Spatiotemporal land use random forest model for estimating metropolitan NO2 exposure in Japan

Do Machine Learning Methods Improve Prediction of Ambient Air Pollutants with High Spatial Contrast? A Systematic Review

BACKGROUND & OBJECTIVE

The use of machine learning for air pollution modelling is rapidly increasing. We conducted a systematic review of studies comparing statistical and machine learning models predicting the spatiotemporal variation of ambient nitrogen dioxide (NO2), ultrafine particles (UFPs) and black carbon (BC) to determine whether and in which scenarios machine learning generates more accurate predictions.

METHODS

Web of Science and Scopus were searched up to June 13, 2024. All records were screened by two independent reviewers. Differences in the coefficient of determination (R2) and Root Mean Square Error (RMSE) between best statistical and machine learning methods were compared across categories of methodological elements.

RESULTS

A total of 38 studies with 46 model comparisons (30 for NO2, 8 for UFPs and 8 for BC) were included. Linear non-regularized methods and Random Forest were most frequently used. Machine learning outperformed statistical models in 34 comparisons. Mean differences (95% confidence intervals) in R2 and RMSE between best machine learning and statistical models were 0.12 (0.08, 0.17) and 20% (11%, 29%) respectively. Tree-based methods performed best in 12 of 17 multi-model comparisons. Nonlinear or regularization regression methods were used in only 12 comparisons and provided similar performance to machine learning methods.

CONCLUSION

This systematic review suggests that machine learning methods, especially tree-based methods, may be superior to linear non-regularized methods for predicting ambient concentrations of NO2, UFPs and BC. Additional comparison studies using nonlinear, regularized and a wider array of machine learning methods are needed to confirm their relative performance. Future air pollution studies would also benefit from more explicit and standardized reporting of methodologies and results.

Long-term spatiotemporal variation of benzo[a]pyrene in Japan: Significant decrease in ambient concentrations, human exposure, and health risk

Although Benzo[a]pyrene (BaP) is considered carcinogenic to humans, the health effects of exposure to ambient levels have not been sufficiently investigated. This study estimated the long-term spatiotemporal variation of BaP in Japan over nearly two decades at a fine spatial resolution of 1 km. This study aimed to obtain an accurate spatiotemporal distribution of BaP that can be used in epidemiological studies on the health effects of ambient BaP exposure. The annual BaP concentrations were estimated using an ensemble machine learning approach using various predictors, including the concentrations and emission intensities of the criteria air pollutants, and meteorological, land use, and traffic-related variables. The model performance, evaluated by location-based cross-validation, exhibited satisfactory accuracy (R2 of 0.693). Densely populated areas showed higher BaP levels and greater temporal reduction, whereas BaP levels remained higher in some industrial areas. The population-weighted BaP in 2018 was 0.12 ng m-3, a decrease of approximately 70% from its 2000 value of 0.44 ng m-3, which was also reflected in the estimated excess number of lung cancer incidences. Accordingly, the proportion of BaP exposure below 0.12 ng m-3, which is the BaP concentration associated with an excess lifetime cancer risk of 10-5, reached 67% in 2018. Our estimates can be used in epidemiological studies to assess the health effects of BaP exposure at ambient concentrations.

Capturing Exposure Disparities with Chemical Transport Models: Evaluating the Suitability of Downscaling Using Land Use Regression

High resolution exposure surfaces are essential to capture disparities in exposure to traffic-related air pollution in urban areas. In this study, we develop an approach to downscale Chemical Transport Model (CTM) simulations to a hyperlocal level (∼100m) in the Greater Toronto Area (GTA) under three scenarios where emissions from cars, trucks and buses are zeroed out, thus capturing the burden of each transportation mode. This proposed approach statistically fuses CTMs with Land-Use Regression using machine learning techniques. With this proposed downscaling approach, changes in air pollutant concentrations under different scenarios are appropriately captured by downscaling factors that are trained to reflect the spatial distribution of emission reductions. Our validation analysis shows that high-resolution models resulted in better performance than coarse models when compared with observations at reference stations. We used this downscaling approach to assess disparities in exposure to nitrogen dioxide (NO2) for populations composed of renters, low-income households, recent immigrants, and visible minorities. Individuals in all four categories were disproportionately exposed to the burden of cars, trucks, and buses. We conducted this analysis at spatial resolutions of 12, 4, 1 km, and 100 m and observed that disparities were significantly underestimated when using coarse spatial resolutions. This reinforces the need for high-spatial resolution exposure surfaces for environmental justice analyses.

Mixed effects of ambient air pollutants on oocyte-related outcomes: A novel insight from women undergoing assisted reproductive technology

Air pollution is widely acknowledged as a significant risk factor for human health, especially reproductive health. Nevertheless, many studies have disregarded the potentially mixed effects of air pollutants on reproductive outcomes. We performed a retrospective cohort study involving 8048 women with 9445 cycles undergoing In Vitro Fertilization (IVF) and Intracytoplasmic Sperm Injection (ICSI) in China, from 2017 to 2021. A land-use random forest model was applied to estimate daily residential exposure to air pollutants, including sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), ozone (O3), and fine particulate matter (PM2.5). Individual and joint associations between air pollutants and oocyte-related outcomes of ART were evaluated. In 90 days prior to oocyte pick-up to oocyte pick-up (period A), NO2, O3 and CO was negatively associated with total oocyte yield. In the 90 days prior to oocyte pick-up to start of gonadotropin medication (Gn start, period B), there was a negative dose-dependent association of exposure to five air pollutants with total oocyte yield and mature oocyte yield. In Qgcomp analysis, increasing the multiple air pollutants mixtures by one quartile was related to reducing the number of oocyte pick-ups by -2.00 % (95 %CI: -2.78 %, -1.22 %) in period A, -2.62 % (95 %CI: -3.40 %, -1.84 %) in period B, and -0.98 % (95 %CI: -1.75 %, -0.21 %) in period C. During period B, a 1-unit increase in the WQS index of multiple air pollutants exposure was associated with fewer number of total oocyte (-1.27 %, 95 %CI: -2.16 %, -0.36 %) and mature oocyte (-1.42 %, 95 %CI: -2.41 %, -0.43 %). O3 and NO2 were major contributors with adverse effects on the mixed associations. Additionally, period B appears to be the susceptible window. Our study implies that exposure to air pollution adversely affects oocyte-related outcomes, which raises concerns about the potential adverse impact of air pollution on women's reproductive health.

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

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

Estimate of near-surface NO2 concentrations in Fenwei Plain, China, based on TROPOMI data and random forest model

Nitrogen dioxide (NO2) concentration is a crucial indicator of ground-level air quality, and elevated concentrations can adversely affect human health and the atmospheric environment. In this study, we utilized Tropospheric Monitoring Instrument (TROPOMI) tropospheric NO2 vertical column density data (VCD) and multi-source geographic data to establish a random forest regression (RF) model that accurately estimates NO2 concentrations near the ground in the Fenwei Plain. The model addresses the inherent limitations of traditional ground-based monitoring and provides data support for analyzing regional pollution spatial and temporal characteristics. (1) The RF model based on TROPOMI and geographic data demonstrates high estimation accuracy, with monthly average RF model fit and validation coefficient of determination (R2) reaching 0.949 and 0.875, respectively. (2) A complex nonlinear relationship exists between near-surface NO2 concentration and multi-source geographic data. The RF model's estimations reveal clear seasonal and regional variations in near-surface NO2 concentration. Concentrations are generally highest in winter, followed by spring and autumn, and lowest in summer. The high NO2 concentrations are primarily mainly distributed in the plains and river valleys with low elevation and dense population density. The model estimation results also indicate that the estimated effect is better when the NO2 concentration fluctuates less and anthropogenic emission reduction measures significantly impact the NO2 concentration near the ground. (3) The population exposure risk results indicate that most cities in the Fenwei Plain face varying exposure risks. These findings offer valuable insights for regional NO2 pollution management.

