An improved meteorological variables-based aerosol optical depth estimation method by combining a physical mechanism model with a two-stage model.

A two-stage model integrating a spatiotemporal linear mixed effect (STLME) and a geographic weight regression (GWR) model is proposed to improve the meteorological variables-based aerosol optical depth (AOD) retrieval method (Elterman retrieval model-ERM). The proposed model is referred to as the STG-ERM model. The STG-ERM model is applied over the Beijing-Tianjin-Hebei (BTH) region in China for the years 2019 and 2020. The results show that data coverage increased by 39.0% in 2019 and 40.5% in 2020. Cross-validation of the retrieval results versus multi-angle implementation of atmospheric correction (MAIAC) AOD shows the substantial improvement of the STG-ERM model over earlier meteorological models for AOD estimation, with a determination coefficient (R2) of daily AOD of 0.86, root mean squared prediction error (RMSE) and the relative prediction error (RPE) of 0.10 and 36.14% in 2019 and R2 of 0.86, RMSE of 0.12 and RPE of 37.86% in 2020. The fused annual mean AOD indicates strong spatial variation with high value in south plain and low value in northwestern mountainous areas of the BTH region. The overall spatial seasonal mean AOD ranges from 0.441 to 0.586, demonstrating strongly seasonal variation. The coverage of STG-ERM retrieved AOD, as determined in this exercise by leaving out part of the meteorological data, affects the accuracy of fused AOD. The coverage of the meteorological data has smaller impact on the fused AOD in the districts with low annual mean AOD of less than 0.35 than that in the districts with high annual mean AOD of greater than 0.6. If available, continuous daily meteorological data with high spatiotemporal resolution can improve the model performance and the accuracy of fused AOD. The STG-ERM model may serve as a valuable approach to provide data to fill gaps in satellite-retrieved AOD products.

Evaluation and analysis of long-term MODIS MAIAC aerosol products in China.

NASA has released the latest Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle Implementation of Atmospheric Correction (MAIAC) Collection 6 (C6) and Collection 6.1 (C6.1) aerosol optical depth (AOD) products with 1 km spatial resolution. This study validated and compared C6 and C6.1 MAIAC AOD products with AERONET observations in terms of accuracy and stability, and analyzed the spatiotemporal characteristics of AOD at different scales in China. The results show that the overall accuracy of MAIAC products is good, with correlation coefficient (R) > 0.9, mean bias (BIAS) < 0.015, and the fraction within the expected error (EE) > 68 %. However, after the algorithm update, the accuracy of Terra MAIAC aerosol products C6.1 has significantly decreased. The accuracy of the products varies with the region. The accuracy of C6.1 in North China, Central East China, and West China, is comparable to or even exceeds that of C6, but performs poorly in South China. In addition, the stability of the updated C6.1 MAIAC aerosol products has not seen significantly improvement. The metrics of no product can all meet the stability goals of the Global Climate Observing System (GCOS, 0.02 per decade) in China. C6.1 improves the retrieval frequency in many regions and temporarily solves the problem of AOD discontinuity at the boundaries of different aerosol models in C6, but there are some fixed climatological AOD blocks (AOD = 0.014) in the eastern Tibetan Plateau region. Both C6 and C6.1 can capture similar annual variation characteristics of AOD to those observed at the AERONET sites. The study provides possible references for improving the MAIAC algorithm and building long-term stable aerosol records.

[Remote Sensing Model for Estimating Atmospheric PM2.5 Concentration in the Guangdong-Hong Kong-Macao Greater Bay Area].

