A Satellite-Based Spatio-Temporal Machine Learning Model to Reconstruct Daily PM2.5 Concentrations across Great Britain

Modelling extensions for multi-location studies in environmental epidemiology

Multi-location studies are increasingly used in environmental epidemiology. Their application is supported by designs and statistical techniques developed in the last decades, which however have known limitations. In this contribution, we propose an improved modelling framework that addresses these issues. Specifically, this flexible framework allows the direct modelling of demographic differences across locations, defining geographical variations linked to multiple vulnerability factors, capturing spatial heterogeneity and predicting risks to new locations, and improving the assessment of uncertainty. We illustrate these new developments in an analysis of temperature-mortality associations in Italian cities, providing fully reproducible R code and data.

Effect modification of air pollution on the association between heat and mortality in five European countries

BACKGROUND

Evidence suggests that air pollution modifies the association between heat and mortality. However, most studies have been conducted in cities without rural data. This time-series study examined potential effect modification of particulate matter (PM) and ozone (O3) on heat-related mortality using small-area data from five European countries, and explored the influence of area characteristics.

METHODS

We obtained daily non-accidental death counts from both urban and rural areas in Norway, England and Wales, Germany, Italy, and the Attica region of Greece during the warm season (2000-2018). Daily mean temperatures and air pollutant concentrations were estimated by spatial-temporal models. Heat effect modification by air pollution was assessed in each small area by over-dispersed Poisson regression models with a tensor smoother between temperature and air pollution. We extracted temperature-mortality relationships at the 5th (low), 50th (medium), and 95th (high) percentiles of pollutant distributions. At each air pollution level, we estimated heat-related mortality for a temperature increase from the 75th to the 99th percentile. We applied random-effects meta-analysis to derive the country-specific and overall associations, and mixed-effects meta-regression to examine the influence of urban-rural and coastal typologies and greenness on the heat effect modification by air pollution.

RESULTS

Heat-related mortality risks increased with higher PM levels, rising by 6.4% (95% CI: -2.0%-15.7%), 10.7% (2.6%-19.5%), and 14.1% (4.4%-24.6%) at low, medium, and high PM levels, respectively. This effect modification was consistent in urban and rural regions but more pronounced in non-coastal regions. In addition, heat-mortality associations were slightly stronger at high O3 levels, particularly in regions with low greenness.

CONCLUSION

Our analyses of both urban and rural data indicate that air pollution may intensify heat-related mortality, particularly in non-coastal and less green regions. The synergistic effect of heat and air pollution implies a potential pathway of reducing heat-related health impacts by improving air quality.

Development of land use regression, dispersion, and hybrid models for prediction of outdoor air pollution exposure in Barcelona

BACKGROUND

Air pollution is the leading environmental risk factor for health. Assessing outdoor air pollution exposure with detailed spatial and temporal variability in urban areas is crucial for evaluating its health effects.

AIM

We developed and compared Land Use Regression (LUR), dispersion (DM), and hybrid (HM) models to estimate outdoor concentrations for NO2, PM2.5, black carbon (BC), and PM2.5-constituents (Fe, Cu, Zn) in Barcelona.

METHODS

Two monitoring campaigns were conducted. In the first, NO2 concentrations were measured twice at 984 home addresses and in the second, NO2, PM2.5, and BC were measured four times at 34 points across Barcelona. LUR and DM were constructed using conventional techniques, while HM was developed using Random Forest (RF). Model performance was evaluated using leave-one-out cross-validation (LOOCV) and 10-fold cross-validation (10-CV) for LUR and HM, and by comparing DM and LUR estimates with routine monitoring stations. NO2 levels estimated by all models were externally validated using the home monitoring campaign. Agreement between models was assessed using Spearman correlation (rs) and Bland-Altman (BA) plots.

RESULTS

Models showed moderate to good performance. LUR exhibited R2LOOCV of 0.62 (NO2), 0.45 (PM2.5), 0.83 (BC), and 0.85 to 0.89 (PM2.5-constituents). DM model comparison showed R2 values of 0.39 (NO2), 0.26 (PM2.5), and 0.65 (BC). HM models had higher R210-CV 0.64 (NO2), 0.66 (PM2.5), 0.86 (BC), and 0.44 to 0.70 (PM2.5-constituents). Validation for NO2 showed R2 values of 0.56 (LUR), 0.44 (DM), and 0.64 (HM). Correlations between models varied from -0.38 to 0.92 for long-term exposure, and - 0.23 to 0.94 for short-term exposure. BA plots showed good agreement between models, especially for NO2 and BC.

CONCLUSIONS

Our models varied substantially, with some models performing better in validation samples (NO2 and BC). Future health studies should use the most accurate methods to minimize bias from exposure measurement error.

