtze River Delta region of China using random forest model and the Top-of-Atmosphere reflectance

Previous studies that have used remote sensing data to estimate the PM2.5 concentrations mainly focused on the retrieval of aerosol optical depth (AOD) with moderate-to-low spatial resolution. However, the complex process of retrieving AOD from satellite Top-of-Atmosphere (TOA) reflectance always generates the missingness of AOD values due to the limitation of AOD retrieval algorithms. This study validated the possibility of using satellite TOA reflectance for estimating PM2.5 concentrations, rather than using conventional AOD products retrieved from remote sensing imageries. Given that the TOA-PM2.5 relationship cannot be accurately expressed by simple linear correlation, we developed a random forest model that integrated satellite TOA reflectance from Moderate Resolution Imaging Spectroradiometer (MODIS) Level 1B product, meteorological fields and land-use variables to estimate the ground-level PM2.5 concentrations. The highly-polluted Yangtze River Delta (YRD) region of eastern China was employed as our study case. The results showed that our model performed well with a site-based and a time-based CV R2 of 0.92 and 0.88, respectively. The derived annual and seasonal distributions of PM2.5 concentrations exhibited high PM2.5 values in northern YRD region (i.e., Jiangsu province) and relatively low values in southern region (i.e., Zhejiang province), which shared a similar distribution trend with the observed PM2.5 concentrations. This study demonstrated the outstanding performance of random forest model using satellite TOA reflectance, and also provided an effective method for remotely sensed PM2.5 estimation in regions where AOD retrievals are unavailable.