PM2.5 is extremely harmful to the atmospheric environment and human health, and a timely and accurate understanding of PM2.5 with high spatial and temporal resolution plays an important role in the prevention and control of air pollution. Based on multi-angle implementation of atmospheric correction algorithm (MAIAC), 1 km AOD products, ERA5 meteorological data, and pollutant concentrations (CO, O3, NO2, SO2, PM10, and PM2.5) in the Guangdong-Hong Kong-Macao Greater Bay Area during 2015-2020, a geographically and temporally weighted regression model (GTWR), BP neural network model (BPNN), support vector machine regression model (SVR), and random forest model (RF) were established, respectively, to estimate PM2.5 concentration. The results showed that the estimation ability of the RF model was better than that of the BPNN, SVR, and GTWR models. The correlation coefficients of the BPNN, SVR, GTWR, and RF models were 0.922, 0.920, 0.934, and 0.981, respectively. The RMSE values were 7.192, 7.101, 6.385, and 3.670 µg·m-3. The MAE values were 5.482, 5.450, 4.849, and 2.323 µg·m-3, respectively. The RF model had the best effect during winter, followed by that during summer, and again during spring and autumn, with correlation coefficients above 0.976 in the prediction of different seasons. The RF model could be used to predict the PM2.5 concentration in the Greater Bay Area. In terms of time, the daily ρ(PM2.5) of cities in the Greater Bay Area showed a trend of "decreasing first and then increasing" in 2021, with the highest values ranging from 65.550 µg·m-3 to 112.780 µg·m-3 and the lowest values ranging from 5.000 µg·m-3 to 7.899 µg·m-3. The monthly average concentration showed a U-shaped distribution, and the concentration began to decrease in January and gradually increased after reaching a trough in June. Seasonally, it was characterized by the highest concentration during winter, the lowest during summer, and the transition during spring and autumn. The annual average ρ(PM2.5) of the Greater Bay Area was 28.868 µg·m-3, which was lower than the secondary concentration limit. Spatially, there was a "northwest to southeast" decreasing distribution of PM2.5 in 2021, and the high-pollution areas clustered in the central part of the Greater Bay Area, represented by Foshan. Low concentration areas were mainly distributed in the eastern part of Huizhou, Hong Kong, Macao, Zhuhai, and other coastal areas. The spatial distribution of PM2.5 in different seasons also showed heterogeneity and regionality. The RF model estimated the PM2.5 concentration with high accuracy, which provides a scientific basis for the health risk assessment associated with PM2.5 pollution in the Greater Bay Area.

Evaluating TROPOMI and MODIS performance to capture the dynamic of air pollution in São Paulo state: A case study during the COVID-19 outbreak.

Atmospheric pollutant data retrieved through satellite sensors are continually used to assess changes in air quality in the lower atmosphere. During the COVID-19 pandemic, several studies started to use satellite measurements to evaluate changes in air quality in many different regions worldwide. However, although satellite data is continuously validated, it is known that its accuracy may vary between monitored areas, requiring regionalized quality assessments. Thus, this study aimed to evaluate whether satellites could measure changes in the air quality of the state of São Paulo, Brazil, during the COVID-19 outbreak; and to verify the relationship between satellite-based data [Tropospheric NO2 column density and Aerosol Optical Depth (AOD)] and ground-based concentrations [NO2 and particulate material (PM; coarse: PM10 and fine: PM2.5)]. For this purpose, tropospheric NO2 obtained from the TROPOMI sensor and AOD retrieved from MODIS sensor data by using the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm were compared with concentrations obtained from 50 automatic ground monitoring stations. The results showed low correlations between PM and AOD. For PM10, most stations showed correlations lower than 0.2, which were not significant. The results for PM2.5 were similar, but some stations showed good correlations for specific periods (before or during the COVID-19 outbreak). Satellite-based Tropospheric NO2 proved to be a good predictor for NO2 concentrations at ground level. Considering all stations with NO2 measurements, correlations >0.6 were observed, reaching 0.8 for specific stations and periods. In general, it was observed that regions with a more industrialized profile had the best correlations, in contrast with rural areas. In addition, it was observed about 57% reductions in tropospheric NO2 throughout the state of São Paulo during the COVID-19 outbreak. Variations in air pollutants were linked to the region economic vocation, since there were reductions in industrialized areas (at least 50% of the industrialized areas showed >20% decrease in NO2) and increases in areas with farming and livestock characteristics (about 70% of those areas showed increase in NO2). Our results demonstrate that Tropospheric NO2 column densities can serve as good predictors of NO2 concentrations at ground level. For MAIAC-AOD, a weak relationship was observed, requiring the evaluation of other possible predictors to describe the relationship with PM. Thus, it is concluded that regionalized assessment of satellite data accuracy is essential for assertive estimates on a regional/local level. Good quality information retrieved at specific polluted areas does not assure a worldwide use of remote sensor data.

Accounting for the aerosol type and additional satellite-borne aerosol products improves the prediction of PM2.5 concentrations.

