Estimating Daily PM2.5 and PM10 over Italy Using an Ensemble Model

A multi-resolution ensemble model of three decision-tree-based algorithms to predict daily NO2 concentration in France 2005-2022

Understanding and managing the health effects of Nitrogen Dioxide (NO2) requires high resolution spatiotemporal exposure maps. Here, we developed a multi-stage multi-resolution ensemble model that predicts daily NO2 concentration across continental France from 2005 to 2022. Innovations of this work include the computation of daily predictions at a 200m resolution in large urban areas and the use of a spatio-temporal blocking procedure to avoid data leakage and ensure fair performance estimation. Predictions were obtained after three cascading stages of modeling: (1) predicting NO2 total column density from Ozone Monitoring Instrument satellite; (2) predicting daily NO2 concentrations at a 1km spatial resolution using a large set of potential predictors such as predictions obtained from stage 1, land-cover and road traffic data; and (3) predicting residuals from stage 2 models at a 200m resolution in large urban areas. The latter two stages used a generalized additive model to ensemble predictions of three decision-tree algorithms (random forest, extreme gradient boosting and categorical boosting). Cross-validated performances of our ensemble models were overall very good, with a ten-fold cross-validated R2 for the 1 km model of 0.83, and of 0.69 for the 200 m model. All three basis learners participated in the ensemble predictions to various degrees depending on time and space. In sum, our multi-stage approach was able to predict daily NO2 concentrations with a relatively low error. Ensembling the predictions maximizes the chance of obtaining accurate values if one basis learner fails in a specific area or at a particular time, by relying on the other learners. To the best of our knowledge, this is the first study aiming to predict NO2 concentrations in France with such a high spatiotemporal resolution, large spatial extent, and long temporal coverage. Exposure estimates are available to investigate NO2 health effects in epidemiological studies.

Nationwide estimation of daily ambient PM2.5 from 2008 to 2020 at 1 km2 in India using an ensemble approach

High-resolution assessment of historical levels is essential for assessing the health effects of ambient air pollution in the large Indian population. The diversity of geography, weather patterns, and progressive urbanization, combined with a sparse ground monitoring network makes it challenging to accurately capture the spatiotemporal patterns of ambient fine particulate matter (PM2.5) pollution in India. We developed a model for daily average ambient PM2.5 between 2008 and 2020 based on monitoring data, meteorology, land use, satellite observations, and emissions inventories. Daily average predictions at each 1 km × 1 km grid from each learner were ensembled using a Gaussian process regression with anisotropic smoothing over spatial coordinates, and regression calibration was used to account for exposure error. Cross-validating by leaving monitors out, the ensemble model had an R2 of 0.86 at the daily level in the validation data and outperformed each component learner (by 5-18%). Annual average levels in different zones ranged between 39.7 μg/m3 (interquartile range: 29.8-46.8) in 2008 and 30.4 μg/m3 (interquartile range: 22.7-37.2) in 2020, with a cross-validated (CV)-R2 of 0.94 at the annual level. Overall mean absolute daily errors (MAE) across the 13 years were between 14.4 and 25.4 μg/m3. We obtained high spatial accuracy with spatial R2 greater than 90% and spatial MAE ranging between 7.3-16.5 μg/m3 with relatively better performance in urban areas at low and moderate elevation. We have developed an important validated resource for studying PM2.5 at a very fine spatiotemporal resolution, which allows us to study the health effects of PM2.5 across India and to identify areas with exceedingly high levels.

Dynamic Traffic Data in Machine-Learning Air Quality Mapping Improves Environmental Justice Assessment

Air pollution poses a critical public health threat around many megacities but in an uneven manner. Conventional models are limited to depict the highly spatial- and time-varying patterns of ambient pollutant exposures at the community scale for megacities. Here, we developed a machine-learning approach that leverages the dynamic traffic profiles to continuously estimate community-level year-long air pollutant concentrations in Los Angeles, U.S. We found the introduction of real-world dynamic traffic data significantly improved the spatial fidelity of nitrogen dioxide (NO2), maximum daily 8-h average ozone (MDA8 O3), and fine particulate matter (PM2.5) simulations by 47%, 4%, and 15%, respectively. We successfully captured PM2.5 levels exceeding limits due to heavy traffic activities and providing an "out-of-limit map" tool to identify exposure disparities within highly polluted communities. In contrast, the model without real-world dynamic traffic data lacks the ability to capture the traffic-induced exposure disparities and significantly underestimate residents' exposure to PM2.5. The underestimations are more severe for disadvantaged communities such as black and low-income groups, showing the significance of incorporating real-time traffic data in exposure disparity assessment.

Estimating daily air temperature and pollution in Catalonia: A comprehensive spatiotemporal modelling of multiple exposures

Environmental epidemiology studies require models of multiple exposures to adjust for co-exposure and explore interactions. We estimated spatiotemporal exposure to surface air temperature and pollution (PM2.5, PM10, NO2, O3) at high spatiotemporal resolution (daily, 250 m) for 2018-2020 in Catalonia. Innovations include the use of TROPOMI products, a data split for remote sensing gap-filling evaluation, estimation of prediction uncertainty, and use of explainable machine learning. We compiled meteorological and air quality station measurements, climate and atmospheric composition reanalyses, remote sensing products, and other spatiotemporal data. We performed gap-filling of remotely-sensed products using Random Forest (RF) models and validated them using Out-Of-Bag (OOB) samples and a structured data split. The exposure modelling workflow consisted of: 1) PM2.5 station imputation with PM10 data; 2) quantile RF (QRF) model fitting; and 3) geostatistical residual spatial interpolation. Prediction uncertainty was estimated using QRF. SHAP values were used to examine variable importance and the fitted relationships. Model performance was assessed via nested CV at the station level. Evaluation of the gap-filling models using the structured split showed error underestimation when using OOB. Temperature models had the best performance (R2 =0.98) followed by the gaseous air pollutants (R2 =0.81 for NO2 and 0.86 for O3), while the performance of the PM2.5 and PM10 models was lower (R2 =0.57 and 0.63 respectively). Predicted exposure patterns captured urban heat island effects, dust advection events, and NO2 hotspots. SHAP values estimated a high importance of TROPOMI tropospheric NO2 columns in PM and NO2 models, and confirmed that the fitted associations conformed to prior knowledge. Our work highlights the importance of correctly validating gap-filling models and the potential of TROPOMI measurements. Moderate performance in PM models can be partly explained by the poor station coverage. Our exposure estimates can be used in epidemiological studies potentially accounting for exposure uncertainty.

Application of satellite remote sensing data and random forest approach to estimate ground-level PM2.5 concentration in Northern region of Thailand

Numerous epidemiological studies have shown that particulate matter with aerodynamic diameter up to 2.5 μm (PM2.5) is associated with many health consequences, where PM2.5 concentration obtained from the monitoring station was normally applied as the exposure level, so that the concentration of PM2.5 in unmonitored areas has not been captured. The satellite-derived aerosol optical depth (AOD) product is then used to spatially predict ground truth of PM2.5 concentration that covers the locations with no air quality monitoring station, but this method has seldom been developed in Thailand. This study aimed at estimating ground-level PM2.5 concentration at 3 km × 3 km spatial resolution over Northern region of Thailand in 2021 using the random forest model integrating the Moderate Resolution Imaging Spectroradiometer (MODIS) AOD products from Terra and Aqua satellites, meteorological factors, and land use data. A random forest model contained 100 decision trees was utilized to train the model, and 10-fold cross-validation approach was implemented to validate the model performance. The good consistency between actual (observed) and predicted concentrations of PM2.5 in Northern region of Thailand was observed, where a coefficient of determination (R2) and root mean square error (RMSE) of the model fitting were 0.803 and 14.30 μg/m3, respectively, and those of 10-fold cross-validation approach were 0.796 and 14.64 μg/m3, respectively. The three most important predictors for estimating the ground-level concentrations of PM2.5 in this study were normalized difference vegetation index (NDVI), relative humidity, and number of fire hotspot, respectively. Findings from this study revealed that integrating the MODIS AOD, meteorological variables, and land use data into the random forest model precisely and accurately estimated ground-level PM2.5 concentration over Northern region of Thailand that can be further used to investigate the effects of PM2.5 exposure on health consequences, even in unmonitored locations, in epidemiological studies.