Assessing the Spatiotemporal Characteristics, Factor Importance, and Health Impacts of Air Pollution in Seoul by Integrating Machine Learning into Land-Use Regression Modeling at High Spatiotemporal Resolutions

Previous studies have characterized spatial patterns of air pollution with land-use regression (LUR) models. However, the spatiotemporal characteristics of air pollution, the contribution of various factors to them, and the resultant health impacts have yet to be evaluated comprehensively. This study integrates machine learning (random forest) into LUR modeling (LURF) with intensive evaluations to develop high spatiotemporal resolution prediction models to estimate daily and diurnal PM2.5 and NO2 in Seoul, South Korea, at the spatial resolution of 500 m for a year (2019) and to then evaluate the contribution of driving factors and quantify the resultant premature mortality. Our results show that incorporating the random forest algorithm into our LUR model improves the model performance. Meteorological conditions have a great influence on daily models, while land-use factors play important roles in diurnal models. Our health assessment using dynamic population data estimates that PM2.5 and NO2 pollution, when combined, causes a total of 11,183 (95% CI: 5837-16,354) premature mortalities in Seoul in 2019, of which 64.9% are due to PM2.5, while the remaining are attributable to NO2. The air pollution-attributable health impacts in Seoul are largely caused by cardiovascular diseases including stroke. This study pinpoints the significant spatiotemporal variations and health impact of PM2.5 and NO2 in Seoul, providing essential data for epidemiological research and air quality management.

Assessment of NO2 population exposure from 2005 to 2020 in China

Nitrogen dioxide (NO₂) is a major air pollutant with serious environmental and human health impacts. A random forest model was developed to estimate ground-level NO₂ concentrations in China at a monthly time scale based on ground-level observed NO₂ concentrations, tropospheric NO₂ column concentration data from the Ozone Monitoring Instrument (OMI), and meteorological covariates (the MAE, RMSE, and R² of the model were 4.16 µg/m³, 5.79 µg/m³, and 0.79, respectively, and the MAE, RMSE, and R² of the cross-validation were 4.3 µg/m³, 5.82 µg/m³, and 0.77, respectively). On this basis, this article analyzed the spatial and temporal variation in NO₂ population exposure in China from 2005 to 2020, which effectively filled the gap in the long-term NO₂ population exposure assessment in China. NO₂ population exposure over China has significant spatial aggregation, with high values mainly distributed in large urban clusters in the north, east, south, and provincial capitals in the west. The NO₂ population exposure in China shows a continuous increasing trend before 2012 and a continuous decreasing trend after 2012. The change in NO₂ population exposure in western and southern cities is more influenced by population density compared to northern cities. NO₂ pollution in China has substantially improved from 2013 to 2020, but Urumqi, Lanzhou, and Chengdu still maintain high NO₂ population exposure. In these cities, the Environmental Protection Agency (EPA) could reduce NO₂ population exposure through more monitoring instruments and limiting factory emissions.

Random forest algorithm-based classification model of pesticide aquatic toxicity to fishes

Aquatic toxicity of pesticides can result in poisoning of many non-target organisms, of which various fishes are the most prominent one. It is a challenge to predict the toxicity (LC₅₀) classes of organic pesticides to various fish species from global QSAR models with a larger applicability domain. In this paper, by applying the random forest (RF) algorithm for a two-class problem, only eight molecular descriptors were used to develop a quantitative structure-activity relationship (QSAR) model for 1106 toxicity data (96 h, LC₅₀) of organic pesticides to various fish species including Oncorhynchus mykiss, Lepomis macrochirus, Pimephales promelas, Brachydanio rerio, Cyprinodon, Cyprinus carpio, etc. By the prediction of the optimal RF Model I (ntree =280, mtry = 3 and nodesize = 5), the training set (885 organic pesticides) has the prediction accuracies of 99.6% for Class 1 (LC₅₀ ≤ 10) and 96.7% for Class 2 (LC₅₀ > 10); the test set (221 organic pesticides) has the accuracies being 90.8% for Class 1 and 91.2% for Class 2. The optimal RF Model I is satisfactory compared with other QSAR model reported in the literature, although its descriptor subset is small.

Land use change and climate dynamics in the Rift Valley Lake Basin, Ethiopia

Land use and climate dynamics have a pronounced impact on water resources, biodiversity, land degradation, and productivity at all scales. Thus, in this study, we present the spatio-temporal dynamics of land use change and climate aiming to provide a scientific evidence about gains and losses in major land use categories and associated drivers and significancy and homogeneity of climate change. To this end, Landsat images and historical climate data have been used to determine the dynamics. In addition, population census data and land use policy have been considered to assess the potential drivers of land use change. The spatio-temporal land use dynamics have been evaluated using transition matrix and dynamics index. Likewise, shifts in the climate data were analyzed using change point analysis and three homogenous climate zones have been identified using principal component analysis. The results show that, from 1989 to 2019, the areal percentage of agricultural land increased by 27.5%, settlement by 0.8%, and barren land 0.4% while the natural vegetation, wetland, water body, and grass land decreased by 24.5%, 1.6%, 0.5%, and 2.1%, respectively. The land use dynamics have been stronger in the first decade of the study period. An abrupt shift of climate has occurred in the 1980s. In the last four decades, rainfall shows a not significant decreasing trend. However, a significant increasing trend has been observed for temperature. Rapid population growth, agricultural expansion policy, and climate variability have been identified as the underlying drivers of land use dynamics.

Fetal Exposure to Air Pollution in Late Pregnancy Significantly Increases ADHD-Risk Behavior in Early Childhood

BACKGROUND

Air pollution nowadays has seriously threatened the health of the Chinese population, especially in the vulnerable groups of fetuses, infants and toddlers. In particular, the effects of air pollution on children's neurobehavioral development have attracted widespread attention. Moreover, the early detection of a sensitive period is very important for the precise intervention of the disease. However, such studies focusing on hyperactive behaviors and susceptible window identification are currently lacking in China.

OBJECTIVES

The study aims to explore the correlation between air pollution exposure and hyperactive behaviors during the early life stage and attempt to identify whether a susceptible exposure window exists that is crucial for further precise intervention.

METHODS

Based on the Longhua Child Cohort Study, we collected the basic information and hyperactivity index of 26,052 children using a questionnaire conducted from 2015 to 2017, and the Conners' Parent Rating Scale-revised (CPRS-48) was used to assess hyperactive behaviors. Moreover, the data of air pollution concentration (PM₁₀, PM_2.5, NO₂, CO, O₃ and SO₂) were collected from the monitoring station between 2011 to 2017, and a land-use random forest model was used to evaluate the exposure level of each subject. Furthermore, Distributed lag non-linear models (DLNMs) were applied for statistic analysis.

RESULTS

The risk of child hyperactivity was found to be positively associated with early life exposure to PM₁₀, PM_2.5 and NO₂. In particular, for an increase of per 10 µg/m³ in PM₁₀, PM_2.5 and NO₂ exposure concentration during early life, the risk of child hyperactivity increased significantly during the seventh month of pregnancy to the fourth month after birth, with the strongest association in the ninth month of pregnancy (PM₁₀: OR = 1.043, 95% CI: 1.016-1.071; PM_2.5: OR = 1.062, 95% CI: 1.024-1.102; NO₂: OR = 1.043, 95% CI: 1.016-1.071). However, no significant associations among early life exposure to CO, O₃ and SO₂ and child hyperactive behaviors were observed.

CONCLUSIONS

Early life exposure to PM₁₀, PM_2.5 and NO₂ is associated with an increased risk of child ADHD-like behaviors at the age around 3 years, and the late-prenatal and early postnatal periods might be the susceptible exposure windows.