Fine airborne particles (diameter <2.5 µm; PM2.5) are recognized as a major threat to human health due to their physicochemical properties: composition, size, shape, etc. However, normally only size-fraction-specific particle concentrations are monitored. Interestingly, although the aerosol type is reported as part of the aerosol optical depth retrieval from satellite observations, it has not been utilized, to date, as an auxiliary information/co-variate for PM2.5 prediction. We developed Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) models that account for this information when predicting surface PM2.5. The models take as input only widely available data: satellite aerosol products with full cover and surface meteorological data. Distinct models were developed for AOD of specific aerosol types. Both the RF and XGBoost models performed well, showing moderate-to-high cross-validated adjusted R2 (RF: 0.753-0.909; XGBoost: 0.741-0.903), depending on the aerosol type and other covariates. The weighted performance of the specific aerosol-type models was higher than of the RF and XGBoost baseline models, where all the AOD retrievals were used together (the common practice). Our approach can provide improved risk estimates due to exposure to PM2.5, better resolved radiative forcing calculations, and tailored abatement surveillance of specific pollutants/sources.

Deterioration of air quality associated with the 2020 US wildfires.

The wildfires of August and September 2020 in the western part of the United States were characterized by an unparalleled duration and wide geographical coverage. A particular consequence of massive wildfires includes serious health effects due to short and long-term exposure to poor air quality. Using a variety of data sources including aerosol optical depth (AOD) and ultraviolet aerosol index (UVAI), obtained with the Moderate-Resolution Imaging Spectroradiometer (MODIS), Multi-Angle Implementation of Atmospheric Correction (MAIAC) and Tropospheric Monitoring Instrument (TROPOMI), combined with meteorological information from the European Center for Medium-Range Weather Forecasts (ECMWF) and other supporting data, the impact of wildfires on air quality is examined in the three western US states, California, Oregon, and Washington, and areas to the east. The results show that smoke aerosols not only led to a significant deterioration in air quality in these states but also affected all other states, Canada, and surrounding ocean areas. The wildfires increased the average daily surface concentration of PM2.5 posing significant health risks, especially for vulnerable populations. Large amounts of black carbon (BC) aerosols were emitted into the atmosphere. AOD and UVAI exceeded 1 and 2 over most of the country. In parts of the three western states, those values reached 3.7 and 6.6, respectively. Moreover, a reanalysis based on MERRA-2 (Modern-Era Retrospective Analysis for Research and Applications, version 2) showed that the maximum values of BC surface mass concentration during the wildfires were about 370 µg/m3. These various indicators provide a better understanding of the extent of environmental and atmospheric degradation associated with these forest fires.

Characteristics of the severe March 2021 Gobi Desert dust storm and its impact on air pollution in China.

A dust storm that formed in the north of China and the southeastern part of Mongolia in March 2021 significantly deteriorated air quality over a large area of East Asia. According to the synoptic pattern, the cause of the dust storm was a cyclone with a significant drop in pressure leading to high winds and dry components of the soil over parts of the Gobi Desert becoming airborne. Data obtained from ground-based air quality monitoring stations show that the observed hourly PM10 concentration greatly exceeded the recommended maximum of 150 µg/m3 with readings above 1500 µg/m3 in the cities of Tianjin, Baoding, Zhengzhou, Luoyang, Zhoukou. In Shijiazhuang, Taiyuan, Jinnan, Xining, Baotou, and Jining. In Handan, it was over 2000 µg/m3. Cities where PM10 concentration exceeded 3900 µg/m3 included Lanzhou, Hohhot, Changzhou, Alashan, Yan'an, Yulin, Hami, Jiuquan, Heze, Hotan, and Baiyin. Concentrations exceeded 7000 µg/m3 on March 15th over parts of the provinces of Inner Mongolia, Gansu and Ningxia, in the cities of Ordos, Jinchang, Wuwei and Zhongwei. According to satellite data, the area of dust covered approximately 450,000 km2. MODIS and TROPOMI data demonstrated high aerosol optical depth (AOD) (more than 1) with a high ultraviolet aerosol index (UVAI) (more than 2), confirming the predominance of dust particles during the storm. Data from CALIPSO show the presence of a dense layer of dust extending from the earth's surface to a height of about 8 km. The Dust Regional Atmospheric Model (BSC-DREAM8b) demonstrates high synchrony with the satellite's surface dust concentration data, ranging from 640 to 1280 µg/m3, and exceeding 2650 µg/m3 in some areas. The purpose of this study is to analyze data from ground-based sensors, satellites, and atmospheric models to better understand the March 2021 dust storm event. The results may be useful for the implementation of protective and preventive measures for both the environment and human health, including air quality control.