Air pollution mixture complexity and its effect on PM2.5-related mortality: A multicountry time-series study in 264 cities

Background

Fine particulate matter (PM2.5) occurs within a mixture of other pollutant gases that interact and impact its composition and toxicity. To characterize the local toxicity of PM2.5, it is useful to have an index that accounts for the whole pollutant mix, including gaseous pollutants. We consider a recently proposed pollutant mixture complexity index (PMCI) to evaluate to which extent it relates to PM2.5 toxicity.

Methods

The PMCI is constructed as an index spanning seven different pollutants, relative to the PM2.5 levels. We consider a standard two-stage analysis using data from 264 cities in the Northern Hemisphere. The first stage estimates the city-specific relative risks between daily PM2.5 and all-cause mortality, which are then pooled into a second-stage meta-regression model with which we estimate the effect modification from the PMCI.

Results

We estimate a relative excess risk of 1.0042 (95% confidence interval: 1.0023, 1.0061) for an interquartile range increase (from 1.09 to 1.95) of the PMCI. The PMCI predicts a substantial part of within-country relative risk heterogeneity with much less between-country heterogeneity explained. The Akaike information criterion and Bayesian information criterion of the main model are lower than those of alternative meta-regression models considering the oxidative capacity of PM2.5 or its composition.

Conclusions

The PMCI represents an efficient and simple predictor of local PM2.5-related mortality, providing evidence that PM2.5 toxicity depends on the surrounding gaseous pollutant mix. With the advent of remote sensing for pollutants, the PMCI can provide a useful index to track air quality.

Reconstructing individual-level exposures in cohort analyses of environmental risks: an example with the UK Biobank

Recent developments in linkage procedures and exposure modelling offer great prospects for cohort analyses on the health risks of environmental factors. However, assigning individual-level exposures to large population-based cohorts poses methodological and practical problems. In this contribution, we illustrate a linkage framework to reconstruct environmental exposures for individual-level epidemiological analyses, discussing methodological and practical issues such as residential mobility and privacy concerns. The framework outlined here requires the availability of individual residential histories with related time periods, as well as high-resolution spatio-temporal maps of environmental exposures. The linkage process is carried out in three steps: (1) spatial alignment of the exposure maps and residential locations to extract address-specific exposure series; (2) reconstruction of individual-level exposure histories accounting for residential changes during the follow-up; (3) flexible definition of exposure summaries consistent with alternative research questions and epidemiological designs. The procedure is exemplified by the linkage and processing of daily averages of air pollution for the UK Biobank cohort using gridded spatio-temporal maps across Great Britain. This results in the extraction of exposure summaries suitable for epidemiological analyses of both short and long-term risk associations and, in general, for the investigation of temporal dependencies. The linkage framework presented here is generally applicable to multiple environmental stressors and can be extended beyond the reconstruction of residential exposures. IMPACT: This contribution describes a linkage framework to assign individual-level environmental exposures to population-based cohorts using high-resolution spatio-temporal exposure. The framework can be used to address current limitations of exposure assessment for the analysis of health risks associated with environmental stressors. The linkage of detailed exposure information at the individual level offers the opportunity to define flexible exposure summaries tailored to specific study designs and research questions. The application of the framework is exemplified by the linkage of fine particulate matter (PM2.5) exposures to the UK Biobank cohort.

High resolution mapping of nitrogen dioxide and particulate matter in Great Britain (2003-2021) with multi-stage data reconstruction and ensemble machine learning methods

In this contribution, we applied a multi-stage machine learning (ML) framework to map daily values of nitrogen dioxide (NO2) and particulate matter (PM10 and PM2.5) at a 1 km2 resolution over Great Britain for the period 2003-2021. The process combined ground monitoring observations, satellite-derived products, climate reanalyses and chemical transport model datasets, and traffic and land-use data. Each feature was harmonized to 1 km resolution and extracted at monitoring sites. Models used single and ensemble-based algorithms featuring random forests (RF), extreme gradient boosting (XGB), light gradient boosting machine (LGBM), as well as lasso and ridge regression. The various stages focused on augmenting PM2.5 using co-occurring PM10 values, gap-filling aerosol optical depth and columnar NO2 data obtained from satellite instruments, and finally the training of an ensemble model and the prediction of daily values across the whole geographical domain (2003-2021). Results show a good ensemble model performance, calculated through a ten-fold monitor-based cross-validation procedure, with an average R2 of 0.690 (range 0.611-0.792) for NO2, 0.704 (0.609-0.786) for PM10, and 0.802 (0.746-0.888) for PM2.5. Reconstructed pollution levels decreased markedly within the study period, with a stronger reduction in the latter eight years. The pollutants exhibited different spatial patterns, while NO2 rose in close proximity to high-traffic areas, PM demonstrated variation at a larger scale. The resulting 1 km2 spatially resolved daily datasets allow for linkage with health data across Great Britain over nearly two decades, thus contributing to extensive, extended, and detailed research on the long-and short-term health effects of air pollution.