Unveiling the migration of Cr and Cd to biochar from pyrolysis of manure and sludge using machine learning

Heavy metal (HM) in biochar derived from pyrolysis of sludge or manure is the main issue for its large-scale application in soils for carbon sequestration. However, there is a paucity of efficient approaches to predict and comprehend the HM migration during pyrolysis for preparing low HM-contained biochar. Herein, the data on the feedstock information (FI), additive, total concentration of feedstock (FTC) of HM Cr and Cd, and pyrolysis condition, were extracted from the literature, to predict total concentration (TC) and retention rate (RR) of Cr and Cd in sludge/manure biochar using ML for mapping their migration during pyrolysis. Two datasets for Cr and Cd were compiled with 388 and 292 data points from 48 and 37 peer-review papers. The results indicated that the TC and RR of Cr and Cd could be predicted by the Random Forest model with test R² of 0.74-0.98. Their TC and RR in biochar were dominated by the FTC and FI, respectively; while pyrolysis temperature was the most important to Cd RR. Moreover, potassium-based inorganic additives decreased the TC and RR of Cr while increased those of Cd. The predictive models and insights provided by this work could aid the understanding of HM migration during manure and sludge pyrolysis and guide the preparation of low HM-contained biochar.

Relationship of long-term air pollution exposure with asthma and rhinitis in Italy: an innovative multipollutant approach

BACKGROUND

air pollution is a complex mixture; novel multipollutant approaches could help understanding the health effects of multiple concomitant exposures to air pollutants.

AIM

to assess the relationship of long-term air pollution exposure with the prevalence of respiratory/allergic symptoms and diseases in an Italian multicenter study using single and multipollutant approaches.

METHODS

14420 adults living in 6 Italian cities (Ancona, Pavia, Pisa, Sassari, Turin, Verona) were investigated in 2005-2011 within 11 different study cohorts. Questionnaire information about risk factors and health outcomes was collected. Machine learning derived mean annual concentrations of PM₁₀, PM_2.5, NO₂ and mean summer concentrations of O₃ (μg/m³) at residential level (1-km resolution) were used for the period 2013-2015. The associations between the four pollutants and respiratory/allergic symptoms/diseases were assessed using two approaches: a) logistic regression models (single-pollutant models), b) principal component logistic regression models (multipollutant models). All the models were adjusted for age, sex, education level, smoking habits, season of interview, climatic index and included a random intercept for cohorts.

RESULTS

the three-year average (± standard deviation) pollutants concentrations at residential level were: 20.3 ± 6.8 μg/m³ for PM_2.5, 29.2 ± 7.0 μg/m³ for PM₁₀, 28.0 ± 11.2 μg/m³ for NO₂, and 70.9 ± 4.3 μg/m³ for summer O_3. Through the multipollutant models the following associations emerged: PM₁₀ and PM_2.5 were related to 14-25% increased odds of rhinitis, 23-34% of asthma and 30-33% of night awakening; NO₂ was related to 6-9% increased odds of rhinitis, 7-8% of asthma and 12% of night awakening; O₃ was associated with 37% increased odds of asthma attacks. Overall, the Odds Ratios estimated through the multipollutant models were attenuated when compared to those of the single-pollutant models.

CONCLUSIONS

this study enabled to obtain new information about the health effects of air pollution on respiratory/allergic outcomes in adults, applying innovative methods for exposure assessment and multipollutant analyses.

An ensemble mixed spatial model in estimating long-term and diurnal variations of PM2.5 in Taiwan

Meteorology, human activities, and other emission sources drive diurnal cyclic patterns of air pollution. Previous studies mainly focused on the variation of PM_2.5 concentrations during daytime rather than nighttime. In addition, assessing the spatial variations of PM_2.5 in large areas is a critical issue for environmental epidemiological studies to clarify the health effects from PM_2.5 exposures. In terms of air pollution spatial modelling, using only a single model might lose information in capturing spatial and temporal correlation between predictors and pollutant levels. Hence, this study aimed to propose an ensemble mixed spatial model that incorporated Kriging interpolation, land-use regression (LUR), machine learning, and stacking ensemble approach to estimate long-term PM_2.5 variations for nearly three decades in daytime and nighttime. Three steps of model development were applied: 1) linear based LUR and Hybrid Kriging-LUR were used to determine influential predictors; 2) machine learning algorithms were used to enhance model prediction accuracy; 3) predictions from the selected machine learning models were fitted and evaluated again to build the final ensemble mixed spatial model. The results showed that prediction performance increased from 0.514 to 0.895 for daily, 0.478 to 0.879 for daytime, and 0.523 to 0.878 for nighttime when applying the proposed ensemble mixed spatial model compared with LUR. Results of overfitting test and extrapolation ability test confirmed the robustness and reliability of the developed models. The distance to the nearest thermal power plant, density of soil and pebbles fields, and funeral facilities might affect the variation of PM_2.5 levels between daytime and nighttime. The PM_2.5 level was higher in daytime compared with nighttime with little difference, revealing the importance of estimating nighttime PM_2.5 variations. Our findings also clarified the emission sources in daytime and nighttime, which serve as valuable information for air pollution control strategies establishment.

Influence of land-sea breeze on PM[Formula: see text] prediction in central and southern Taiwan using composite neural network

PM[Formula: see text] prediction plays an important role for governments in establishing policies to control the emission of excessive atmospheric pollutants to protect the health of citizens. However, traditional machine learning methods that use data collected from ground-level monitoring stations have reached their limit with poor model generalization and insufficient data. We propose a composite neural network trained with aerosol optical depth (AOD) and weather data collected from satellites, as well as interpolated ocean wind features. We investigate the model outputs of different components of the composite neural network, concluding that the proposed composite neural network architecture yields significant improvements in overall performance compared to each component and the ensemble model benchmarks. The monthly analysis also demonstrates the superiority of the proposed architecture for stations where land-sea breezes frequently occur in the southern and central Taiwan in the months when land-sea breeze dominates the accumulation of PM[Formula: see text].

Global estimates of daily ambient fine particulate matter concentrations and unequal spatiotemporal distribution of population exposure: a machine learning modelling study

BACKGROUND

Short-term exposure to ambient PM_2·5 is a leading contributor to the global burden of diseases and mortality. However, few studies have provided the global spatiotemporal variations of daily PM_2·5 concentrations over recent decades.

METHODS

In this modelling study, we implemented deep ensemble machine learning (DEML) to estimate global daily ambient PM_2·5 concentrations at 0·1° × 0·1° spatial resolution between Jan 1, 2000, and Dec 31, 2019. In the DEML framework, ground-based PM_2·5 measurements from 5446 monitoring stations in 65 countries worldwide were combined with GEOS-Chem chemical transport model simulations of PM_2·5 concentration, meteorological data, and geographical features. At the global and regional levels, we investigated annual population-weighted PM_2·5 concentrations and annual population-weighted exposed days to PM_2·5 concentrations higher than 15 μg/m³ (2021 WHO daily limit) to assess spatiotemporal exposure in 2000, 2010, and 2019. Land area and population exposures to PM_2·5 above 5 μg/m³ (2021 WHO annual limit) were also assessed for the year 2019. PM_2·5 concentrations for each calendar month were averaged across the 20-year period to investigate global seasonal patterns.