A machine learning model to estimate ambient PM2.5 concentrations in industrialized highveld region of South Africa.

Exposure to fine particulate matter (PM2.5) has been linked to a substantial disease burden globally, yet little has been done to estimate the population health risks of PM2.5 in South Africa due to the lack of high-resolution PM2.5 exposure estimates. We developed a random forest model to estimate daily PM2.5 concentrations at 1 km2 resolution in and around industrialized Gauteng Province, South Africa, by combining satellite aerosol optical depth (AOD), meteorology, land use, and socioeconomic data. We then compared PM2.5 concentrations in the study domain before and after the implementation of the new national air quality standards. We aimed to test whether machine learning models are suitable for regions with sparse ground observations such as South Africa and which predictors played important roles in PM2.5 modeling. The cross-validation R2 and Root Mean Square Error of our model was 0.80 and 9.40 µg/m3, respectively. Satellite AOD, seasonal indicator, total precipitation, and population were among the most important predictors. Model-estimated PM2.5 levels successfully captured the temporal pattern recorded by ground observations. Spatially, the highest annual PM2.5 concentration appeared in central and northern Gauteng, including northern Johannesburg and the city of Tshwane. Since the 2016 changes in national PM2.5 standards, PM2.5 concentrations have decreased in most of our study region, although levels in Johannesburg and its surrounding areas have remained relatively constant. This is anadvanced PM2.5 model for South Africa with high prediction accuracy at the daily level and at a relatively high spatial resolution. Our study provided a reference for predictor selection, and our results can be used for a variety of purposes, including epidemiological research, burden of disease assessments, and policy evaluation.

Estimating the Impact of COVID-19 on the PM2.5 Levels in China with a Satellite-Driven Machine Learning Model.

China implemented an aggressive nationwide lockdown procedure immediately after the COVID-19 outbreak in January 2020. As China emerges from the impact of COVID-19 on national economic and industrial activities, it has become the site of a large-scale natural experiment to evaluate the impact of COVID-19 on regional air quality. However, ground measurements of fine particulate matters (PM2.5) concentrations do not offer comprehensive spatial coverage, especially in suburban and rural regions. In this study, we developed a machine learning method with satellite aerosol remote sensing data, meteorological fields and land use parameters as major predictor variables to estimate spatiotemporally resolved daily PM2.5 concentrations in China. Our study period consists of a reference semester (1 November 2018-30 April 2019) and a pandemic semester (1 November 2019-30 April 2020), with six modeling months in each semester. Each period was then divided into subperiod 1 (November and December), subperiod 2 (January and February) and subperiod 3 (March and April). The reference semester model obtained a 10-fold cross-validated R2 (RMSE) of 0.79 (17.55 µg/m3) and the pandemic semester model obtained a 10-fold cross-validated R2 (RMSE) of 0.83 (13.48 µg/m3) for daily PM2.5 predictions. Our prediction results showed high PM2.5 concentrations in the North China Plain, Yangtze River Delta, Sichuan Basin and Xinjiang Autonomous Region during the reference semester. PM2.5 levels were lowered by 4.8 µg/m3 during the pandemic semester compared to the reference semester and PM2.5 levels during subperiod 2 decreased most, by 18%. The southeast region was affected most by the COVID-19 outbreak with PM2.5 levels during subperiod 2 decreasing by 31%, followed by the Northern Yangtze River Delta (29%) and Pearl River Delta (24%).

[High-resolution Estimation of Spatio-temporal Variation in PM2.5 Concentrations in the Beijing-Tianjin-Hebei Region].