Air Pollution in Relation to COVID-19 Morbidity and Mortality: A Large Population-Based Cohort Study in Catalonia, Spain (COVAIR-CAT)

INTRODUCTION

Evidence from epidemiological studies based on individual-level data indicates that air pollution may be associated with coronavirus disease 2019 (COVID-19) severity. We aimed to test whether (1) long-term exposure to air pollution is associated with COVID-19-related hospital admission or mortality in the general population; (2) short-term exposure to air pollution is associated with COVID-19-related hospital admission following COVID-19 diagnosis; (3) there are vulnerable population subgroups; and (4) the influence of long-term air pollution exposure on COVID-19-related hospital admissions differed from that for other respiratory infections.

METHODS

We constructed a cohort covering nearly the full population of Catalonia through registry linkage, with follow- up from January 1, 2015, to December 31, 2020. Exposures at residential addresses were estimated using newly developed spatiotemporal models of nitrogen dioxide (NO23), particulate matter ≤2.5 μm in aerodynamic diameter (PM2.5), particulate matter ≤10 μm in aerodynamic diameter (PM10), and maximum 8-hr-average ozone (O3) at a spatial resolution of 250 m for the period 2018-2020.

RESULTS

The general population cohort included 4,660,502 individuals; in 2020 there were 340,608 COVID-19 diagnoses, 47,174 COVID-19-related hospital admissions, and 10,001 COVID-19 deaths. Mean (standard deviation) annual exposures were 26.2 (10.3) μg/m3 for NO2, 13.8 (2.2) μg/m3 for PM2.5, and 91.6 (8.2) μg/m3 for O3. In Aim 1, an increase of 16.1 μg/m3 NO2 was associated with a 25% (95% confidence interval [CI]: 22%-29%) increase in hospitalizations and an 18% (10%-27%) increase in deaths. In Aim 2, cumulative air pollution exposure over the previous 7 days was positively associated with COVID-19-related hospital admission in the second pandemic wave (June 20 to December 31, 2020). Associations of exposure were driven by exposure on the day of the hospital admission (lag0). Associations between short-term exposure to air pollution and COVID-19-related hospital admission were similar in all population subgroups. In Aim 3, individuals with lower individual- and area-level socioeconomic status (SES) were identified as particularly vulnerable to the effects of long-term exposure to NO2 and PM2.5 on COVID-19-related hospital admission. In Aim 4, long-term exposure to air pollution was associated with hospital admission for influenza and pneumonia: (6%; 95% CI: 2-11 per 16.4-μg/m3 NO2 and 5%; 1-8 per 2.6-μg/m3 PM2.5) as well as for all lower respiratory infections (LRIs) (18%; 14-22 per 16.4-μg/m3 NO2 and 14%; 11-17 per 2.6-μg/m3 PM2.5) before the COVID-19 pandemic. Associations for COVID-19-related hospital admission were larger than those for influenza or pneumonia for NO2, PM2.5, and O3 when adjusted for NO2.

CONCLUSIONS

Linkage across several registries allowed the construction of a large population-based cohort, tracking COVID-19 cases from primary care and testing data to hospital admissions, and death. Long- and short-term exposure to ambient air pollution were positively associated with severe COVID-19 events. The effects of long-term air pollution exposure on COVID-19 severity were greater among those with lower individual- and area-level SES.

Parsimonious Random-Forest-Based Land-Use Regression Model Using Particulate Matter Sensors in Berlin, Germany

Machine learning (ML) methods are widely used in particulate matter prediction modelling, especially through use of air quality sensor data. Despite their advantages, these methods' black-box nature obscures the understanding of how a prediction has been made. Major issues with these types of models include the data quality and computational intensity. In this study, we employed feature selection methods using recursive feature elimination and global sensitivity analysis for a random-forest (RF)-based land-use regression model developed for the city of Berlin, Germany. Land-use-based predictors, including local climate zones, leaf area index, daily traffic volume, population density, building types, building heights, and street types were used to create a baseline RF model. Five additional models, three using recursive feature elimination method and two using a Sobol-based global sensitivity analysis (GSA), were implemented, and their performance was compared against that of the baseline RF model. The predictors that had a large effect on the prediction as determined using both the methods are discussed. Through feature elimination, the number of predictors were reduced from 220 in the baseline model to eight in the parsimonious models without sacrificing model performance. The model metrics were compared, which showed that the parsimonious_GSA-based model performs better than does the baseline model and reduces the mean absolute error (MAE) from 8.69 µg/m3 to 3.6 µg/m3 and the root mean squared error (RMSE) from 9.86 µg/m3 to 4.23 µg/m3 when applying the trained model to reference station data. The better performance of the GSA_parsimonious model is made possible by the curtailment of the uncertainties propagated through the model via the reduction of multicollinear and redundant predictors. The parsimonious model validated against reference stations was able to predict the PM2.5 concentrations with an MAE of less than 5 µg/m3 for 10 out of 12 locations. The GSA_parsimonious performed best in all model metrics and improved the R2 from 3% in the baseline model to 17%. However, the predictions exhibited a degree of uncertainty, making it unreliable for regional scale modelling. The GSA_parsimonious model can nevertheless be adapted to local scales to highlight the land-use parameters that are indicative of PM2.5 concentrations in Berlin. Overall, population density, leaf area index, and traffic volume are the major predictors of PM2.5, while building type and local climate zones are the less significant predictors. Feature selection based on sensitivity analysis has a large impact on the model performance. Optimising models through sensitivity analysis can enhance the interpretability of the model dynamics and potentially reduce computational costs and time when modelling is performed for larger areas.