IoT and Satellite Sensor Data Integration for Assessment of Environmental Variables: A Case Study on NO2

This paper introduces a novel approach to increase the spatiotemporal resolution of an arbitrary environmental variable. This is achieved by utilizing machine learning algorithms to construct a satellite-like image at any given time moment, based on the measurements from IoT sensors. The target variables are calculated by an ensemble of regression models. The observed area is gridded, and partitioned into Voronoi cells based on the IoT sensors, whose measurements are available at the considered time. The pixels in each cell have a separate regression model, and take into account the measurements of the central and neighboring IoT sensors. The proposed approach was used to assess NO2 data, which were obtained from the Sentinel-5 Precursor satellite and IoT ground sensors. The approach was tested with three different machine learning algorithms: 1-nearest neighbor, linear regression and a feed-forward neural network. The highest accuracy yield was from the prediction models built with the feed-forward neural network, with an RMSE of 15.49 ×10-6 mol/m2.

Modelling and investigating the impacts of climatic variables on ozone concentration in Malaysia using correlation analysis with random forest, decision tree regression, linear regression, and support vector regression

Climate change is generally known to impact ozone concentration globally. However, the intensity varies across regions and countries. Therefore, local studies are essential to accurately assess the correlation of climate change and ozone concentration in different countries. This study investigates the effects of climatic variables on ozone concentration in Malaysia in order to understand the nexus between climate change and ozone concentration. The selected data was obtained from ten (10) air monitoring stations strategically mounted in urban-industrial and residential areas with significant emissions of pollutants. Correlation analysis and four machine learning algorithms (random forest, decision tree regression, linear regression, and support vector regression) were used to analyze ozone and meteorological dataset in the study area. The analysis was carried out during the southwest monsoon due to the rise of ozone in the dry season. The results show a very strong correlation between temperature and ozone. Wind speed also exhibits a moderate to strong correlation with ozone, while relative humidity is negatively correlated. The highest correlation values were obtained at Bukit Rambai, Nilai, Jaya II Perai, Ipoh, Klang and Petaling Jaya. These locations have high industries and are well urbanized. The four machine learning algorithms exhibit high predictive performances, generally ascertaining the predictive accuracy of the climatic variables. The random forest outperformed other algorithms with a very high R² of 0.970, low RMSE of 2.737 and MAE of 1.824, followed by linear regression, support vector regression and decision tree regression, respectively. This study's outcome indicates a linkage between temperature and wind speed with ozone concentration in the study area. An increase of these variables will likely increase the ozone concentration posing threats to lives and the environment. Therefore, this study provides data-driven insights for decision-makers and other stakeholders in ensuring good air quality for sustainable cities and communities. It also serves as a guide for the government for necessary climate actions to reduce the effect of climate change on air pollution and enabling sustainable cities in accordance with the UN's SDGs 13 and 11, respectively.

Fine-Grained Urban Air Quality Mapping from Sparse Mobile Air Pollution Measurements and Dense Traffic Density

Urban air quality mapping has been widely applied in urban planning, air pollution control and personal air pollution exposure assessment. Urban air quality maps are traditionally derived using measurements from fixed monitoring stations. Due to high cost, these stations are generally sparsely deployed in a few representative locations, leading to a highly generalized air quality map. In addition, urban air quality varies rapidly over short distances (<1 km) and is influenced by meteorological conditions, road network and traffic flow. These variations are not well represented in coarse-grained air quality maps generated by conventional fixed-site monitoring methods but have important implications for characterizing heterogeneous personal air pollution exposures and identifying localized air pollution hotspots. Therefore, fine-grained urban air quality mapping is indispensable. In this context, supplementary low-cost mobile sensors make mobile air quality monitoring a promising alternative. Using sparse air quality measurements collected by mobile sensors and various contextual factors, especially traffic flow, we propose a context-aware locally adapted deep forest (CLADF) model to infer the distribution of NO2 by 100 m and 1 h resolution for fine-grained air quality mapping. The CLADF model exploits deep forest to construct a local model for each cluster consisting of nearest neighbor measurements in contextual feature space, and considers traffic flow as an important contextual feature. Extensive validation experiments were conducted using mobile NO2 measurements collected by 17 postal vans equipped with low-cost sensors operating in Antwerp, Belgium. The experimental results demonstrate that the CLADF model achieves the lowest RMSE as well as advances in accuracy and correlation, compared with various benchmark models, including random forest, deep forest, extreme gradient boosting and support vector regression.

Spatio-Temporal Evolution Dynamic, Effect and Governance Policy of Construction Land Use in Urban Agglomeration: Case Study of Yangtze River Delta, China

The urban construction land change is the most obvious and complex spatial phenomenon in urban agglomerations which has attracted extensive attention of scholars in different fields. Yangtze River Delta Urban Agglomeration is the most mature urban agglomeration in China, a typical representative in both China and the world. This paper analyzes the evolution dynamic, effect and governance policy of urban construction land in Yangtze River Delta Urban Agglomeration 2011–2020 using a combination of BCG model, decoupling model and GIS tools. The findings are as follows. (1) There are large intercity differences in urban construction land in urban agglomerations, but the spatial heterogeneity is gradually decreasing. (2) The change trends and evolution patterns of urban construction land in urban agglomerations are increasingly diversified, with emergence of a variety of types such as rapid growth, slow growth, inverted U-shape, stars, cows, question and dogs. (3) The population growth, economic development and income improvement corresponding to the change of urban construction land in urban agglomerations have no desirable effect, with most cities in the expansive negative decoupling state. (4) The decoupling types show increasingly complex changes, in evolution, degeneration and unchanged states. Affected by economic transformation and the outbreak of COVID-19, an increasing number of cities are in strong negative decoupling and degeneration states, threatening the sustainable development of urban agglomerations. (5) Based on the division of urban agglomerations into three policy areas of Transformation Leading, Land Dependent, and Land Reduction, the response strategies for each are proposed, and a differentiated land use zoning management system is established.

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

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

Developing air pollution concentration fields for health studies using multiple methods: Cross-comparison and evaluation

Past air pollution epidemiological studies have used a wide range of methods to develop concentration fields for health analyses. The fields developed differ considerably among these methods. The reasons for these differences and comparisons of their strengths, as well as the limitations for estimating exposures, remains under-investigated. Here, we applied nine methods to develop fields of eight pollutants (carbon monoxide (CO), nitrogen dioxide (NO₂), sulfur dioxide (SO₂), ozone (O₃), fine particulate matter (PM_2.5), and three speciated PM_2.5 constituents including elemental carbon (EC), organic carbon (OC), and sulfate (SO₄)) for the metropolitan Atlanta region for five years. The nine methods are Central Monitor (CM), Site Average (SA), Inverse Distance Weighting (IDW), Kriging (KRIG), Land Use Regression (LUR), satellite Aerosol Optical Depth (AOD), CMAQ model, CMAQ with kriging adjustment (CMAQ-KRIG), and CMAQ based data fusion (CMAQ-DF). Additionally, we applied an increasingly popular method, Random Forest (RF), and compared its results for NO₂ and PM_2.5 with other methods. For statistical evaluation, we focused our discussion on the temporal coefficient of determination, although other metrics are also calculated. Raw output from the CMAQ model contains modeling biases and errors, which are partially mitigated by fusing observational data in the CMAQ-KRIG and CMAQ-DF methods. Based on analyses that simulated model responses to more limited input data, the RF model is more robust and outperforms LUR for PM_2.5. These results suggest RF may have potential in air pollution health studies, especially when limited measurement data are available. The RF method has several important weaknesses, including a relatively poor performance for NO₂, diagnostic challenges, and computational intensiveness. The results of this study will help to improve our understanding of the strengths and weaknesses of different methods for estimating air pollutant exposures in epidemiological studies.