This study developed a two-stage statistical model (linear mixed effect (LME) model+geographical weight regression (GWR) model) to determine the spatio-temporal variation of PM2.5 concentrations in the Beijing-Tianjin-Hebei (BTH) region with full-coverage, high resolution, and high accuracy. The model employs multi-angle implementation of atmospheric correction aerosol optical depth (MAIAC AOD) data, with a 1 km spatial resolution, as the main predictor and meteorological data/land-use data as the auxiliary predictors. To determine the characteristics of heavy PM2.5 pollution in the BTH region, unique predictors such as AOD2 were also introduced into the two-stage model. The two-stage model was used to estimate the daily PM2.5 concentrations with a 1 km resolution. After being cross-validated against ground observations, the R2 of PM2.5 was found to be 0.94, with a slope value of 0.95 and RMSPE value of 13.14 µg·m-3. Compared to previous studies such as LME, the two-stage model has much higher accuracy, suitable for estimating PM2.5 concentrations. The PM2.5 concentration in the BTH region ranged from 0 to 89.89 µg·m-3 in 2017, with a mean value of 44.96 µg·m-3. The spatio-temporal variability of PM2.5 over the BTH region was significant, exhibiting high values over the southwestern plain, moderate values over the northeastern plain, and low values over the mountainous plateau. In terms of seasonal variation, PM2.5 concentrations were high in winter, low in summer, and moderate in spring and autumn. The estimated PM2.5 concentrations, with high spatio-temporal resolution, are useful for exposure assessments in epidemiological studies and identifying the spatio-temporal variation of pollution sources at a fine spatial scale. The results show that the locations of vital pollution sources over the severely polluted south-central Hebei piedmont plain may have changed since the implementation of the Air Pollution Prevention and Control Action Plan. This study could provide a scientific basis for the prevention and control of air pollution in the BHT region.

[MAIAC AOD and PM2.5 Mass Concentrations Characteristics and Correlation Analysis in Beijing-Tianjin-Hebei and Surrounding Areas].

Based on the MAIAC AOD and PM2.5 mass concentration data for the Beijing-Tianjin-Hebei region and surrounding areas from 2014 to 2018, the temporal and spatial differences in Aerosol Optical Depth (AOD) and PM2.5 mass concentrations were explored and their correlation was analyzed by linear regression. The results show that the daily average concentration of PM2.5 exceeds the standard for 33% and 57% of measurements based on the daily average standard values of the World Health Organization IT.1 and IT.2, respectively, indicating serious pollution levels. The annual average concentrations of PM2.5 and Terra and Aqua MAIAC AOD all show downward trends. The PM2.5 concentrations are high in winter and spring and low in summer and autumn; Terra and Aqua AOD values are high in spring and summer and low in autumn and winter. The seasonal and annual average concentrations of PM2.5 and AOD all show the regional pattern of "low in the north and high in the south". High-value areas are mainly located in southern Hebei, southwestern Shanxi, western Shandong, and northern Henan, while low-value areas are mainly located in northwestern Shanxi, northern Hebei, and eastern Shandong. The annual average concentration of PM2.5 is between 27 and 99µg·m-3, and the annual average AOD is between 0.20 and 0.69. The correlation between Aqua AOD and PM2.5 concentration is strong whereas and the correlations between Terra AOD, Aqua AOD, and PM2.5vary significantly in different seasons; overall, correlations are strongest in spring and winter and weakest in summer and autumn. After vertical-humidity correction, the correlation between satellite AOD and PM2.5 data is significantly improved.

Prediction of PM2.5 concentrations at unsampled points using multiscale geographically and temporally weighted regression.

Numerous statistical models have established the relationship between ambient fine particulate matter (PM2.5, with an aerodynamic diameter of less than 2.5 µm) and satellite aerosol optical depth (AOD) along with other meteorological/land-related covariates. However, all the models assumed that all covariates affect the PM2.5 concentration at the same scale, and none could provide a posterior uncertainty analysis at each regression point. Therefore, a multiscale geographically and temporally weighted regression (MGTWR) model was proposed by specifying a unique bandwidth for each covariate. However, the lack of a method for predicting values at unsampled points in the MGTWR model greatly restricts its corresponding application. Thus, this study developed a method for inferring unsampled points and used the posterior uncertainty assessment value to improve the model accuracy. With the aid of the high-resolution satellite multi-angle implementation of atmospheric correction (MAIAC) AOD product, daily PM2.5 concentrations with a 1 km × 1 km resolution were generated over the Beijing-Tianjin-Hebei region between 2013 and 2019. The coefficient of determination (R2) and root mean square error (RMSE) of the fitted MGTWR results vary from 0.90 to 0.94 and from 10.66 to 25.11 µg/m3, respectively. The sample-based and site-based cross-validation R2 and RMSE vary from 0.81 to 0.89 and from 14.40 to 34.43 µg/m3 respectively, demonstrating the effectiveness of the proposed inference method at unsampled points. With the uncertainty constraint, the sample-based and site-based validated MGTWR R2 results for all years are further improved by approximately 0.02-0.04, demonstrating the effectiveness of the posterior uncertainty assessment constraint method. These results suggest that the inference method proposed in this study is promising to overcome the defects of the MGTWR model in inferring the prediction values at unsampled points and could consequently enhance the wide applications of MGTWR modeling.