Model misspecification, measurement error, and apparent supralinearity in the concentration-response relationship between PM2.5 and mortality

A growing number of studies have produced results that suggest the shape of the concentration-response (C-R) relationship between PM2.5 exposure and mortality is "supralinear" such that incremental risk is higher at the lowest exposure levels than at the highest exposure levels. If the C-R function is in fact supralinear, then there may be significant health benefits associated with reductions in PM2.5 below the current US National Ambient Air Quality Standards (NAAQS), as each incremental tightening of the PM2.5 NAAQS would be expected to produce ever-greater reductions in mortality risk. In this paper we undertake a series of tests with simulated cohort data to examine whether there are alternative explanations for apparent supralinearity in PM2.5 C-R functions. Our results show that a linear C-R function for PM2.5 can falsely appear to be supralinear in a statistical estimation process for a variety of reasons, such as spatial variation in the composition of total PM2.5 mass, the presence of confounders that are correlated with PM2.5 exposure, and some types of measurement error in estimates of PM2.5 exposure. To the best of our knowledge, this is the first simulation-based study to examine alternative explanations for apparent supralinearity in C-R functions.

Predicting particulate matter, nitrogen dioxide, and ozone across Great Britain with high spatiotemporal resolution based on random forest models

In Great Britain, limited studies have employed machine learning methods to predict air pollution especially ozone (O3) with high spatiotemporal resolution. This study aimed to address this gap by developing random forest models for four key pollutants (fine and inhalable particulate matter [PM2.5 and PM10], nitrogen dioxide [NO2] and O3) by integrating multiple-source predictors at a daily level and 1-km resolution. The out-of-bag R2 (root mean squared error, RMSE) between predictions from models and measurements from monitoring stations in 2006-2013 was 0.85 (3.63 μg/m3) for PM2.5, 0.77 (6.00 μg/m3) for PM10, 0.85 (9.71 μg/m3) for NO2, and 0.85 (9.39 μg/m3) for maximum daily 8-h average (MDA8) O3 at daily level, and the predicting accuracy was higher at monthly and annual level. The high-resolution predictions captured characterized spatiotemporal patterns of the four pollutants. Higher concentrations of PM2.5, PM10, and NO2 were distributed in densely populated southern regions of Great Britain while O3 showed an inverse spatial pattern in general, which could not be fully depicted by monitoring stations. Therefore, predictions produced in this study could improve exposure assessment with less exposure misclassification and flexible exposure windows for future epidemiological studies to investigate the impact of air pollution across Great Britain.

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.

Mortality burden attributable to exceptional PM2.5 air pollution events in Australian cities: A health impact assessment

Background

People living in Australian cities face increased mortality risks from exposure to extreme air pollution events due to bushfires and dust storms. However, the burden of mortality attributable to exceptional PM2.5 levels has not been well characterised. We assessed the burden of mortality due to PM2.5 pollution events in Australian capital cities between 2001 and 2020.

Methods

For this health impact assessment, we obtained data on daily counts of deaths for all non-accidental causes and ages from the Australian National Vital Statistics Register. Daily concentrations of PM2.5 were estimated at a 5 km grid cell, using a Random Forest statistical model of data from air pollution monitoring sites combined with a range of satellite and land use-related data. We calculated the exceptional PM2.5 levels for each extreme pollution exposure day using the deviation from a seasonal and trend loess decomposition model. The burden of mortality was examined using a relative risk concentration-response function suggested in the literature.

Findings

Over the 20-year study period, we estimated 1454 (95 % CI 987, 1920) deaths in the major Australian cities attributable to exceptional PM2.5 exposure levels. The mortality burden due to PM2.5 exposure on extreme pollution days was considerable. Variations were observed across Australia. Despite relatively low daily PM2.5 levels compared to global averages, all Australian cities have extreme pollution exposure days, with PM2.5 concentrations exceeding the World Health Organisation Air Quality Guideline standard for 24-h exposure. Our analysis results indicate that nearly one-third of deaths from extreme air pollution exposure can be prevented with a 5 % reduction in PM2.5 levels on days with exceptional pollution.

Interpretation

Exposure to exceptional PM2.5 events was associated with an increased mortality burden in Australia's cities. Policies and coordinated action are needed to manage the health risks of extreme air pollution events due to bushfires and dust storms under climate change.