Development and Evaluation of Spatio-Temporal Air Pollution Exposure Models and Their Combinations in the Greater London Area, UK

Land use regression (LUR) and dispersion/chemical transport models (D/CTMs) are frequently applied to predict exposure to air pollution concentrations at a fine scale for use in epidemiological studies. Moreover, the use of satellite aerosol optical depth data has been a key predictor especially for particulate matter pollution and when studying large populations. Within the STEAM project we present a hybrid spatio-temporal modeling framework by (a) incorporating predictions from dispersion modeling of nitrogen dioxide (NO₂), ozone (O₃) and particulate matter with an aerodynamic diameter equal or less than 10 μm (PM10) and less than 2.5 μm (PM2.5) into a spatio-temporal LUR model; and (b) combining the predictions LUR and dispersion modeling and additionally, only for PM2.5, from an ensemble machine learning approach using a generalized additive model (GAM). We used air pollution measurements from 2009 to 2013 from 62 fixed monitoring sites for O3, 115 for particles and up to 130 for NO₂, obtained from the dense network in the Greater London Area, UK. We assessed all models following a 10-fold cross validation (10-fold CV) procedure. The hybrid models performed better compared to separate LUR models. Incorporation of the dispersion estimates in the LUR models as a predictor, improved the LUR model fit: CV-R² increased to 0.76 from 0.71 for NO₂, to 0.79 from 0.57 for PM10, to 0.81 to 0.66 for PM2.5 and to 0.75 from 0.62 for O₃. The CV-R² obtained from the hybrid GAM framework was also increased compared to separate LUR models (CV-R² = 0.80 for NO₂, 0.76 for PM10, 0.79 for PM2.5 and 0.75 for O₃). Our study supports the combined use of different air pollution exposure assessment methods in a single modeling framework to improve the accuracy of spatio-temporal predictions for subsequent use in epidemiological studies.

Machine learning-based estimation of ground-level NO2 concentrations over China

Most current scientific research on NO₂ remote sensing focuses on tropospheric NO₂ column concentrations rather than ground-level NO₂ concentrations; however, ground-level NO₂ concentrations are more related to anthropogenic emissions and human health. This study proposes a machine learning estimation method for retrieving the ground-level NO₂ concentrations throughout China based on the tropospheric NO₂ column concentrations from the TROPOspheric Monitoring Instrument (TROPOMI) and multisource geographic data from 2018 to 2020. This method adopts the XGBoost machine learning model characterized by a strong fitting ability and complex model structure, which can explain the complex nonlinear and high-order relationships between ground-measured NO₂ and its influencing factors. The R² values between the retrievals and the validation and test datasets are 0.67 and 0.73, respectively, which suggests that the proposed method can reliably retrieve the ground-level NO₂ concentrations across China. The distribution characteristics, seasonal variations and interannual differences in ground-level NO₂ concentrations are further analyzed based on the retrieval results, demonstrating that the ground-level NO₂ concentrations exhibit significant geographical and seasonal variations, with high concentrations in winter and low concentrations in summer, and the highly polluted regions are concentrated mainly in Beijing-Tianjin-Hebei (BTH), the Yangtze River Delta (YRD), the Pearl River Delta (PRD), Cheng-Yu District (CY) and other urban agglomerations. Finally, the interannual variation in the ground-level NO₂ concentrations indicates that pollution decreased continuously from 2018 to 2020.

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.

A multi-step machine learning approach to assess the impact of COVID-19 lockdown on NO2 attributable deaths in Milan and Rome, Italy

BACKGROUND

Air pollution is one of the main concerns for the health of European citizens, and cities are currently striving to accomplish EU air pollution regulation. The 2020 COVID-19 lockdown measures can be seen as an unintended but effective experiment to assess the impact of traffic restriction policies on air pollution. Our objective was to estimate the impact of the lockdown measures on NO2 concentrations and health in the two largest Italian cities.

METHODS

NO2 concentration datasets were built using data deriving from a 1-month citizen science monitoring campaign that took place in Milan and Rome just before the Italian lockdown period. Annual mean NO2 concentrations were estimated for a lockdown scenario (Scenario 1) and a scenario without lockdown (Scenario 2), by applying city-specific annual adjustment factors to the 1-month data. The latter were estimated deriving data from Air Quality Network stations and by applying a machine learning approach. NO2 spatial distribution was estimated at a neighbourhood scale by applying Land Use Random Forest models for the two scenarios. Finally, the impact of lockdown on health was estimated by subtracting attributable deaths for Scenario 1 and those for Scenario 2, both estimated by applying literature-based dose-response function on the counterfactual concentrations of 10 μg/m3.

RESULTS

The Land Use Random Forest models were able to capture 41-42% of the total NO2 variability. Passing from Scenario 2 (annual NO2 without lockdown) to Scenario 1 (annual NO2 with lockdown), the population-weighted exposure to NO2 for Milan and Rome decreased by 15.1% and 15.3% on an annual basis. Considering the 10 μg/m3 counterfactual, prevented deaths were respectively 213 and 604.

CONCLUSIONS

Our results show that the lockdown had a beneficial impact on air quality and human health. However, compliance with the current EU legal limit is not enough to avoid a high number of NO2 attributable deaths. This contribution reaffirms the potentiality of the citizen science approach and calls for more ambitious traffic calming policies and a re-evaluation of the legal annual limit value for NO2 for the protection of human health.

Land use regression modelling of NO2 in São Paulo, Brazil

BACKGROUND

Air pollution is a major global public health problem. The situation is most severe in low- and middle-income countries, where pollution control measures and monitoring systems are largely lacking. Data to quantify the exposure to air pollution in low-income settings are scarce.

METHODS

In this study, land use regression models (LUR) were developed to predict the outdoor nitrogen dioxide (NO₂) concentration in the study area of the Western Region Birth Cohort in São Paulo. NO₂ measurements were performed for one week in winter and summer at eighty locations. Additionally, weekly measurements at one regional background location were performed over a full one-year period to create an annual prediction.

RESULTS

Three LUR models were developed (annual, summer, winter) by using a supervised stepwise linear regression method. The winter, summer and annual models explained 52 %, 75 % and 66 % of the variance (R²) respectively. Cross-holdout validation tests suggest robust models. NO₂ levels ranged from 43.2 μg/m³ to 93.4 μg/m³ in the winter and between 28.1 μg/m³ and 72.8 μg/m³ in summer. Based on our annual prediction, about 67 % of the population living in the study area is exposed to NO₂ values over the WHO suggested annual guideline of 40 μg/m³ annual average.

CONCLUSION

In this study we were able to develop robust models to predict NO₂ residential exposure. We could show that average measures, and therefore the predictions of NO_2, in such a complex urban area are substantially high and that a major variability within the area and especially within the season is present. These findings also suggest that in general a high proportion of the population is exposed to high NO₂ levels.

Land use land cover mapping using advanced machine learning classifiers

Due to the recent climate changes such as floods and droughts, there is a need for Land Use Land Cover (LULC) mapping to monitor environmental changes that have effects on ecology, policy management, health and disaster management. As such, in this study, two well-known machine learning classifiers, namely, Support Vector Machine (SVM) and Random Forest (RF), are used for land cover mapping. In addition, two advanced deep learning algorithms, namely, the GAMLP and FSMLP, that are based on the Multi-layer Perceptron (MLP) function are developed in MATLAB programming language. The GAMLP uses a Genetic Algorithm (GA) to optimise parameters of the MLP function and, on the other hand, the FSMLP uses a derivative-free function for optimisation of the MLP function parameters. Three different scenarios using Landsat-8 imagery with spatial resolutions of 30 and 15 m are defined to investigate the effects of data pre-processing on the final predicted LULC map. Results based on the statistical indices, including overall accuracy (OA) and kappa index, show that the developed MLP-based algorithms have relatively high accuracies with higher than 98% correct classification. Besides the statistical indices, final LULC maps are interpreted visually where the GAMLP and FSMLP give the best results for the pre-processed Landsat-8 imagery with a spatial resolution of 15 m, but they have the worst outcomes for the unprocessed Landsat-8 imagery compared to SVM and RF classifiers visually and statistically.