Season, not lockdown, improved air quality during COVID-19 State of Emergency in Nigeria.

Globally, ambient air pollution claims ~9 million lives yearly, prompting researchers to investigate changes in air quality. Of special interest is the impact of COVID-19 lockdown. Many studies reported substantial improvements in air quality during lockdowns compared with pre-lockdown or as compared with baseline values. Since the lockdown period coincided with the onset of the rainy season in some tropical countries such as Nigeria, it is unclear if such improvements can be fully attributed to the lockdown. We investigate whether significant changes in air quality in Nigeria occurred primarily due to statewide COVID-19 lockdown. We applied a neural network approach to derive monthly average ground-level fine aerosol optical depth (AODf) across Nigeria from year 2001-2020, using the Multi-angle Implementation of Atmospheric Correction (MAIAC) AODs from Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) satellites, AERONET aerosol optical properties, meteorological and spatial parameters. During the year 2020, we found a 21% or 26% decline in average AODf level across Nigeria during lockdown (April) as compared to pre-lockdown (March), or during the easing phase-1 (May) as compared to lockdown, respectively. Throughout the 20-year period, AODf levels were highest in January and lowest in May or June, but not April. Comparison of AODf levels between 2020 and 2019 shows a small decline (1%) in pollution level in April of 2020 compare to 2019. Using a linear time-lag model to compare changes in AODf levels for similar months from 2002 to 2020, we found no significant difference (Levene's test and ANCOVA; α = 0.05) in the pollution levels by year, which indicates that the lockdown did not significantly improve air quality in Nigeria. Impact analysis using multiple linear regression revealed that favorable meteorological conditions due to seasonal change in temperature, relative humidity, planetary boundary layer height, wind speed and rainfall improved air quality during the lockdown.

Retrospective assessment of pregnancy exposure to particulate matter from desert dust on a Caribbean island: could satellite-based aerosol optical thickness be used as an alternative to ground PM10 concentration?

Desert dust transported from the Saharan-Sahel region to the Caribbean Sea is responsible for peak exposures of particulate matter (PM). This study explored the potential added value of satellite aerosol optical thickness (AOT) measurements, compared to the PM concentration at ground level, to retrospectively assess exposure during pregnancy. MAIAC MODIS AOT retrievals in blue band (AOT470) were extracted for the French Guadeloupe archipelago. AOT470 values and PM10 concentrations were averaged over pregnancy for 906 women (2005-2008). Regression modeling was used to examine the AOT470-PM10 relationship during pregnancy and test the association between dust exposure estimates and preterm birth. Moderate agreement was shown between mean AOT470 retrievals and PM10 ground-based measurements during pregnancy (R2 = 0.289). The magnitude of the association between desert dust exposure and preterm birth tended to be lower using the satellite method compared to the monitor method. The latter remains an acceptable trade-off between epidemiological relevance and exposure misclassification, in areas with few monitoring stations and complex topographical/meteorological conditions, such as tropical islands.

Estimating hourly PM2.5 concentrations in Beijing with satellite aerosol optical depth and a random forest approach.

Assessing short-term exposure to PM2.5 requires the concentration distribution at a high spatiotemporal resolution. Abundant researches have derived the daily predictions of fine particles, but estimating hourly PM2.5 is still a challenge restrained by the input data. The recent aerosol optical depth (AOD) product from Himawari-8 provides hourly satellite observations informative to modelling. In this study, we developed separate random forest models with and without AOD and combined the estimates to obtain a full-coverage hourly PM2.5 distribution. 10-fold cross validation R2 ranged from 0.92 to 0.95 and root mean square errors from 14.1 to 16.9 µg/m3, indicating the good model performance. Spatial convolutional layers of PM2.5 measurements and temporal accumulation effects of meteorological features were added into the model. They turned out to be of the most important predictors and improved the performance significantly. Finally, we mapped hourly PM2.5 at a 1-km resolution in Beijing during a pollution episode in 2019 and studied the pollution pattern. The study proposed a method to obtain 24-h full-coverage hourly PM2.5 estimates which are useful for acute exposure assessment in epidemiological researches.