Integrating Augmented In Situ Measurements and a Spatiotemporal Machine Learning Model To Back Extrapolate Historical Particulate Matter Pollution over the United Kingdom: 1980-2019

Historical PM2.5 data are essential for assessing the health effects of air pollution exposure across the life course or early life. However, a lack of high-quality data sources, such as satellite-based aerosol optical depth before 2000, has resulted in a gap in spatiotemporally resolved PM2.5 data for historical periods. Taking the United Kingdom as an example, we leveraged the light gradient boosting model to capture the spatiotemporal association between PM2.5 concentrations and multi-source geospatial predictors. Augmented PM2.5 from PM10 measurements expanded the spatiotemporal representativeness of the ground measurements. Observations before and after 2009 were used to train and test the models, respectively. Our model showed fair prediction accuracy from 2010 to 2019 [the ranges of coefficients of determination (R2) for the grid-based cross-validation are 0.71-0.85] and commendable back extrapolation performance from 1998 to 2009 (the ranges of R2 for the independent external testing are 0.32-0.65) at the daily level. The pollution episodes in the 1980s and pollution levels in the 1990s were also reproduced by our model. The 4-decade PM2.5 estimates demonstrated that most regions in England witnessed significant downward trends in PM2.5 pollution. The methods developed in this study are generalizable to other data-rich regions for historical air pollution exposure assessment.

Characterising personal, household, and community PM2.5 exposure in one urban and two rural communities in China

BACKGROUND

Cooking and heating in households contribute importantly to air pollution exposure worldwide. However, there is insufficient investigation of measured fine particulate matter (PM2.5) exposure levels, variability, seasonality, and inter-spatial dynamics associated with these behaviours.

METHODS

We undertook parallel measurements of personal, household (kitchen and living room), and community PM2.5 in summer (May-September 2017) and winter (November 2017-Janauary 2018) in 477 participants from one urban and two rural communities in China. After stringent data cleaning, there were 67,326-80,980 person-hours (ntotal = 441; nsummer = 384; nwinter = 364; 307 had repeated PM2.5 data in both seasons) of processed data per microenvironment. Age- and sex-adjusted geometric means of PM2.5 were calculated by key participant characteristics, overall and by season. Spearman correlation coefficients between PM2.5 levels across different microenvironments were computed.

FINDINGS

Overall, 26.4 % reported use of solid fuel for both cooking and heating. Solid fuel users had 92 % higher personal and kitchen 24-h average PM2.5 exposure than clean fuel users. Similarly, they also had a greater increase (83 % vs 26 %) in personal and household PM2.5 from summer to winter, whereas community levels of PM2.5 were 2-4 times higher in winter across different fuel categories. Compared with clean fuel users, solid fuel users had markedly higher weighted annual average PM2.5 exposure at personal (78.2 [95 % CI 71.6-85.3] μg/m3 vs 41.6 [37.3-46.5] μg/m3), kitchen (102.4 [90.4-116.0] μg/m3 vs 52.3 [44.8-61.2] μg/m3) and living room (62.1 [57.3-67.3] μg/m3 vs 41.0 [37.1-45.3] μg/m3) microenvironments. There was a remarkable diurnal variability in PM2.5 exposure among the participants, with 5-min moving average from 10 μg/m3 to 700-1200 μg/m3 across different microenvironments. Personal PM2.5 was moderately correlated with living room (Spearman r: 0.64-0.66) and kitchen (0.52-0.59) levels, but only weakly correlated with community levels, especially in summer (0.15-0.34) and among solid fuel users (0.11-0.31).

CONCLUSION

Solid fuel use for cooking and heating was associated with substantially higher personal and household PM2.5 exposure than clean fuel users. Household PM2.5 appeared a better proxy of personal exposure than community PM2.5.

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.

Prediction of daily mean and one-hour maximum PM2.5 concentrations and applications in Central Mexico using satellite-based machine-learning models

BACKGROUND

Machine-learning algorithms are becoming popular techniques to predict ambient air PM_2.5 concentrations at high spatial resolutions (1 × 1 km) using satellite-based aerosol optical depth (AOD). Most machine-learning models have aimed to predict 24 h-averaged PM_2.5 concentrations (mean PM_2.5) in high-income regions. Over Mexico, none have been developed to predict subdaily peak levels, such as the maximum daily 1-h concentration (max PM_2.5).

OBJECTIVE

Our goal was to develop a machine-learning model to predict mean PM_2.5 and max PM_2.5 concentrations in the Mexico City Metropolitan Area from 2004 through 2019.

METHODS

We present a new modeling approach based on extreme gradient boosting (XGBoost) and inverse-distance weighting that uses AOD, meteorology, and land-use variables. We also investigated applications of our mean PM_2.5 predictions that can aid local authorities in air-quality management and public-health surveillance, such as the co-occurrence of high PM_2.5 and heat, compliance with local air-quality standards, and the relationship of PM_2.5 exposure with social marginalization.