FINDINGS

Our DEML model showed good performance in capturing the global variability in ground-measured daily PM_2·5, with a cross-validation R² of 0·91 and root mean square error of 7·86 μg/m³. Globally, across 175 countries, the mean annual population-weighted PM_2·5 concentration for the period 2000-19 was estimated at 32·8 μg/m³ (SD 0·6). During the two decades, population-weighted PM_2·5 concentration and annual population-weighted exposed days (PM_2·5 >15 μg/m³) decreased in Europe and northern America, whereas exposures increased in southern Asia, Australia and New Zealand, and Latin America and the Caribbean. In 2019, only 0·18% of the global land area and 0·001% of the global population had an annual exposure to PM_2·5 at concentrations lower than 5 μg/m³, with more than 70% of days having daily PM_2·5 concentrations higher than 15 μg/m³. Distinct seasonal patterns were indicated in many regions of the world.

INTERPRETATION

The high-resolution estimates of daily PM_2·5 provide the first global view of the unequal spatiotemporal distribution of PM_2·5 exposure for a recent 20-year period, which is of value for assessing short-term and long-term health effects of PM_2·5, especially for areas where monitoring station data are not available.

FUNDING

Australian Research Council, Australian Medical Research Future Fund, and the Australian National Health and Medical Research Council.

The impact of fire-specific PM2.5 calibration on health effect analyses

The quantification of PM_2.5 concentrations solely stemming from both wildfire and prescribed burns (hereafter referred to as 'fire') is viable using the Community Multiscale Air Quality (CMAQ), although CMAQ outputs are subject to biases and uncertainties. To reduce the biases in CMAQ-based outputs, we propose a two-stage calibration strategy that improves the accuracy of CMAQ-based fire PM_2.5 estimates. First, we calibrated CMAQ-based non-fire PM_2.5 to ground PM_2.5 observations retrieved during non-fire days using an ensemble-based model. We estimated fire PM_2.5 concentrations in the second stage by multiplying the calibrated non-fire PM_2.5 obtained from the first stage by location- and time-specific conversion ratios. In a case study, we estimated fire PM_2.5 during the Washington 2016 fire season using the proposed calibration approach. The calibrated PM_2.5 better agreed with ground PM_2.5 observations with a 10-fold cross-validated (CV) R² of 0.79 compared to CMAQ-based PM_2.5 estimates with R² of 0.12. In the health effect analysis, we found significant associations between calibrated fire PM_2.5 and cardio-respiratory hospitalizations across the fire season: relative risk (RR) for cardiovascular disease = 1.074, 95% confidence interval (CI) = 1.021-1.130 in October; RR = 1.191, 95% CI = 1.099-1.291 in November; RR for respiratory disease = 1.078, 95% CI = 1.005-1.157 in October; RR = 1.153, 95% CI = 1.045-1.272 in November. However, the results were inconsistent when non-calibrated PM_2.5 was used in the analysis. We found that calibration affected health effect assessments in the present study, but further research is needed to confirm our findings.

Using satellite data on remote transportation of air pollutants for PM2.5 prediction in northern Taiwan

Accurate PM2.5 prediction is part of the fight against air pollution that helps governments to manage environmental policy. Satellite Remote sensing aerosol optical depth (AOD) processed by The Multi-Angle Implementation of Atmospheric Correlation (MAIAC) algorithm allows us to observe the transportation of remote pollutants between regions. The paper proposes a composite neural network model, the Remote Transported Pollutants (RTP) model, for such long-range pollutant transportation that predicts more accurate local PM2.5 concentrations given such satellite data. The proposed RTP model integrates several deep learning components and learns from the heterogeneous features of various domains. We also detected remote transportation pollution events (RTPEs) at two reference sites from the AOD data. Extensive experiments using real-world data show that the proposed RTP model outperforms the base model that does not account for RTPEs by 17%-30%, 23%-26% and 18%-22% and state-of-the-art models that account for RTPEs by 12%-22%, 12%-14%, and 10%-11% at +4h to +24h, +28h to +48 hours, and +52h to +72h hours respectively.

Optimizing modeling windows to better capture the long-term variation of PM2.5 concentrations in China during 2005-2019

Including data of different time intervals during model development influences the predicting accuracy of PM2.5 but has not been widely discussed. Therefore, we included modeling data with multiple time windows to identify optimized modeling time windows for capturing the long-term variation of PM2.5 in China during 2005-2019. In general, we incorporated PM2.5 measurements, aerosol optical depth (AOD), meteorological parameters, land use data, and other predictors to train random forest models. The study period was separated into two phases (2013-2019 and 2005-2012) according to the availability of PM2.5 measurements. First, we trained models with two strategies of choosing time windows to compare model performance in predicting PM2.5 from 2013 to 2019, when measurements were available. Strategy 1a (ST1a) refers to training one model with all available data, and Strategy 1b (ST1b) refers to training multiple models each with one-year data. Second, we trained models with additional ten strategies (ST2a-ST2j) based on data from different time windows during 2013-2019 to compare the accuracy in predicting PM2.5 before 2013, when measurements were unavailable. The internal and external cross-validation (CV) indicated that the model performance of ST1b was better than ST1a. Predictions based on ST1a tended to underestimate PM2.5 levels in 2013 and 2014 when PM2.5 concentrations were high, and overestimate after 2017 when PM2.5 dropped dramatically. The external CV of predicting historical PM2.5 was the most robust in ST2i (averaged predictions from two models developed by 2013 and 2014 data, respectively). Models with data closer to historical years and PM2.5 levels performed better in predicting historical PM2.5 concentrations. Our results suggested that training models with data of current-years performed better during 2013-2019, and with data of 2013 and 2014 performed better in predicting historical PM2.5 before 2013 in China. The comparison provided evidence for choosing optimized time windows when predicting long-term PM2.5 concentrations in China.

A gap-filling hybrid approach for hourly PM2.5 prediction at high spatial resolution from multi-sourced AOD data

Despite a growing interest in the satellite derived estimation of ground-level PM_2.5 concentrations, modeling hourly PM_2.5 levels at high spatial resolution with complete coverage for a large study domain remains a challenge. The primary modeling challenges lie in the presence of missing data in aerosol optical depth (AOD) and the limited data resolution for a single-platformed satellite AOD product. To address these issues, we developed a gap-filling hybrid approach to estimate full coverage hourly ground-level PM_2.5 concentrations at a high spatial resolution of 1 km using multi-platformed and multi-scale satellite derived AOD products. Specifically, we filled the gaps and downscaled the multi-sourced AOD from Geostationary Ocean Color Imager (GOCI), Multi-Angle Implementation of Atmospheric Correction (MAIAC), and Modern-Era Retrospective Analysis for Research and Applications - version 2 (MERRA-2), using a hybrid data fusion approach. The fused hourly AOD with full coverage was then used for hourly PM_2.5 predictions at a high spatial resolution of 1 km. We demonstrated the application of the proposed approach and assessed its performance using the data collected from northeastern Asia from 2015 to 2019. Our fused hourly AOD data showed high accuracy with the mean absolute error of 0.14 and correlation coefficient of 0.94, in validation against Aerosol Robotic Network (AERONET) AOD. Our AOD-based PM_2.5 prediction model showed a good prediction accuracy with cross-validated R² of 0.85 and root mean squared error of 12.40 μg/m³, respectively. Given that the highly resolved PM_2.5 predictions captured both the temporal trend and the peak of PM_2.5 pollution scenarios, we concluded that the proposed hybrid approach can effectively combine multi-sourced satellite AOD and derive subsequent PM_2.5 distributions at high spatial and temporal resolutions.