Advancing methodologies for applying machine learning and evaluating spatiotemporal models of fine particulate matter (PM2.5) using satellite data over large regions.

Reconstructing the distribution of fine particulate matter (PM2.5) in space and time, even far from ground monitoring sites, is an important exposure science contribution to epidemiologic analyses of PM2.5 health impacts. Flexible statistical methods for prediction have demonstrated the integration of satellite observations with other predictors, yet these algorithms are susceptible to overfitting the spatiotemporal structure of the training datasets. We present a new approach for predicting PM2.5 using machine-learning methods and evaluating prediction models for the goal of making predictions where they were not previously available. We apply extreme gradient boosting (XGBoost) modeling to predict daily PM2.5 on a 1×1 km2 resolution for a 13 state region in the Northeastern USA for the years 2000-2015 using satellite-derived aerosol optical depth and implement a recursive feature selection to develop a parsimonious model. We demonstrate excellent predictions of withheld observations but also contrast an RMSE of 3.11 µg/m3 in our spatial cross-validation withholding nearby sites versus an overfit RMSE of 2.10 µg/m3 using a more conventional random ten-fold splitting of the dataset. As the field of exposure science moves forward with the use of advanced machine-learning approaches for spatiotemporal modeling of air pollutants, our results show the importance of addressing data leakage in training, overfitting to spatiotemporal structure, and the impact of the predominance of ground monitoring sites in dense urban sub-networks on model evaluation. The strengths of our resultant modeling approach for exposure in epidemiologic studies of PM2.5 include improved efficiency, parsimony, and interpretability with robust validation while still accommodating complex spatiotemporal relationships.

Weight-of-evidence approach to identify regionally representative sites for air-quality monitoring network: Satellite data-based analysis.

The methodology discussed in Lekinwala et al., 2020, hereinafter referred to as the 'parent article', is used to setup a nation-wide network for background PM2.5 measurement at strategic locations, optimally placing sites to obtain maximum regionally representative PM2.5 concentrations with minimum number of sites. Traditionally, in-situ PM2.5 measurements are obtained for several potential sites and compared to identify the most regionally representative sites [4], Wongphatarakul et al., 1998) at the location. The 'parent article' proposes the use of satellite-derived proxy for aerosol (Aerosol Optical Depth, AOD) data in the absence of in-situ PM2.5 measurements. This article focuses on the details about satellite-data processing which forms part of the methodology discussed in the 'parent article'. Following are some relevant aspects:•High resolution AOD is retrieved from Moderate Resolution Imaging Spectroradiometer (MODIS) instruments aboard NASA's Aqua and Terra satellite using Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm. The data is stored as grids of size 1200  ×  1200 and a total of seven such grids cover the Indian land mass. These grids were merged, regridded and multiplied by conversion factors from GEOS-Chem Chemical Transport Model to obtain PM2.5 values. Standard set of tools like CDO and NCL are used to manipulate the satellite-data (*.nc files).•The PM2.5 values are subjected to various statistical analysis using metrics like coefficient of divergence (CoD), Pearson correlation coefficient (PCC) and mutual information (MI).•Computations for CoD, MI are performed using Python codes developed in-house while a function in NumPy module of Python was used for PCC calculations.

Spatiotemporal Imputation of MAIAC AOD Using Deep Learning with Downscaling.

Aerosols have adverse health effects and play a significant role in the climate as well. The Multiangle Implementation of Atmospheric Correction (MAIAC) provides Aerosol Optical Depth (AOD) at high temporal (daily) and spatial (1 km) resolution, making it particularly useful to infer and characterize spatiotemporal variability of aerosols at a fine spatial scale for exposure assessment and health studies. However, clouds and conditions of high surface reflectance result in a significant proportion of missing MAIAC AOD. To fill these gaps, we present an imputation approach using deep learning with downscaling. Using a baseline autoencoder, we leverage residual connections in deep neural networks to boost learning and parameter sharing to reduce overfitting, and conduct bagging to reduce error variance in the imputations. Downscaled through a similar auto-encoder based deep residual network, Modern-Era Retrospective analysis for Research and Applications Version 2 (MERRA-2) GMI Replay Simulation (M2GMI) data were introduced to the network as an important gap-filling feature that varies in space to be used for missingness imputations. Imputing weekly MAIAC AOD from 2000 to 2016 over California, a state with considerable geographic heterogeneity, our full (non-full) residual network achieved mean R2 = 0.94 (0.86) [RMSE = 0.007 (0.01)] in an independent test, showing considerably better performance than a regular neural network or non-linear generalized additive model (mean R2 = 0.78-0.81; mean RMSE = 0.013-0.015). The adjusted imputed as well as combined imputed and observed MAIAC AOD showed strong correlation with Aerosol Robotic Network (AERONET) AOD (R = 0.83; R2 = 0.69, RMSE = 0.04). Our results show that we can generate reliable imputations of missing AOD through a deep learning approach, having important downstream air quality modeling applications.