Estimating monthly concentrations of ambient key air pollutants in Japan during 2010-2015 for a national-scale birth cohort

Exposure to ambient air pollution is associated with maternal and child health. Some air pollutants exhibit similar behavior in the atmosphere, and some interact with each other; thus, comprehensive assessments of individual air pollutants are required. In this study, we developed national-scale monthly models for six air pollutants (NO, NO₂, SO₂, O₃, PM_2.5, and suspended particulate matter (SPM)) to obtain accurate estimates of pollutant concentrations at 1 km × 1 km resolution from 2010 through 2015 for application to the Japan Environment and Children's Study (JECS), which is a large-scale birth cohort study. We developed our models in the land use regression framework using random forests in conjunction with kriging. We evaluated the model performance via 5-fold location-based cross-validation. We successfully predicted monthly NO (r² = 0.65), NO₂ (r² = 0.84), O₃ (r² = 0.86), PM_2.5 (r² = 0.79), and SPM (r² = 0.64) concentrations. For SO₂, a satisfactory model could not be developed (r² = 0.45) because of the low SO₂ concentrations in Japan. The performance of our models is comparable to those reported in previous studies at similar temporal and spatial scales. The model predictions in conjunction with the JECS will reveal the critical windows of prenatal and infancy exposure to ambient air pollutants, thus contributing to the development of environmental policies on air pollution.

A land use regression model using machine learning and locally developed low cost particulate matter sensors in Uganda

The application of land use regression (LUR) modeling for estimating air pollution exposure has been used only rarely in sub-Saharan Africa (SSA). This is generally due to a lack of air quality monitoring networks in the region. Low cost air quality sensors developed locally in sub-Saharan Africa presents a sustainable operating mechanism that may help generate the air monitoring data needed for exposure estimation of air pollution with LUR models. The primary objective of our study is to investigate whether a network of locally developed low-cost air quality sensors can be used in LUR modeling for accurately predicting monthly ambient fine particulate matter (PM2.5) air pollution in urban areas of central and eastern Uganda. Secondarily, we aimed to explore whether the application of machine learning (ML) can improve LUR predictions compared to ordinary least squares (OLS) regression. We used data for the entire year of 2020 from a network of 23 PM2.5 low-cost sensors located in urban municipalities of eastern and central Uganda. Between January 1, 2020 and December 31, 2020, these sensors collected highly time-resolved measurement data of PM2.5 air concentrations. We used monthly-averaged PM2.5 concentration data for LUR prediction modeling of monthly PM2.5 concentrations. We used eight different ML base-learner algorithms as well as ensemble modeling. We applied 5-fold cross validation (80% training/20% test random splits) to evaluate the models with resampling and Root mean squared error (RMSE). The relative explanatory power and accuracy of the ML algorithms were evaluated by comparing coefficient of determination (R²) and RMSE, using OLS as the reference approach. The overall average PM2.5 concentration during the study period was 52.22 μg/m³ (IQR: 38.11, 62.84 μg/m³)-well above World Health Organization PM2.5 ambient air guidelines. From the base-learner and ensemble models, RMSE and R² values ranged between 7.65 μg/m³ - 16.85 μg/m³ and 0.24-0.84, respectively. Extreme gradient boosting (xgbTree) performed best out of the base learner algorithms (R² = 0.84; RMSE = 7.65 μg/m³). Model performance from ensemble modeling with Lasso and Elastic-Net Regularized Generalized Linear Models (glmnet) did not outperform xgbTree, but prediction performance was comparable to that of xgbTree. The most important temporal and spatial predictors of monthly PM2.5 levels were monthly precipitation, percent of the population using solid fuels for cooking, distance to Lake Victoria, and greenspace (NDVI) within a 500-m buffer of air monitors. In conclusion, data from locally developed low-cost PM sensors provide evidence that they can be used for spatio-temporal prediction modeling of air pollution exposures in Uganda. Moreover, the non-parametric ML and ensemble approaches to LUR modeling clearly outperformed OLS regression algorithm for the prediction of monthly PM2.5 concentrations. Deploying low-cost air quality sensors in concert with implementation of data quality control measures, can help address the critical need for expanding and improving air quality monitoring in resource-constrained settings of sub-Saharan Africa. These low-cost sensors, in conjunction with non-parametric ML algorithms, may provide a rapid path forward for PM2.5 exposure assessment and to spur air pollution epidemiology research in the region.

Underestimated or overestimated? Dynamic assessment of hourly PM2.5 exposure in the metropolitan area based on heatmap and micro-air monitoring stations

Spatio-temporal distributions of air pollution and population are two important factors influencing the patterns of mortality and diseases. Past studies have quantified the adverse effects of long-term exposure to air pollution. However, the dynamic changes of air pollution levels and population mobility within a day are rarely taken into consideration, especially in metropolitan areas. In this study, we use the high-resolution PM_2.5 data from the micro-air monitoring stations, and hourly population mobility simulated by the heatmap based on Location Based Service (LBS) big data to evaluate the hourly active PM_2.5 exposure in a typical Chinese metropolis. The dynamic "active population exposure" is compared spatiotemporally with the static "census population exposure" based on census data. The results show that over 12 h on both study periods, 45.83% of suburbs' population-weighted exposure (PWE) is underestimated, while 100% of rural PWE and more than 34.78% of downtown's PWE are overestimated, with the relative difference reaching from -11 μg/m³ to 7 μg/m³. More notably, the total PWE of the active population at morning peak hours on weekdays is worse than previously realized, about 12.41% of people are exposed to PM_2.5 over 60 μg/m³, about twice as much as that in census scenario. The commuters who live in the suburbs and work in downtown may suffer more from PM_2.5 exposure and uneven environmental resource distribution. This study proposes a new approach of calculating population exposure which can also be extended to quantify other environmental issues and related health burdens.

Statistical Approaches for Forecasting Primary Air Pollutants: A Review

Air pollutant forecasting can be used to quantitatively estimate pollutant reduction trends. Combining bibliometrics with the evolutionary tree and Markov chain methods can achieve a superior quantitative analysis of research hotspots and trends. In this work, we adopted a bibliometric method to review the research status of statistical prediction methods for air pollution, used evolutionary trees to analyze the development trend of such research, and applied the Markov chain to predict future research trends for major air pollutants. The results indicate that papers mainly focused on the effects of air pollution on human diseases, urban pollution exposure models, and land use regression (LUR) methods. Particulate matter (PM), nitrogen oxides (NOx), and ozone (O3) were the most investigated pollutants. Artificial neural network (ANN) methods were preferred in studies of PM and O3, while LUR were more widely used in studies of NOx. Additionally, multi-method hybrid techniques gradually became the most widely used approach between 2010 and 2018. In the future, the statistical prediction of air pollution is expected to be based on a mixed method to simultaneously predict multiple pollutants, and the interaction between pollutants will be the most challenging aspect of research on air pollution prediction. The research results summarized in this paper provide technical support for the accurate prediction of atmospheric pollution and the emergency management of regional air quality.

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

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

On-field test and data calibration of a low-cost sensor for fine particles exposure assessment

BACKGROUND

Accurate individual exposure assessment is crucial for evaluating the health effects of particulate matter (PM). Various portable monitors built upon low-cost optical sensors have emerged. However, the main challenge for their application is to guarantee accuracy of measurements.

OBJECTIVE

To assess the performance of a newly developed PM sensor, and to develop methods for post-hoc data calibration to optimize its data quality.

METHOD

We conducted a series of laboratory experiments and field evaluations to quantify the reproducibility within Plantower PM sensors 7003 (PMS 7003) and the consistency between sensors and two established PM_2.5 measurement methods [tapered element oscillating microbalances (TEOM) and gravimetric method (GM)]. Post-hoc data calibration methods for sensors were based on a multiple linear regression model (MLRM) and a random forest model (RFM). Ratios of raw and calibrated readings over the data of reference methods were calculated to examine the improvement after calibration.