GIS-based air quality modelling: spatial prediction of PM10 for Selangor State, Malaysia using machine learning algorithms

Rapid urbanization has caused severe deterioration of air quality globally, leading to increased hospitalization and premature deaths. Therefore, accurate prediction of air quality is crucial for mitigation planning to support urban sustainability and resilience. Although some studies have predicted air pollutants such as particulate matter (PM) using machine learning algorithms (MLAs), there is a paucity of studies on spatial hazard assessment with respect to the air quality index (AQI). Incorporating PM in AQI studies is crucial because of its easily inhalable micro-size which has adverse impacts on ecology, environment, and human health. Accurate and timely prediction of the air quality index can ensure adequate intervention to aid air quality management. Therefore, this study undertakes a spatial hazard assessment of the air quality index using particulate matter with a diameter of 10 μm or lesser (PM10) in Selangor, Malaysia, by developing four machine learning models: eXtreme Gradient Boosting (XGBoost), random forest (RF), K-nearest neighbour (KNN), and Naive Bayes (NB). Spatially processed data such as NDVI, SAVI, BU, LST, Ws, slope, elevation, and road density was used for the modelling. The model was trained with 70% of the dataset, while 30% was used for cross-validation. Results showed that XGBoost has the highest overall accuracy and precision of 0.989 and 0.995, followed by random forest (0.989, 0.993), K-nearest neighbour (0.987, 0.984), and Naive Bayes (0.917, 0.922), respectively. The spatial air quality maps were generated by integrating the geographical information system (GIS) with the four MLAs, which correlated with Malaysia's air pollution index. The maps indicate that air quality in Selangor is satisfactory and posed no threats to health. Nevertheless, the two algorithms with the best performance (XGBoost and RF) indicate that a high percentage of the air quality is moderate. The study concludes that successful air pollution management policies such as green infrastructure practice, improvement of energy efficiency, and restrictions on heavy-duty vehicles can be adopted in Selangor and other Southeast Asian cities to prevent deterioration of air quality in the future.

Ensemble averaging using remote sensing data to model spatiotemporal PM10 concentrations in sparsely monitored South Africa

There is a paucity of air quality data in sub-Saharan African countries to inform science driven air quality management and epidemiological studies. We investigated the use of available remote-sensing aerosol optical depth (AOD) data to develop spatially and temporally resolved models to predict daily particulate matter (PM₁₀) concentrations across four provinces of South Africa (Gauteng, Mpumalanga, KwaZulu-Natal and Western Cape) for the year 2016 in a two-staged approach. In stage 1, a Random Forest (RF) model was used to impute Multiangle Implementation of Atmospheric Correction AOD data for days where it was missing. In stage 2, the machine learner algorithms RF, Gradient Boosting and Support Vector Regression were used to model the relationship between ground-monitored PM₁₀ data, AOD and other spatial and temporal predictors. These were subsequently combined in an ensemble model to predict daily PM₁₀ concentrations at 1 km × 1 km spatial resolution across the four provinces. An out-of-bag R² of 0.96 was achieved for the first stage model. The stage 2 cross-validated (CV) ensemble model captured 0.84 variability in ground-monitored PM₁₀ with a spatial CV R² of 0.48 and temporal CV R² of 0.80. The stage 2 model indicated an optimal performance of the daily predictions when aggregated to monthly and annual means. Our results suggest that a combination of remote sensing data, chemical transport model estimates and other spatiotemporal predictors has the potential to improve air quality exposure data in South Africa's major industrial provinces. In particular, the use of a combined ensemble approach was found to be useful for this area with limited availability of air pollution ground monitoring data.

High-Resolution Urban Air Quality Mapping for Multiple Pollutants Based on Dense Monitoring Data and Machine Learning

Spatially explicit urban air quality information is important for urban fine-management and public life. However, existing air quality measurement methods still have some limitations on spatial coverage and system stability. A micro station is an emerging monitoring system with multiple sensors, which can be deployed to provide dense air quality monitoring data. Here, we proposed a method for urban air quality mapping at high-resolution for multiple pollutants. By using the dense air quality monitoring data from 448 micro stations in Lanzhou city, we developed a decision tree model to infer the distribution of citywide air quality at a 500 m × 500 m × 1 h resolution, with a coefficient of determination (R²) value of 0.740 for PM_2.5, 0.754 for CO and 0.716 for SO₂. Meanwhile, we also show that the deployment density of the monitoring stations can have a significant impact on the air quality inference results. Our method is able to show both short-term and long-term distribution of multiple important pollutants in the city, which demonstrates the potential and feasibility of dense monitoring data combined with advanced data science methods to support urban atmospheric environment fine-management, policy making, and public health studies.

Predicting Daily PM2.5 Exposure with Spatially Invariant Accuracy Using Co-Existing Pollutant Concentrations as Predictors

The spatiotemporal variation of PM2.5 should be accurately estimated for epidemiological studies. However, the accuracy of prediction models may change over geographical space, which is not conducive for proper exposure assessment. In this study, we developed a prediction model to estimate daily PM2.5 concentrations from 2010 to 2017 in the Kansai region of Japan with co-existing pollutant concentrations as predictors. The overall objective was to obtain daily estimates over the study domain with spatially homogeneous accuracy. We used random forest algorithm to model the relationship between the daily PM2.5 concentrations and various predictors. The model performance was evaluated via spatial and temporal cross-validation and the daily PM2.5 surface was estimated from 2010 to 2017 at a 1 km × 1 km resolution. We achieved R2 values of 0.91 and 0.92 for spatial and temporal cross-validation, respectively. The prediction accuracy for each monitoring site was found to be consistently high, regardless of the distance to the nearest monitoring location, up to 10 km. Even for distances greater than 10 km, the mean R2 value was 0.88. Our approach yielded spatially homogeneous prediction accuracy, which is beneficial for epidemiological studies. The daily PM2.5 estimates will be used in a related birth cohort study to evaluate the potential impact on human health.

Estimation of the Near-Surface Ozone Concentration with Full Spatiotemporal Coverage across the Beijing-Tianjin-Hebei Region Based on Extreme Gradient Boosting Combined with a WRF-Chem Model

With the intensification of global warming and economic development in China, the near-surface ozone (O3) concentration has been increasing recently, especially in the Beijing-Tianjin-Hebei (BTH) region, which is the political and economic center of China. However, O3 has been measured in real time only over the past few years, and the observational records are discontinuous. Therefore, we propose a new method (WRFC-XGB) to establish a near-surface O3 concentration dataset in the BTH region by integrating the Weather Research and Forecasting with Chemistry (WRF-Chem) model with the extreme gradient boosting (XGBoost) algorithm. Based on this method, the 8-h maximum daily average (MDA8) O3 concentrations are obtained with full spatiotemporal coverage at a spatial resolution of 0.1° × 0.1° across the BTH region in 2018. Two evaluation methods, sample- and station-based 10-fold cross-validation (10-CV), are used to assess our method. The sample-based (station-based) 10-CV evaluation results indicate that WRFC-XGB can achieve excellent accuracy with a high coefficient of determination (R2) of 0.95 (0.91), low root mean square error (RMSE) of 13.50 (17.70) µg m−3, and mean absolute error (MAE) of 9.60 (12.89) µg m−3. In addition, superb spatiotemporal consistencies are confirmed for this model, including the estimation of high O3 concentrations, and our WRFC-XGB model outperforms traditional models and previous studies in data mining. In addition, the proposed model can be applied to estimate the O3 concentration when it has not been measured. Furthermore, the spatial distribution analysis of the MDA8 O3 in 2018 reveals that O3 pollution in the BTH region exhibits significant seasonality. Heavy O3 pollution episodes mainly occur in summer, and the high O3 loading is distributed mainly in the southern BTH areas, which will pose challenges to atmospheric environmental governance for local governments.