Developing an Advanced PM2.5 Exposure Model in Lima, Peru.

It is well recognized that exposure to fine particulate matter (PM2.5) affects health adversely, yet few studies from South America have documented such associations due to the sparsity of PM2.5 measurements. Lima's topography and aging vehicular fleet results in severe air pollution with limited amounts of monitors to effectively quantify PM2.5 levels for epidemiologic studies. We developed an advanced machine learning model to estimate daily PM2.5 concentrations at a 1 km2 spatial resolution in Lima, Peru from 2010 to 2016. We combined aerosol optical depth (AOD), meteorological fields from the European Centre for Medium-Range Weather Forecasts (ECMWF), parameters from the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem), and land use variables to fit a random forest model against ground measurements from 16 monitoring stations. Overall cross-validation R2 (and root mean square prediction error, RMSE) for the random forest model was 0.70 (5.97 µg/m3). Mean PM2.5 for ground measurements was 24.7 µg/m3 while mean estimated PM2.5 was 24.9 µg/m3 in the cross-validation dataset. The mean difference between ground and predicted measurements was -0.09 µg/m3 (Std.Dev. = 5.97 µg/m3), with 94.5% of observations falling within 2 standard deviations of the difference indicating good agreement between ground measurements and predicted estimates. Surface downwards solar radiation, temperature, relative humidity, and AOD were the most important predictors, while percent urbanization, albedo, and cloud fraction were the least important predictors. Comparison of monthly mean measurements between ground and predicted PM2.5 shows good precision and accuracy from our model. Furthermore, mean annual maps of PM2.5 show consistent lower concentrations in the coast and higher concentrations in the mountains, resulting from prevailing coastal winds blown from the Pacific Ocean in the west. Our model allows for construction of long-term historical daily PM2.5 measurements at 1 km2 spatial resolution to support future epidemiological studies.

Impacts of snow and cloud covers on satellite-derived PM2.5 levels.

Satellite aerosol optical depth (AOD) has been widely employed to evaluate ground fine particle (PM2.5) levels, whereas snow/cloud covers often lead to a large proportion of non-random missing AOD values. As a result, the fully covered and unbiased PM2.5 estimates will be hard to generate. Among the current approaches to deal with the data gap issue, few have considered the cloud-AOD relationship and none of them have considered the snow-AOD relationship. This study examined the impacts of snow and cloud covers on AOD and PM2.5 and made full- coverage PM2.5 predictions by considering these impacts. To estimate missing AOD values, daily gap-filling models with snow/cloud fractions and meteorological covariates were developed using the random forest algorithm. By using these models in New York State, a daily AOD data set with a 1-km resolution was generated with a complete coverage. The "out-of-bag" R2 of the gap-filling models averaged 0.93 with an interquartile range from 0.90 to 0.95. Subsequently, a random forest-based PM2.5 prediction model with the gap-filled AOD and covariates was built to predict fully covered PM2.5 estimates. A ten-fold cross-validation for the prediction model showed a good performance with an R2 of 0.82. In the gap-filling models, the snow fraction was of higher significance to the snow season compared with the rest of the year. The prediction models fitted with/without the snow fraction also suggested the discernible changes in PM2.5 patterns, further confirming the significance of this parameter. Compared with the methods without considering snow and cloud covers, our PM2.5 prediction surfaces showed more spatial details and reflected small-scale terrain-driven PM2.5 patterns. The proposed methods can be generalized to the areas with extensive snow/cloud covers and large proportions of missing satellite AOD data for predicting PM2.5 levels with high resolutions and complete coverage.