RESULTS

Strong correlations (≥0.82) and relatively small relative standard deviations (16-21%) between sensors were found during the laboratory and the field sampling. Compared with the reference methods, moderate to strong coefficients of determination (0.56-0.83) were observed; however, significant deviations were presented. After calibration, the ratios of PMS measurements over that of two reference methods both became convergent.

CONCLUSIONS

Our study validated low-cost optical PM sensors under a wide range of PM_2.5 concentrations (8-167 μg/m³). Our findings indicated potential applicability of PM sensors in PM_2.5 exposure assessment, and confirmed a need of calibration. Linear calibration methods may be sufficient for ambient monitoring using TEOM as a reference, while nonlinear calibration methods may be more appropriate for indoor monitoring using GM as a reference.

Estimating the Daily NO2 Concentration with High Spatial Resolution in the Beijing–Tianjin–Hebei Region Using an Ensemble Learning Model

Nitrogen dioxide (NO2) is an important pollutant related to human activities, which has short-term and long-term effects on human health. An ensemble learning model was constructed and applied to estimate daily NO2 concentrations in the Beijing–Tianjin–Hebei region between 2010 and 2016. A variety of predictive variables included satellite-based troposphere NO2 vertical column concentration, meteorology, elevation, gross domestic product (GDP), population, land-use variables, and road network. The ensemble learning model achieved two things: a 0.01° × 0.01° grid resolution and the estimation of historical data for the years 2010–2013. The ensemble model showed good performance, whereby the R2 of tenfold cross-validation was 0.72 and the R2 of test validation was 0.71. Meteorological hysteretic effects were incorporated into the model, where the one-day lagged boundary layer height contributed the most. The annual NO2 estimation showed little change from 2010 to 2016. The seasonal NO2 estimation from highest to lowest occurred in winter, autumn, spring, and summer. In the annual maps and seasonal maps, the NO2 estimations in the northwest region were lower than those in the southeast region, and there was a heavily polluted band in the south of the Taihang Mountains. In coastal areas, the annual NO2 estimations were higher than the NO2 monitored values. The drawback of the model is underestimation at high values and overestimation at low values. This study indicates that the ensemble learning model has excellent performance in the simulation of NO2 with high spatial and temporal resolution. Furthermore, the research framework in this study can be a generally applied for drawing implications for other regions, especially for other cities in China.

Modeling estrogenic activity in streams throughout the Potomac and Chesapeake Bay watersheds

Endocrine-disrupting compounds (EDCs), specifically estrogenic endocrine-disrupting compounds, vary in concentration and composition in surface waters under the influence of different landscape sources and landcover gradients. Estrogenic activity in surface waters may lead to adverse effects in aquatic species at both individual and population levels, often observed through the presence of intersex and vitellogenin induction in male fish. In the Chesapeake Bay Watershed, located on the mid-Atlantic coast of the USA, intersex has been observed in several sub-watersheds where previous studies have identified specific landscape sources of EDCs in tandem with observed fish health effects. Previous work in the Potomac River Watershed (PRW), the largest basin within the Chesapeake Bay Watershed, was leveraged to build random forest regression models to predict estrogenic activity at unsampled reaches in both the Potomac River and larger Chesapeake Bay Watersheds (CBW). Model outputs including important variables, partial dependence plots, and predicted values of estrogenic activity at unsampled reaches provide insight into drivers of estrogenic activity at different seasons and scales. Using the US Environmental Protection Agency effects-based threshold of 1.0 ng/L 17 β-estradiol equivalents, catchments predicted to exceed this value were categorized as at risk for adverse effects from exposure to estrogenic compounds and evaluated relative to healthy watersheds and recreation access locations throughout the PRW. Results show immediate catchment scale models are more reliable than upstream models, and the best predictive variables differ by season and scale. A small percentage of healthy watersheds (< 13%) and public access sites were classified as at risk using the "Total" (annual) model in the CBW. This study is the first Potomac River Watershed assessment of estrogenic activity, providing a new foundation for future risk assessment and management design efforts, with additional context provided for the entire Chesapeake Bay Watershed.

Early Evidence of Inactivated Enterovirus 71 Vaccine Impact Against Hand, Foot, and Mouth Disease in a Major Center of Ongoing Transmission in China, 2011-2018: A Longitudinal Surveillance Study

BACKGROUND

Enterovirus 71 (EV71) is a major causative agent of hand, foot, and mouth disease (HFMD), associated with severe manifestations of the disease. Pediatric immunization with inactivated EV71 vaccine was initiated in 2016 in the Asia-Pacific region, including China. We analyzed a time series of HFMD cases attributable to EV71, coxsackievirus A16 (CA16), and other enteroviruses in Chengdu, a major transmission center in China, to assess early impacts of immunization.

METHODS

Reported HFMD cases were obtained from China's notifiable disease surveillance system. We compared observed postvaccination incidence rates during 2017-2018 with counterfactual predictions made from a negative binomial regression and a random forest model fitted to prevaccine years (2011-2015). We fit a change point model to the full time series to evaluate whether the trend of EV71 HFMD changed following vaccination.

RESULTS

Between 2011 and 2018, 279 352 HFMD cases were reported in the study region. The average incidence rate of EV71 HFMD in 2017-2018 was 60% (95% prediction interval [PI], 41%-72%) lower than predicted in the absence of immunization, corresponding to an estimated 6911 (95% PI, 3246-11 542) EV71 cases averted over 2 years. There were 52% (95% PI, 42%-60%) fewer severe HFMD cases than predicted. However, the incidence rate of non-CA16 and non-EV71 HFMD was elevated in 2018. We identified a significant decline in the trend of EV71 HFMD 4 months into the postvaccine period.

CONCLUSIONS

We provide the first real-world evidence that programmatic vaccination against EV71 is effective against childhood HFMD and present an approach to detect early vaccine impact or intended consequences from surveillance data.

Lag Variables in Nitrogen Oxide Concentration Modelling: A Case Study in Wrocław, Poland

Due to the unwavering interest of both residents and authorities in the air quality of urban agglomerations, we pose the following question in this paper: What impact do current and past meteorological factors and traffic flow intensity have on air quality? What is the impact of lagged variables on the fit of an explanation model, and how do they affect its ability to predict? We focused on NO2 and NOx concentrations, and conducted this research using hourly data from the city of Wrocław (western Poland) from 2015 to 2017; we used multi-objective optimization to determine the optimal delays. It turned out that for both NO2 and NOx, the past values for traffic flow, wind speed, and sunshine duration are more important than the current ones. We built random forest models on each of the pollutants for both the current and past values and discovered that including a lagged variable increases the resulting R2 from 0.51 to 0.56 for NO2 and from 0.46 to 0.52 for NOx. We also analyzed the feature importance in each model, and found that for NO2, a wind speed delay of more than three hours causes a significant decrease, while the importance of relative humidity increases with a seven-hour delay; likewise, wind speed increases the importance for NOx prediction with a two-hour delay. We concluded that, in pollutant concentration modeling, the possibility of a delayed effect of the independent variables should always be considered, because it can significantly increase the performance of the model and suggest unexpected relationships or dependencies.

A Machine Learning Approach to Investigate the Surface Ozone Behavior

The concentration of surface ozone (O3) strongly depends on environmental and meteorological variables through a series of complex and non-linear functions. This study aims to explore the performances of an advanced machine learning (ML) method, the boosted regression trees (BRT) technique, in exploring the relationships between surface O3 and its driving factors, and in predicting the levels of O3 concentrations. To this end, a BRT model was trained on hourly data of air pollutants and meteorological parameters, acquired, over the 2016–2018 period, in a rural area affected by an anthropic source of air pollutants. The abilities of the BRT model in ranking, visualizing, and predicting the relationship between ground-level O3 concentrations and its driving factors were analyzed and illustrated. A comparison with a multiple linear regression (MLR) model was performed based on several statistical indicators. The results obtained indicated that the BRT model was able to account for 81% of changes in O3 concentrations; it slightly outperforms the MLR model in terms of the predictions accuracy and allows a better identification of the main factors influencing O3 variability on a local scale. This knowledge is expected to be useful in defining effective measures to prevent and/or mitigate the health damages associated with O3 exposure.