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.

Rapid assessments of light-duty gasoline vehicle emissions using on-road remote sensing and machine learning

In-time and accurate assessments of on-road vehicle emissions play a central role in urban air quality and health policymaking. However, official insight is hampered by the Inspection/Maintenance (I/M) procedure conducted in the laboratory annually. It not only has a large gap to real-world situations (e.g., meteorological conditions) but also is incapable of regular supervision. Here we build a unique dataset including 103,831 light-duty gasoline vehicles, in which on-road remote sensing (ORRS) measurements are linked to the I/M records based on the vehicle identification numbers and license plates. On this basis, we develop an ensemble model framework that integrates three machining learning algorithms, including neural network (NN), extreme gradient boosting (XGBoost), and random forest (RF). We demonstrate that this ensemble model could rapidly assess the vehicle-specific emissions (i.e., CO, HC, and NO). In particular, the model performs quite well for the passing vehicles under normal conditions (i.e., lower VSP (<18 kw/t), temperature (6-32 °C), relative humidity (<80%), and wind speed (<5 m/s)). Together with the current emission standard, we identify a large number of the 'dirty' (2.33%) or 'clean' (74.92%) vehicles in the real world. Our results show that the ORRS measurements, assisted by the machine-learning-based ensemble model developed here, can realize day-to-day supervision of on-road vehicle-specific emissions. This approach framework provides a valuable opportunity to reform the I/M procedures globally and mitigate urban air pollution deeply.

Deep Ensemble Machine Learning Framework for the Estimation of PM2.5 Concentrations

BACKGROUND

Accurate estimation of historical PM2.5 (particle matter with an aerodynamic diameter of less than 2.5μm) is critical and essential for environmental health risk assessment.

OBJECTIVES

The aim of this study was to develop a multiple-level stacked ensemble machine learning framework for improving the estimation of the daily ground-level PM2.5 concentrations.

METHODS

An innovative deep ensemble machine learning framework (DEML) was developed to estimate the daily PM2.5 concentrations. The framework has a three-stage structure: At the first stage, four base models [gradient boosting machine (GBM), support vector machine (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost)] were used to generate a new data set of PM2.5 concentrations for training the next-stage learners. At the second stage, three meta-models [RF, XGBoost, and Generalized Linear Model (GLM)] were used to estimate PM2.5 concentrations using a combination of the original data set and the predictions from the first-stage models. At the third stage, a nonnegative least squares (NNLS) algorithm was employed to obtain the optimal weights for PM2.5 estimation. We took the data from 133 monitoring stations in Italy as an example to implement the DEML to predict daily PM2.5 at each 1km×1km grid cell from 2015 to 2019 across Italy. We evaluated the model performance by performing 10-fold cross-validation (CV) and compared it with five benchmark algorithms [GBM, SVM, RF, XGBoost, and Super Learner (SL)].

RESULTS

The results revealed that the PM2.5 prediction performance of DEML [coefficients of determination (R2)=0.87 and root mean square error (RMSE)=5.38μg/m3] was superior to any benchmark models (with R2 of 0.51, 0.76, 0.83, 0.70, and 0.83 for GBM, SVM, RF, XGBoost, and SL approach, respectively). DEML displayed reliable performance in capturing the spatiotemporal variations of PM2.5 in Italy.

DISCUSSION

The proposed DEML framework achieved an outstanding performance in PM2.5 estimation, which could be used as a tool for more accurate environmental exposure assessment. https://doi.org/10.1289/EHP9752.

Substantial Reduction in Particulate Matter Air Pollution across Europe during 2006-2019: A Spatiotemporal Modeling Analysis

Air pollution poses the largest environmental health risk in Europe. Particulate matter (PM) concentrations are the most harmful pollutants representing the main air quality indicator in the Sustainable Development Goals (SDGs). The air quality surveillance in Europe is based on a monitoring network that is too coarse for a comprehensive evaluation of the air pollution burden. We link raw pollutant data with remotely sensed products using Bayesian geostatistical models and for the first time estimate pan-European near-surface concentrations of both fine (PM2.5) and coarse (PM10) particles at 1 km2 spatial resolution during 2006-2019. We evaluate the compliance with the air quality thresholds set by the World Health Organization (WHO) and the European Union (EU) and assess country-wise trends. The results show that during the last 14 years, PM10 and PM2.5 concentrations declined by 36.5% (95% credible interval: 30.3, 41.9%) and 39.1% (26.6, 50.5%), respectively. The number of people exposed to PM10 levels above the WHO thresholds decreased from 78.3% (52.6, 91.8%) in 2006 to 28.4% (16.2, 43.7%) in 2019; for PM2.5, the decrease was smaller: from 91.0% (61.3, 99.1%) exposed in 2006 to 53.6% (33.5, 76.3%) in 2019. Although there is a clear improvement in the overall picture, stricter measures are needed to ensure compliance with the WHO guidelines.

Full-coverage spatiotemporal mapping of ambient PM2.5 and PM10 over China from Sentinel-5P and assimilated datasets: Considering the precursors and chemical compositions

Ambient concentrations of particulate matters (PM_2.5 and PM₁₀) are significant indicators for monitoring the air quality relevant to living conditions. At present, most remote sensing based approaches for the estimation of PM_2.5 and PM₁₀ employed Aerosol Optical Depth (AOD) products as the main variate. Nevertheless, the coverage of missing data is generally large in AOD products, which can cause deviations in practical applications of estimated PM_2.5 and PM₁₀ (e.g., air quality monitoring and exposure evaluation). To efficiently address this issue, our study explores a novel approach using the datasets of the precursors & chemical compositions for PM_2.5 and PM₁₀ instead of AOD products. Specifically, the daily full-coverage ambient concentrations of PM_2.5 and PM₁₀ are estimated at 5-km (0.05°) spatial girds across China based on Sentinel-5P and assimilated datasets (GEOS-FP). The estimation models are acquired via an advanced ensemble learning method named Light Gradient Boosting Machine in this paper. For comparison, the Deep Blue AOD product from VIIRS is adopted in a similar framework as a baseline (AOD-based). Validation results show that the ambient concentrations are well estimated through the proposed approach, with the space-based Cross-Validation R²s and RMSEs of 0.88 (0.83) and 11.549 (22.9) μg/m³ for PM_2.5 (PM₁₀), respectively. Meanwhile, the proposed approach achieves better performance than the AOD-based in different cases (e.g., overall and seasonal). Compared to the related previous works over China, the estimation accuracy of our method is also satisfactory. Regarding the mapping, the estimated results through the proposed approach display consecutive spatial distribution and can exactly express the seasonal variations of PM_2.5 and PM₁₀. The proposed approach could efficiently present daily full-coverage results at 5-km spatial grids. It has a large potential to be extended for providing global accurate ambient concentrations of PM_2.5 and PM₁₀ at multiple temporal scales (e.g., daily and annual).