RESULTS

Our models for mean and max PM_2.5 exhibited good performance, with overall cross-validated mean absolute errors (MAE) of 3.68 and 9.20 μg/m³, respectively, compared to mean absolute deviations from the median (MAD) of 8.55 and 15.64 μg/m³. In 2010, everybody in the study region was exposed to unhealthy levels of PM_2.5. Hotter days had greater PM_2.5 concentrations. Finally, we found similar exposure to PM_2.5 across levels of social marginalization.

SIGNIFICANCE

Machine learning algorithms can be used to predict highly spatiotemporally resolved PM_2.5 concentrations even in regions with sparse monitoring.

IMPACT

Our PM_2.5 predictions can aid local authorities in air-quality management and public-health surveillance, and they can advance epidemiological research in Central Mexico with state-of-the-art exposure assessment methods.

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.

PM2.5 Concentrations Variability in North China Explored with a Multi-Scale Spatial Random Effect Model

Compiling fine-resolution geospatial PM_2.5 concentrations data is essential for precisely assessing the health risks of PM_2.5 pollution exposure as well as for evaluating environmental policy effectiveness. In most previous studies, global and local spatial heterogeneity of PM_2.5 is captured by the inclusion of multi-scale covariate effects, while the modelling of genuine scale-dependent variabilities pertaining to the spatial random process of PM_2.5 has not yet been much studied. Consequently, this work proposed a multi-scale spatial random effect model (MSSREM), based a recently developed fixed-rank Kriging method, to capture both the scale-dependent variabilities and the spatial dependence effect simultaneously. Furthermore, a small-scale Monte Carlo simulation experiment was conducted to assess the performance of MSSREM against classic geospatial Kriging models. The key results indicated that when the multiple-scale property of local spatial variabilities were exhibited, the MSSREM had greater ability to recover local- or fine-scale variations hidden in a real spatial process. The methodology was applied to the PM_2.5 concentrations modelling in North China, a region with the worst air quality in the country. The MSSREM provided high prediction accuracy, 0.917 R-squared, and 3.777 root mean square error (RMSE). In addition, the spatial correlations in PM_2.5 concentrations were properly captured by the model as indicated by a statistically insignificant Moran's I statistic (a value of 0.136 with p-value > 0.2). Overall, this study offers another spatial statistical model for investigating and predicting PM_2.5 concentration, which would be beneficial for precise health risk assessment of PM_2.5 pollution exposure.

Spatial Modelling and Prediction with the Spatio-Temporal Matrix: A Study on Predicting Future Settlement Growth

In the past decades, various Earth observation-based time series products have emerged, which have enabled studies and analysis of global change processes. Besides their contribution to understanding past processes, time series datasets hold enormous potential for predictive modeling and thereby meet the demands of decision makers on future scenarios. In order to further exploit these data, a novel pixel-based approach has been introduced, which is the spatio-temporal matrix (STM). The approach integrates the historical characteristics of a specific land cover at a high temporal frequency in order to interpret the spatial and temporal information for the neighborhood of a given target pixel. The provided information can be exploited with common predictive models and algorithms. In this study, this approach was utilized and evaluated for the prediction of future urban/built-settlement growth. Random forest and multi-layer perceptron were employed for the prediction. The tests have been carried out with training strategies based on a one-year and a ten-year time span for the urban agglomerations of Surat (India), Ho-Chi-Minh City (Vietnam), and Abidjan (Ivory Coast). The slope, land use, exclusion, urban, transportation, hillshade (SLEUTH) model was selected as a baseline indicator for the performance evaluation. The statistical results from the receiver operating characteristic curve (ROC) demonstrate a good ability of the STM to facilitate the prediction of future settlement growth and its transferability to different cities, with area under the curve (AUC) values greater than 0.85. Compared with SLEUTH, the STM-based model achieved higher AUC in all of the test cases, while being independent of the additional datasets for the restricted and the preferential development areas.

Machine Learning-Based Approach Using Open Data to Estimate PM2.5 over Europe

Air pollution is currently considered one of the most serious problems facing humans. Fine particulate matter with a diameter smaller than 2.5 micrometres (PM2.5) is a very harmful air pollutant that is linked with many diseases. In this study, we created a machine learning-based scheme to estimate PM2.5 using various open data such as satellite remote sensing, meteorological data, and land variables to increase the limited spatial coverage provided by ground-monitors. A space-time extremely randomised trees model was used to estimate PM2.5 concentrations over Europe, this model achieved good results with an out-of-sample cross-validated R2 of 0.69, RMSE of 5 μg/m3, and MAE of 3.3 μg/m3. The outcome of this study is a daily full coverage PM2.5 dataset with 1 km spatial resolution for the three-year period of 2018–2020. We found that air quality improved throughout the study period over all countries in Europe. In addition, we compared PM2.5 levels during the COVID-19 lockdown during the months March–June with the average of the previous 4 months and the following 4 months. We found that this lockdown had a positive effect on air quality in most parts of the study area except for the United Kingdom, Ireland, north of France, and south of Italy. This is the first study that depends only on open data and covers the whole of Europe with high spatial and temporal resolutions. The reconstructed dataset will be published under free and open license and can be used in future air quality studies.