Estimating daily ground-level PM2.5 in China with random-forest-based spatiotemporal kriging

Ambient fine particulate matter (PM_2.5) plays an important role in cardiovascular- and respiratory-related death. Empirical statistical models have been widely applied to estimate ambient PM_2.5 concentrations with correlated variables. However, empirical statistical models ignore the nonlinear relationship between PM_2.5 and covariates and assume that residuals are independent and identically distributed random variables. Here, a hybrid approach, which integrates random forest (RF) model and spatiotemporal kriging, is proposed to estimate the daily PM_2.5 concentration. The proposed RF-based spatiotemporal kriging (RFSTK) model effectively captures nonlinear interactions among different predictors and accounts for the detailed spatiotemporal dependence of the PM_2.5 concentration. The RFSTK model performs well in predicting the daily PM_2.5 concentration. The 10-fold overall cross-validation R² value is 0.881, the mean absolute error (MAE) is 6.89 μg/m³ and the root-mean-square error (RMSE) is 11.48 μg/m³, indicating better performance than the original RF model (R² = 0.848, MAE = 7.88 μg/m³ and RMSE = 13.26 μg/m³). The spatiotemporal prediction of the PM_2.5 concentration shows that approximately 90.04% of China had a daily exposure to PM_2.5 in 2018 that was below the nation's air quality standard of 75 μg/m³. The proposed hybrid method is entirely general and can be applied to map the ambient PM_2.5 concentration over a large spatiotemporal domain.

Kriging-Based Land-Use Regression Models That Use Machine Learning Algorithms to Estimate the Monthly BTEX Concentration

This paper uses machine learning to refine a Land-use Regression (LUR) model and to estimate the spatial–temporal variation in BTEX concentrations in Kaohsiung, Taiwan. Using the Taiwanese Environmental Protection Agency (EPA) data of BTEX (benzene, toluene, ethylbenzene, and xylenes) concentrations from 2015 to 2018, which includes local emission sources as a result of Asian cultural characteristics, a new LUR model is developed. The 2019 data was then used as external data to verify the reliability of the model. We used hybrid Kriging-land-use regression (Hybrid Kriging-LUR) models, geographically weighted regression (GWR), and two machine learning algorithms—random forest (RF) and extreme gradient boosting (XGBoost)—for model development. Initially, the proposed Hybrid Kriging-LUR models explained each variation in BTEX from 37% to 52%. Using machine learning algorithms (XGBoost) increased the explanatory power of the models for each BTEX, between 61% and 79%. This study compared each combination of the Hybrid Kriging-LUR model and (i) GWR, (ii) RF, and (iii) XGBoost algorithm to estimate the spatiotemporal variation in BTEX concentration. It is shown that a combination of Hybrid Kriging-LUR and the XGBoost algorithm gives better performance than other integrated methods.

Spatio-temporal air pollution modelling using a compositional approach

Air pollutant data are compositional in character because they describe quantitatively the parts of a whole (atmospheric composition). However, it is common to use air pollutant concentrations in statistical models without considering this characteristic of the data and, therefore, without control of common statistical problems, such as spurious correlations and subcompositional incoherence. This paper now proposes a daily multivariate spatio-temporal model with a compositional approach. The air pollution spatio-temporal model is based on a dynamic linear modelling framework with Bayesian inference. The novel modelling methodology was applied in an urban area for carbon monoxide (CO, mg·m-3), sulfur dioxide (SO2, μg·m-3), ozone (O3, μg·m-3), nitrogen dioxide (NO2, μg·m-3), and particulate matter less than 2.5 μm in aerodynamic diameter (PM2.5, μg·m-3). The proposal complemented and improved the conventional approach in air pollution modelling. The main improvements come from a fast multivariate data description, high spatial-correlation, and adequate modelling of air pollutants with high variability.

Estimating historical PM2.5 exposures for three decades (1987-2016) in Japan using measurements of associated air pollutants and land use regression

Accurate estimation of historical PM_2.5 exposures for epidemiological studies is challenging when extensive monitoring data are limited in duration. Here, we develop a national-scale PM_2.5 exposure model for Japan using measurements recorded between 2014 and 2016 to estimate monthly means for 1987 through 2016. Our objective is to obtain accurate PM_2.5 estimates for years prior to implementation of extensive PM_2.5 monitoring, using observations from a limited period. We utilize a neural network to convey the non-linear relationship between the target pollutant and predictors, while incorporating the associated air pollutants. We obtain high R² values of 0.76 and 0.73 through spatial and temporal cross validation, respectively. We evaluate estimation accuracy using an independent data set and achieve an R² of 0.75. Moreover, monthly variations for 2000-2013 are well reproduced with correlation coefficients of greater than 0.78, obtained through a comparison with observations. We estimate monthly means at 1 × 1 km resolution from 1987 through 2016. The estimates show decreases in the area and population weighted means beginning in the 1990s. We successfully estimate monthly mean PM_2.5 concentrations over three decades with outstanding predictive accuracy. Our findings illustrate that the presented approach achieves accurate long-term historical estimations using observations limited in duration.

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

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

Assessing NO2 Concentration and Model Uncertainty with High Spatiotemporal Resolution across the Contiguous United States Using Ensemble Model Averaging

NO₂ is a combustion byproduct that has been associated with multiple adverse health outcomes. To assess NO₂ levels with high accuracy, we propose the use of an ensemble model to integrate multiple machine learning algorithms, including neural network, random forest, and gradient boosting, with a variety of predictor variables, including chemical transport models. This NO₂ model covers the entire contiguous U.S. with daily predictions on 1-km-level grid cells from 2000 to 2016. The ensemble produced a cross-validated R² of 0.788 overall, a spatial R² of 0.844, and a temporal R² of 0.729. The relationship between daily monitored and predicted NO₂ is almost linear. We also estimated the associated monthly uncertainty level for the predictions and address-specific NO₂ levels. This NO₂ estimation has a very high spatiotemporal resolution and allows the examination of the health effects of NO₂ in unmonitored areas. We found the highest NO₂ levels along highways and in cities. We also observed that nationwide NO₂ levels declined in early years and stagnated after 2007, in contrast to the trend at monitoring sites in urban areas, where the decline continued. Our research indicates that the integration of different predictor variables and fitting algorithms can achieve an improved air pollution modeling framework.

Solar radiation forecasting using MARS, CART, M5, and random forest model: A case study for India

Solar radiation is a critical requirement for all solar power plants. As it is a time-varying quantity, the power output of any solar power plant is also time variant in nature. Hence, for the prediction of probable electricity generation for a few days in advance, for any solar power plant, forecasting solar radiation a few days into the future is vital. Hourly forecasting for a few days in advance may help a utility or ISO in the bidding process. In this study, 1-day-ahead to 6-day-ahead hourly solar radiation forecasting was been performed using the MARS, CART, M5 and random forest models. The data required for the forecasting were collected from a solar radiation resource setup, commissioned by an autonomous body of the Government of India in Gorakhpur, India. From the results, it was determined that, for the present study, the random forest model provided the best results, whereas the CART model presented the worst results among all four models considered.