A National-Scale 1-km Resolution PM2.5 Estimation Model over Japan Using MAIAC AOD and a Two-Stage Random Forest Model

Satellite-based models for estimating concentrations of particulate matter with an aerodynamic diameter less than 2.5 μm (PM2.5) have seldom been developed in islands with complex topography over the monsoon area, where the transport of PM2.5 is influenced by both the synoptic-scale winds and local-scale circulations compared with the continental regions. We validated Multi-Angle Implementation of Atmospheric Correction (MAIAC) aerosol optical depth (AOD) with ground observations in Japan and developed a 1-km-resolution national-scale model between 2011 and 2016 to estimate daily PM2.5 concentrations. A two-stage random forest model integrating MAIAC AOD with meteorological variables and land use data was applied to develop the model. The first-stage random forest model was used to impute the missing AOD values. The second-stage random forest model was then utilised to estimate ground PM2.5 concentrations. Ten-fold cross-validation was performed to evaluate the model performance. There was good consistency between MAIAC AOD and ground truth in Japan (correlation coefficient = 0.82 and 74.62% of data falling within the expected error). For model training, the model showed a training coefficient of determination (R2) of 0.98 and a root mean square error (RMSE) of 1.22 μg/m3. For the 10-fold cross-validation, the cross-validation R2 and RMSE of the model were 0.86 and 3.02 μg/m3, respectively. A subsite validation was used to validate the model at the grids overlapping with the AERONET sites, and the model performance was excellent at these sites with a validation R2 (RMSE) of 0.94 (1.78 μg/m3). Additionally, the model performance increased as increased AOD coverage. The top-ten important predictors for estimating ground PM2.5 concentrations were day of the year, temperature, AOD, relative humidity, 10-m-height zonal wind, 10-m-height meridional wind, boundary layer height, precipitation, surface pressure, and population density. MAIAC AOD showed high retrieval accuracy in Japan. The performance of the satellite-based model was excellent, which showed that PM2.5 estimates derived from the model were reliable and accurate. These estimates can be used to assess both the short-term and long-term effects of PM2.5 on health outcomes in epidemiological studies.

A novel multi-model data-driven ensemble approach for the prediction of particulate matter concentration

Accuracy in the prediction of the particulate matter (PM_2.5 and PM₁₀) concentration in the atmosphere is essential for both its monitoring and control. In this study, a novel neuro fuzzy ensemble (NF-E) model was proposed for prediction of hourly PM_2.5 and PM₁₀ concentration. The NF-E involves careful selection of relevant input parameters for base modelling and using an adaptive neuro fuzzy inference system (ANFIS) model as a nonlinear kernel for obtaining ensemble output. The four base models used include ANFIS, artificial neural network (ANN), support vector regression (SVR) and multilinear regression (MLR). The dominant input parameters for developing the base models were selected using two nonlinear approaches (mutual information and single-input single-output ANN-based sensitivity analysis) and a conventional Pearson correlation coefficient. The NF-E model was found to predict both PM_2.5 and PM₁₀ with higher generalization ability and least error. The NF-E model outperformed all the single base models and other linear ensemble techniques with a Nash-Sutcliffe efficiency (NSE) of 0.9594 and 0.9865, mean absolute error (MAE) of 1.63 μg/m³ and 1.66 μg/m³ and BIAS of 0.0760 and 0.0340 in the testing stage for PM_2.5 and PM₁₀, respectively. The NF-E could improve the efficiency of other models by 4-22% for PM_2.5 and 3-20% for PM₁₀ depending on the model.

Using a land use regression model with machine learning to estimate ground level PM2.5

Ambient fine particulate matter (PM_2.5) has been ranked as the sixth leading risk factor globally for death and disability. Modelling methods based on having access to a limited number of monitor stations are required for capturing PM_2.5 spatial and temporal continuous variations with a sufficient resolution. This study utilized a land use regression (LUR) model with machine learning to assess the spatial-temporal variability of PM_2.5. Daily average PM_2.5 data was collected from 73 fixed air quality monitoring stations that belonged to the Taiwan EPA on the main island of Taiwan. Nearly 280,000 observations from 2006 to 2016 were used for the analysis. Several datasets were collected to determine spatial predictor variables, including the EPA environmental resources dataset, a meteorological dataset, a land-use inventory, a landmark dataset, a digital road network map, a digital terrain model, MODIS Normalized Difference Vegetation Index (NDVI) database, and a power plant distribution dataset. First, conventional LUR and Hybrid Kriging-LUR were utilized to identify the important predictor variables. Then, deep neural network, random forest, and XGBoost algorithms were used to fit the prediction model based on the variables selected by the LUR models. Data splitting, 10-fold cross validation, external data verification, and seasonal-based and county-based validation methods were used to verify the robustness of the developed models. The results demonstrated that the proposed conventional LUR and Hybrid Kriging-LUR models captured 58% and 89% of PM_2.5 variations, respectively. When XGBoost algorithm was incorporated, the explanatory power of the models increased to 73% and 94%, respectively. The Hybrid Kriging-LUR with XGBoost algorithm outperformed the other integrated methods. This study demonstrates the value of combining Hybrid Kriging-LUR model and an XGBoost algorithm for estimating the spatial-temporal variability of PM_2.5 exposures.

An interpretable self-adaptive deep neural network for estimating daily spatially-continuous PM2.5 concentrations across China

Accurate estimation of daily spatially-continuous PM_2.5 (fine particulate matter) concentration is a prerequisite to address environmental public health issues, and satellite-based aerosol optical depth (AOD) products have been widely used to estimate PM_2.5 concentrations using statistical-based or machine learning-based models. However, statistical-based models oversimplify the AOD-PM_2.5 relationships, whereas complex machine learning technologies ignore the spatiotemporal heterogeneity of the predictors and demonstrate shortage in interpretation. Besides, large AOD data gaps resulting in PM_2.5 estimation biases have been seldom imputed in previous studies, especially at national scales. To fill the above research gaps, this study attempts to present a feasible methodology to estimate daily spatially-continuous PM_2.5 concentrations in China. The AOD data gaps across China were first imputed via a random forest (RF) model. Then, an interpretable self-adaptive deep neural network (SADNN) model, incorporating AOD, meteorological and other auxiliary predictors, was developed to estimate daily spatially-continuous PM_2.5 concentrations from 2017 to 2018. Five-fold sample (site)-based cross-validation results showed a high accuracy of the SADNN model, with coefficient of determination and root mean square error values equal to 0.86 (0.84) and 13.07 (14.30) μg/m³, respectively, outperforming the standard DNN and the RF model. Furthermore, the SADNN model identified the spatiotemporal patterns of predictor importance, and demonstrated that the boundary layer height, elevation and AOD were the most important predictors both spatially and temporally. And the predictor importance in the Qinghai-Tibet Plateau was different from that in the rest of China. These results enhance our understanding of AOD-PM_2.5 relationships and elucidate the estimated PM_2.5 datasets with complete coverage are applicable for related air pollution studies and epidemiological cohort studies. Moreover, considering the effective nonlinear model capability and interpretability, the SADNN model is beneficial for not only PM_2.5 estimation but also other earth data and scenarios.