High Spatial-Temporal PM2.5 Modeling Utilizing Next Generation Weather Radar (NEXRAD) as a Supplementary Weather Source

PM2.5, a type of fine particulate with a diameter equal to or less than 2.5 micrometers, has been identified as a major source of air pollution, and is associated with many health issues. Research on utilizing various data sources, such as remote sensing and in situ sensors, for PM2.5 concentrations modeling remains a hot topic. In this study, the Next Generation Weather Radar (NEXRAD) is used as a supplementary weather data source, along with European Centre for Medium-Range Weather Forecasts (ECMWF), solar angles, and Geostationary Operational Environmental Satellite (GOES16) Aerosol Optical Depth (AOD) to model high spatial-temporal PM2.5 concentrations. PM2.5 concentrations as well as in situ weather condition variables are collected from the 31 sensors that are deployed in the Dallas Metropolitan area. Four machine learning models with different predictor variables are developed based on an ensemble approach. Since in situ weather observations are not widely available, ECMWF is used as an alternative data source for weather conditions in studies. Hence, the four established models are compared in three groups. Both models in this first group use weather variables collected from deployed sensors, but one uses NEXRAD and the other does not. In the second group, the two models use weather variables retrieved from ECMWF, one using NEXRAD and one without. In the third group, one model uses weather variables from ECMWF, and the other uses in situ weather variables, both without NEXRAD. The first two environmental groups investigate how NEXRAD can enhance model performances with weather variables collected from in situ observations and ECMWF, respectively. The third group explores how effective using ECMWF as an alternative source of weather conditions. Based on the results, the incorporation of NEXRAD achieves an R2 score of 0.86 and 0.83 for groups 1 and 2, respectively, for an improvement of 2.8% and 9.6% over those models without NEXRAD. For group three, the use of ECMWF as an alternative source of in situ weather observations results in a 0.13 R2 drop. For PM2.5 estimation, weather variables including precipitation, temperature, pressure, and surface pressure from ECMWF and deployed sensors, as well as NEXRAD velocity, are shown to be significant factors.

Interpolation-Based Fusion of Sentinel-5P, SRTM, and Regulatory-Grade Ground Stations Data for Producing Spatially Continuous Maps of PM2.5 Concentrations Nationwide over Thailand

Atmospheric pollution has recently drawn significant attention due to its proven adverse effects on public health and the environment. This concern has been aggravated specifically in Southeast Asia due to increasing vehicular use, industrial activity, and agricultural burning practices. Consequently, elevated PM2.5 concentrations have become a matter of intervention for national authorities who have addressed the needs of monitoring air pollution by operating ground stations. However, their spatial coverage is limited and the installation and maintenance are costly. Therefore, alternative approaches are necessary at national and regional scales. In the current paper, we investigated interpolation models to fuse PM2.5 measurements from ground stations and satellite data in an attempt to produce spatially continuous maps of PM2.5 nationwide over Thailand. Four approaches are compared, namely the inverse distance weighted (IDW), ordinary kriging (OK), random forest (RF), and random forest combined with OK (RFK) leveraging on the NO2, SO2, CO, HCHO, AI, and O3 products from the Sentinel-5P satellite, regulatory-grade ground PM2.5 measurements, and topographic parameters. The results suggest that RFK is the most robust, especially when the pollution levels are moderate or extreme, achieving an RMSE value of 7.11 μg/m3 and an R2 value of 0.77 during a 10-day long period in February, and an RMSE of 10.77 μg/m3 and R2 and 0.91 during the entire month of March. The proposed approach can be adopted operationally and expanded by leveraging regulatory-grade stations, low-cost sensors, as well as upcoming satellite missions such as the GEMS and the Sentinel-5.

PM2.5 Modeling and Historical Reconstruction over the Continental USA Utilizing GOES-16 AOD

In this study, we present a nationwide machine learning model for hourly PM2.5 estimation for the continental United States (US) using high temporal resolution Geostationary Operational Environmental Satellites (GOES-16) Aerosol Optical Depth (AOD) data, meteorological variables from the European Center for Medium Range Weather Forecasting (ECMWF) and ancillary data collected between May 2017 and December 2020. A model sensitivity analysis was conducted on predictor variables to determine the optimal model. It turns out that GOES16 AOD, variables from ECMWF, and ancillary data are effective variables in PM2.5 estimation and historical reconstruction, which achieves an average mean absolute error (MAE) of 3.0 μg/m3, and a root mean square error (RMSE) of 5.8 μg/m3. This study also found that the model performance as well as the site measured PM2.5 concentrations demonstrate strong spatial and temporal patterns. Specifically, in the temporal scale, the model performed best between 8:00 p.m. and 11:00 p.m. (UTC TIME) and had the highest coefficient of determination (R2) in Autumn and the lowest MAE and RMSE in Spring. In the spatial scale, the analysis results based on ancillary data show that the R2 scores correlate positively with the mean measured PM2.5 concentration at monitoring sites. Mean measured PM2.5 concentrations are positively correlated with population density and negatively correlated with elevation. Water, forests, and wetlands are associated with low PM2.5 concentrations, whereas developed, cultivated crops, shrubs, and grass are associated with high PM2.5 concentrations. In addition, the reconstructed PM2.5 surfaces serve as an important data source for pollution event tracking and PM2.5 analysis. For this purpose, from May 2017 to December 2020, hourly PM2.5 estimates were made for 10 km by 10 km and the PM2.5 estimates from August through November 2020 during the period of California Santa Clara Unite (SCU) Lightning Complex fires are presented. Based on the quantitative and visualization results, this study reveals that a number of large wildfires in California had a profound impact on the value and spatial-temporal distributions of PM2.5 concentrations.