Ground PM2.5 prediction using imputed MAIAC AOD with uncertainty quantification

Satellite-derived aerosol optical depth (AOD) has been widely used to predict ground-level fine particulate matter (PM_2.5) concentrations, although its utility can be limited due to missing values. Despite recent attempts to address this issue by imputing missing satellite AOD values, the uncertainty associated with the AOD imputation and its impacts on PM_2.5 predictions have been understudied. To fill this gap, we developed a missing data imputation model for the AOD derived from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) and PM_2.5 prediction models using several machine learning methods. We also examined how the uncertainty associated with the imputed AOD and a choice of machine learning algorithm were propagated to PM_2.5 predictions. The application of the proposed imputation model to the data from New York State in the U.S. achieved a superior performance than those related studies, with a cross-validated R² of 0.94 and a Root Mean Square Error of 0.017. We also found that there was considerable uncertainty in PM_2.5 predictions associated with the use of imputed AOD values, although it was not as high as the uncertainty from the machine learning algorithms used in PM_2.5 prediction models. We concluded that the quantification of uncertainties for both AOD imputation and its propagation to AOD-based PM_2.5 prediction is necessary for accurate and reliable PM_2.5 predictions.

The Impact of Individual Mobility on Long-Term Exposure to Ambient PM2.5: Assessing Effect Modification by Travel Patterns and Spatial Variability of PM2.5

The impact of individuals' mobility on the degree of error in estimates of exposure to ambient PM2.5 concentrations is increasingly reported in the literature. However, the degree to which accounting for mobility reduces error likely varies as a function of two related factors-individuals' routine travel patterns and the local variations of air pollution fields. We investigated whether individuals' routine travel patterns moderate the impact of mobility on individual long-term exposure assessment. Here, we have used real-world time-activity data collected from 2013 participants in Erie/Niagara counties, New York, USA, matched with daily PM2.5 predictions obtained from two spatial exposure models. We further examined the role of the spatiotemporal representation of ambient PM2.5 as a second moderator in the relationship between an individual's mobility and the exposure measurement error using a random effect model. We found that the effect of mobility on the long-term exposure estimates was significant, but that this effect was modified by individuals' routine travel patterns. Further, this effect modification was pronounced when the local variations of ambient PM2.5 concentrations were captured from multiple sources of air pollution data ('a multi-sourced exposure model'). In contrast, the mobility effect and its modification were not detected when ambient PM2.5 concentration was estimated solely from sparse monitoring data ('a single-sourced exposure model'). This study showed that there was a significant association between individuals' mobility and the long-term exposure measurement error. However, the effect could be modified by individuals' routine travel patterns and the error-prone representation of spatiotemporal variability of PM2.5.

Impact of different exposure models and spatial resolution on the long-term effects of air pollution

Long-term exposure to air pollution has been related to mortality in several epidemiological studies. The investigations have assessed exposure using various methods achieving different accuracy in predicting air pollutants concentrations. The comparison of the health effects estimates are therefore challenging. This paper aims to compare the effect estimates of the long-term effects of air pollutants (particulate matter with aerodynamic diameter less than 10 μm, PM₁₀, and nitrogen dioxide, NO₂) on cause-specific mortality in the Rome Longitudinal Study, using exposure estimates obtained with different models and spatial resolutions. Annual averages of NO₂ and PM₁₀ were estimated for the year 2015 in a large portion of the Rome urban area (12 × 12 km²) applying three modelling techniques available at increasing spatial resolution: 1) a chemical transport model (CTM) at 1km resolution; 2) a land-use random forest (LURF) approach at 200m resolution; 3) a micro-scale Lagrangian particle dispersion model (PMSS) taking into account the effect of buildings structure at 4 m resolution with results post processed at different buffer sizes (12, 24, 52, 100 and 200 m). All the exposures were assigned at the residential addresses of 482,259 citizens of Rome 30+ years of age who were enrolled on 2001 and followed-up till 2015. The association between annual exposures and natural-cause, cardiovascular (CVD) and respiratory (RESP) mortality were estimated using Cox proportional hazards models adjusted for individual and area-level confounders. We found different distributions of both NO₂ and PM₁₀ concentrations, across models and spatial resolutions. Natural cause and CVD mortality outcomes were all positively associated with NO₂ and PM₁₀ regardless of the model and spatial resolution when using a relative scale of the exposure such as the interquartile range (IQR): adjusted Hazard Ratios (HR), and 95% confidence intervals (CI), of natural cause mortality, per IQR increments in the two pollutants, ranged between 1.012 (1.004, 1.021) and 1.018 (1.007, 1.028) for the different NO₂ estimates, and between 1.010 (1.000, 1.020) and 1.020 (1.008, 1.031) for PM₁₀, with a tendency of larger effect for lower resolution exposures. The latter was even stronger when a fixed value of 10 μg/m³ is used to calculate HRs. Long-term effects of air pollution on mortality in Rome were consistent across different models for exposure assessment, and different spatial resolutions.

Forecasting the Concentration of Particulate Matter in the Seoul Metropolitan Area Using a Gaussian Process Model

Recently, the population of Seoul has been affected by particulate matter in the atmosphere. This problem can be addressed by developing an elaborate forecasting model to estimate the concentration of fine dust in the metropolitan area. We present a forecasting model of the fine dust concentration with an extended range of input variables, compared to existing models. The model takes inputs from holistic perspectives such as topographical features on the surface, chemical sources of the fine dusts, traffic and the human activities in sub-areas, and meteorological data such as wind, temperature, and humidity, of fine dust. Our model was evaluated by the index-of-agreement (IOA) and the root mean-squared error (RMSE) in predicting PM2.5 and PM10 over three subsequent days. Our model variations consist of linear regressions, ARIMA, and Gaussian process regressions (GPR). The GPR showed the best performance in terms of IOA that is over 0.6 in the three-day predictions.

Estimation of High-Resolution PM2.5 over the Indo-Gangetic Plain by Fusion of Satellite Data, Meteorology, and Land Use Variables

Very high spatially resolved satellite-derived ground-level concentrations of particulate matter with an aerodynamic diameter of less than 2.5 μm (PM_2.5) have multiple potential applications, especially in air quality modeling and epidemiological and climatological research. Satellite-derived aerosol optical depth (AOD) and columnar water vapor (CWV), meteorological parameters, and land use data were used as variables within the framework of a linear mixed effect model (LME) and a random forest (RF) model to predict daily ground-level concentrations of PM_2.5 at 1 km × 1 km grid resolution across the Indo-Gangetic Plain (IGP) in South Asia. The RF model exhibited superior performance and higher accuracy compared with the LME model, with better cross-validated explained variance (R² = 0.87) and lower relative prediction error (RPE = 24.5%). The RF model revealed improved performance metrics for increasing averaging periods, from daily to weekly, monthly, seasonal, and annual means, which supported its use in estimating PM_2.5 exposure metrics across the IGP at varying temporal scales (i.e., both short and long terms). The RF-based PM_2.5 estimates showed high PM_2.5 levels over the middle and lower IGP, with the annual mean exceeding 110 μg/m³. As for seasons, winter was the most polluted season, while monsoon was the cleanest. Spatially, the middle and lower IGP showed poorer air quality compared to the upper IGP. In winter, the middle and lower IGP experienced very poor air quality, with mean PM_2.5 concentrations of >170 μg/m³.

A multi-city air pollution population exposure study: Combined use of chemical-transport and random-Forest models with dynamic population data

Cities are severely affected by air pollution. Local emissions and urban structures can produce large spatial heterogeneities. We aim to improve the estimation of NO₂, O₃, PM_2.5 and PM₁₀ concentrations in 6 Italian metropolitan areas, using chemical-transport and machine learning models, and to assess the effect on population exposure by using information on urban population mobility. Three years (2013-2015) of simulations were performed by the Chemical-Transport Model (CTM) FARM, at 1 km resolution, fed by boundary conditions provided by national-scale simulations, local emission inventories and meteorological fields. A downscaling of daily air pollutants at higher resolution (200 m) was then carried out by means of a machine learning Random-Forest (RF) model, considering CTM and spatial-temporal predictors, such as population, land-use, surface greenness and vehicular traffic, as input. RF achieved mean cross-validation (CV) R² of 0.59, 0.72, 0.76 and 0.75 for NO₂, PM₁₀, PM_2.5 and O₃, respectively, improving results from CTM alone. Mean concentration fields exhibited clear geographical gradients caused by climate conditions, local emission sources and photochemical processes. Time series of population weighted exposure (PWE) were estimated for two months of the year 2015 and for five cities, by combining population mobility data (derived from mobile phone traffic volumes data), and concentration levels from the RF model. PWE_RF metric better approximated the observed concentrations compared with the predictions from either CTM alone or CTM and RF combined, especially for pollutants exhibiting strong spatial gradients, such as NO₂. 50% of the population was estimated to be exposed to NO₂ concentrations between 12 and 38 μg/m³ and PM₁₀ between 20 and 35 μg/m³. This work supports the potential of machine learning methods in predicting air pollutant levels in urban areas at high spatial and temporal resolutions.

Estimation of Hourly near Surface Air Temperature Across Israel Using an Ensemble Model

Mapping of near-surface air temperature (Ta) at high spatio-temporal resolution is essential for unbiased assessment of human health exposure to temperature extremes, not least given the observed trend of urbanization and global climate change. Data constraints have led previous studies to focus merely on daily Ta metrics, rather than hourly ones, making them insufficient for intra-day assessment of health exposure. In this study, we present a three-stage machine learning-based ensemble model to estimate hourly Ta at a high spatial resolution of 1 × 1 km2, incorporating remotely sensed surface skin temperature (Ts) from geostationary satellites, reanalysis synoptic variables, and observations from weather stations, as well as auxiliary geospatial variables, which account for spatio-temporal variability of Ta. The Stage 1 model gap-fills hourly Ts at 4 × 4 km2 from the Spinning Enhanced Visible and InfraRed Imager (SEVIRI), which are subsequently fed into the Stage 2 model to estimate hourly Ta at the same spatio-temporal resolution. The Stage 3 model downscales the residuals between estimated and measured Ta to a grid of 1 × 1 km2, taking into account additionally the monthly diurnal pattern of Ts derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) data. In each stage, the ensemble model synergizes estimates from the constituent base learners—random forest (RF) and extreme gradient boosting (XGBoost)—by applying a geographically weighted generalized additive model (GAM), which allows the weights of results from individual models to vary over space and time. Demonstrated for Israel for the period 2004–2017, the proposed ensemble model outperformed each of the two base learners. It also attained excellent five-fold cross-validated performance, with overall root mean square error (RMSE) of 0.8 and 0.9 °C, mean absolute error (MAE) of 0.6 and 0.7 °C, and R2 of 0.95 and 0.98 in Stage 1 and Stage 2, respectively. The Stage 3 model for downscaling Ta residuals to 1 km MODIS grids achieved overall RMSE of 0.3 °C, MAE of 0.5 °C, and R2 of 0.63. The generated hourly 1 × 1 km2 Ta thus serves as a foundation for monitoring and assessing human health exposure to temperature extremes at a larger geographical scale, helping to further minimize exposure misclassification in epidemiological studies.

Soft Computing Applications in Air Quality Modeling: Past, Present, and Future

Air quality models simulate the atmospheric environment systems and provide increased domain knowledge and reliable forecasting. They provide early warnings to the population and reduce the number of measuring stations. Due to the complexity and non-linear behavior associated with air quality data, soft computing models became popular in air quality modeling (AQM). This study critically investigates, analyses, and summarizes the existing soft computing modeling approaches. Among the many soft computing techniques in AQM, this article reviews and discusses artificial neural network (ANN), support vector machine (SVM), evolutionary ANN and SVM, the fuzzy logic model, neuro-fuzzy systems, the deep learning model, ensemble, and other hybrid models. Besides, it sheds light on employed input variables, data processing approaches, and targeted objective functions during modeling. It was observed that many advanced, reliable, and self-organized soft computing models like functional network, genetic programming, type-2 fuzzy logic, genetic fuzzy, genetic neuro-fuzzy, and case-based reasoning are rarely explored in AQM. Therefore, the partially explored and unexplored soft computing techniques can be appropriate choices for research in the field of air quality modeling. The discussion in this paper will help to determine the suitability and appropriateness of a particular model for a specific modeling context.

A Random Forest Approach to Estimate Daily Particulate Matter, Nitrogen Dioxide, and Ozone at Fine Spatial Resolution in Sweden

Air pollution is one of the leading causes of mortality worldwide. An accurate assessment of its spatial and temporal distribution is mandatory to conduct epidemiological studies able to estimate long-term (e.g., annual) and short-term (e.g., daily) health effects. While spatiotemporal models for particulate matter (PM) have been developed in several countries, estimates of daily nitrogen dioxide (NO2) and ozone (O3) concentrations at high spatial resolution are lacking, and no such models have been developed in Sweden. We collected data on daily air pollutant concentrations from routine monitoring networks over the period 2005–2016 and matched them with satellite data, dispersion models, meteorological parameters, and land-use variables. We developed a machine-learning approach, the random forest (RF), to estimate daily concentrations of PM10 (PM<10 microns), PM2.5 (PM<2.5 microns), PM2.5–10 (PM between 2.5 and 10 microns), NO2, and O3 for each squared kilometer of Sweden over the period 2005–2016. Our models were able to describe between 64% (PM10) and 78% (O3) of air pollutant variability in held-out observations, and between 37% (NO2) and 61% (O3) in held-out monitors, with no major differences across years and seasons and better performance in larger cities such as Stockholm. These estimates will allow to investigate air pollution effects across the whole of Sweden, including suburban and rural areas, previously neglected by epidemiological investigations.

Novel Approaches to Air Pollution Exposure and Clinical Outcomes Assessment in Environmental Health Studies

An accurate assessment of pollutants’ exposure and precise evaluation of the clinical outcomes pose two major challenges to the contemporary environmental health research. The common methods for exposure assessment are based on residential addresses and are prone to many biases. Pollution levels are defined based on monitoring stations that are sparsely distributed and frequently distanced far from residential addresses. In addition, the degree of an association between outdoor and indoor air pollution levels is not fully elucidated, making the exposure assessment all the more inaccurate. Clinical outcomes’ assessment, on the other hand, mostly relies on the access to medical records from hospital admissions and outpatients’ visits in clinics. This method differentiates by health care seeking behavior and is therefore, problematic in evaluation of an onset, duration, and severity of an outcome. In the current paper, we review a number of novel solutions aimed to mitigate the aforementioned biases. First, a hybrid satellite-based modeling approach provides daily continuous spatiotemporal estimations with improved spatial resolution of 1 × 1 km2 and 200 × 200 m2 grid, and thus allows a more accurate exposure assessment. Utilizing low-cost air pollution sensors allowing a direct measurement of indoor air pollution levels can further validate these models. Furthermore, the real temporal-spatial activity can be assessed by GPS tracking devices within the individuals’ smartphones. A widespread use of smart devices can help with obtaining objective measurements of some of the clinical outcomes such as vital signs and glucose levels. Finally, human biomonitoring can be efficiently done at a population level, providing accurate estimates of in-vivo absorbed pollutants and allowing for the evaluation of body responses, by biomarkers examination. We suggest that the adoption of these novel methods will change the research paradigm heavily relying on ecological methodology and support development of the new clinical practices preventing adverse environmental effects on human health.