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.

Space-Time Machine Learning Models to Analyze COVID-19 Pandemic Lockdown Effects on Aerosol Optical Depth over Europe

The recent COVID-19 pandemic affected various aspects of life. Several studies established the consequences of pandemic lockdown on air quality using satellite remote sensing. However, such studies have limitations, including low spatial resolution or incomplete spatial coverage. Therefore, in this paper, we propose a machine learning-based scheme to solve the pre-mentioned limitations by training an optimized space-time extra trees model for each year of the study period. The results have shown that our trained models reach a prediction accuracy up to 95% when predicting the missing values in the MODIS MCD19A2 Aerosol Optical Depth (AOD) product. The outcome of the mentioned scheme was a geo-harmonized atmospheric dataset for aerosol optical depth at 550 nm with 1 km spatial resolution and full coverage over Europe. As an application, we used the proposed machine learning based prediction approach in AOD levels analysis. We compared the mean AOD levels between the lockdown period from March to June in 2020 and the mean AOD values of the same period for the past 5 years. We found that AOD levels dropped over most European countries in 2020 but increased in several eastern and western countries. The Netherlands had the most significant average decrease in AOD levels (19%), while Spain had the highest average increase (10%). Moreover, we analyzed the relationship between the relative percentage difference of AOD and four meteorological variables. We found a positive correlation between AOD and relative humidity and a negative correlation between AOD and wind speed. The value of the proposed prediction scheme is further emphasized by taking into consideration that the reconstructed dataset can be used for future air quality studies concerning Europe.

A Novel Recursive Model Based on a Convolutional Long Short-Term Memory Neural Network for Air Pollution Prediction

Deep learning provides a promising approach for air pollution prediction. The existing deep learning-based predicted models generally consider either the temporal correlations of air quality monitoring stations or the nonlinear relationship between the PM2.5 (particulate matter with an aerodynamic diameter of less than 2.5 μm) concentrations and explanatory variables. Spatial correlation has not been effectively incorporated into prediction models, therefore exhibiting poor performance in PM2.5 prediction tasks. Additionally, determining the manner by which to expand longer-term prediction tasks is still challenging. In this paper, to allow for spatiotemporal correlations, a spatiotemporal convolutional recursive long short-term memory (CR-LSTM) neural network model is proposed for predicting the PM2.5 concentrations in long-term prediction tasks by combining a convolutional long short-term memory (ConvLSTM) neural network and a recursive strategy. Herein, the ConvLSTM network was used to capture the complex spatiotemporal correlations and to predict the future PM2.5 concentrations; the recursive strategy was used for expanding the long-term prediction tasks. The CR-LSTM model was used to realize the prediction of the future 24 h of PM2.5 concentrations for 12 air quality monitoring stations in Beijing by configuring both the appropriate time lag derived from the temporal correlations and the spatial neighborhood, including the hourly historical PM2.5 concentrations, the daily mean meteorological data, and the annual nighttime light and normalized difference vegetation index (NDVI). The results showed that the proposed CR-LSTM model achieved better performance (coefficient of determination (R2) = 0.74; root mean square error (RMSE) = 18.96 μg/m3) than other common models, such as multiple linear regression (MLR), support vector regression (SVR), the conventional LSTM model, the LSTM extended (LSTME) model, and the temporal sliding LSTM extended (TS-LSTME) model. The proposed CR-LSTM model, implementing a combination of geographical rules, recursive strategy, and deep learning, shows improved performance in longer-term prediction tasks.

Cytokine fields and the polarization of the immune response

Certain cells that participate in the immune response are known to become polarized in their production of cytokines. It is postulated that, after initial polarization at the site of antigenic encounter, the different types of cell arriving at this site are induced to conform to the local cytokine field, implying that they share common regulatory circuits. As they migrate, these cells might, in turn, spread the particular cytokine field. Therefore, the field is 'infectious' in nature. Propagation of the cytokine field must be regulated somehow. The invasion of the cytokine field into an organ or the entire body could have major immunological consequences.