A kriging-calibrated machine learning method for estimating daily ground-level NO2 in mainland China

It is unclear how to develop a model based on the combined satellite data and ground monitoring data to accurately estimate daily NO₂ levels. Furthermore, the conventional cross-validation (CV) results represent average levels but the model performance may vary greatly from grid to grid. It is an essential issue to evaluate model's prediction ability in different grids and determine the factors affecting model extrapolating ability, which have never been well examined to date. The aim of this study was to compare the ability of three different methods to estimate the daily NO₂ across mainland China during 2014-2016; and to develop a novel two-stage meta-analysis method for exploring the influence of the number and the distribution of nearby sites on grid-level prediction accuracy. For better estimating the daily NO₂ level, we developed and compared three methods, including universal kriging model, satellite-based method (Non-linear exposure-lag-response model & Extreme gradient boosting combined technique) and the kriging-calibrated satellite method. For exploring influencing factors, the two-stage meta-analysis method was purposed. The kriging-calibrated satellite method had an overall CV R-square and root mean square error (RMSE) of 0.85 and 7.87μg/m³, better than the Universal Kriging model and the satellite-based method (CV R² = 0.57 and 0.81). The two-stage meta-analysis method revealed that the model performance did decrease with the sparser distribution of nearby sites. And adding 5 sites within 50 km in the random mode can bring 17.51% improvement in model extrapolating ability. The kriging-calibration can help satellite-based machine learning to provide more accurate NO₂ prediction. Our novel evaluation method can provide the suggestion of adding new sites effectively within a limit budget.

A comparison of linear regression, regularization, and machine learning algorithms to develop Europe-wide spatial models of fine particles and nitrogen dioxide

Empirical spatial air pollution models have been applied extensively to assess exposure in epidemiological studies with increasingly sophisticated and complex statistical algorithms beyond ordinary linear regression. However, different algorithms have rarely been compared in terms of their predictive ability. This study compared 16 algorithms to predict annual average fine particle (PM_2.5) and nitrogen dioxide (NO₂) concentrations across Europe. The evaluated algorithms included linear stepwise regression, regularization techniques and machine learning methods. Air pollution models were developed based on the 2010 routine monitoring data from the AIRBASE dataset maintained by the European Environmental Agency (543 sites for PM_2.5 and 2399 sites for NO₂), using satellite observations, dispersion model estimates and land use variables as predictors. We compared the models by performing five-fold cross-validation (CV) and by external validation (EV) using annual average concentrations measured at 416 (PM_2.5) and 1396 sites (NO₂) from the ESCAPE study. We further assessed the correlations between predictions by each pair of algorithms at the ESCAPE sites. For PM_2.5, the models performed similarly across algorithms with a mean CV R² of 0.59 and a mean EV R² of 0.53. Generalized boosted machine, random forest and bagging performed best (CV R²~0.63; EV R² 0.58-0.61), while backward stepwise linear regression, support vector regression and artificial neural network performed less well (CV R² 0.48-0.57; EV R² 0.39-0.46). Most of the PM_2.5 model predictions at ESCAPE sites were highly correlated (R² > 0.85, with the exception of predictions from the artificial neural network). For NO₂, the models performed even more similarly across different algorithms, with CV R²s ranging from 0.57 to 0.62, and EV R²s ranging from 0.49 to 0.51. The predicted concentrations from all algorithms at ESCAPE sites were highly correlated (R² > 0.9). For both pollutants, biases were low for all models except the artificial neural network. Dispersion model estimates and satellite observations were two of the most important predictors for PM_2.5 models whilst dispersion model estimates and traffic variables were most important for NO₂ models in all algorithms that allow assessment of the importance of variables. Different statistical algorithms performed similarly when modelling spatial variation in annual average air pollution concentrations using a large number of training sites.

Unveiling tropospheric ozone by the traditional atmospheric model and machine learning, and their comparison:A case study in hangzhou, China

Tropospheric ozone in the surface air has become the primary atmospheric pollutant in Hangzhou, China, in recent years. Previous analysis is not enough to decode it for better regulation. Therefore, we use the traditional atmospheric model, Weather Research and Forecasting coupled with Community Multi-scale Air Quality (WRF-CMAQ), and machine learning models, Extreme Learning Machine (ELM), Multi-layer Perceptron (MLP), Random Forest (RF) and Recurrent Neural Network (RNN) to analyze and predict the ozone in the surface air in Hangzhou, China, using meteorology and air pollutants as input. We firstly quantitatively demonstrate that the dew-point deficit, instead of temperature and relative humidity, is the predominant meteorological factor in shaping tropospheric ozone. Urban heat island, daily direct solar radiation time, wind speed and wind direction play trivial role in impacting tropospheric ozone. NO₂ is the primary influential factors both for hourly ozone and daily O₃-8 h due to the titration effect. The most environmental-friendly way to mitigate the ozone pollution is to lower the volatile organic compounds (VOCs) with the highest ozone formation potentials. We deduce that the tropospheric ozone formation process tends to be not only non-linear but also non-smooth. Compared with the traditional atmospheric models, machine learning, whose characteristics are rapid convergence, short calculating time, adaptation of forecasting episodes, small program memory, higher accuracy and less cost, is able to predict tropospheric ozone more accurately.

Forecasting short-term peak concentrations from a network of air quality instruments measuring PM2.5 using boosted gradient machine models

Machine learning algorithms are used successfully in this paper to forecast reliably upcoming short-term high concentration episodes, or peaks (<60-min) of fine particulate air pollution (PM_2.5) 1 h in advance. Results are from a network around Christchurch, New Zealand, with an objective to forecast the occurrence of short-term peaks using a gradient boosted machine with a binary classifier as the response (1 = peak, 0 = no peak). Results are successful, with 80-90% accurate forecasting of whether a peak in PM_2.5 would occur within the next 60-min period. Elevated and variable nitrogen monoxide, nitrogen dioxide, and lower temperatures and wind gusts are found to be important precursors to the occurrence of PM_2.5 peaks. The use of meteorological data from a network of personal weather stations across the monitored area and from the measurement instruments was able to identify local-scale peak differences in the network. Boosted models using hourly-averaged and daily-averaged peaks as the response are developed separately to showcase differences in precursors between short-term and long-term peaks, with recent wind gusts and nitrogen oxides linked to hourly-averaged peaks and aloft air temperatures and atmospheric pressure linked to daily-averaged peaks. Results could prove useful in exposure mitigation strategies (e.g. as a short-term warning system).

Remote Sensing of Wetland Flooding at a Sub-Pixel Scale Based on Random Forests and Spatial Attraction Models

Wetland flooding is significant for the flora and fauna of wetlands. High temporal resolution remote sensing images are widely used for the timely mapping of wetland flooding but have a limitation of their relatively low spatial resolutions. In this study, a novel method based on random forests and spatial attraction models (RFSAM) was proposed to improve the accuracy of sub-pixel mapping of wetland flooding (SMWF) using remote sensing images. A random forests-based SMWF algorithm (RM-SMWF) was developed firstly, and a comprehensive complexity index of a mixed pixel was formulated. Then the RFSAM-SMWF method was developed. Landsat 8 Operational Land Imager (OLI) images of two wetlands of international importance included in the Ramsar List were used to evaluate RFSAM-SMWF against three other SMWF methods, and it consistently achieved more accurate sub-pixel mapping results in terms of visual and quantitative assessments in the two wetlands. The effects of the number of trees in random forests and the complexity threshold on the mapping accuracy of RFSAM-SMWF were also discussed. The results of this study improve the mapping accuracy of wetland flooding from medium-low spatial resolution remote sensing images and therefore benefit the environmental studies of wetlands.

Estimation of daily PM10 and PM2.5 concentrations in Italy, 2013-2015, using a spatiotemporal land-use random-forest model

Particulate matter (PM) air pollution is one of the major causes of death worldwide, with demonstrated adverse effects from both short-term and long-term exposure. Most of the epidemiological studies have been conducted in cities because of the lack of reliable spatiotemporal estimates of particles exposure in nonurban settings. The objective of this study is to estimate daily PM₁₀ (PM < 10 μm), fine (PM < 2.5 μm, PM_2.5) and coarse particles (PM between 2.5 and 10 μm, PM_2.5-10) at 1-km² grid for 2013-2015 using a machine learning approach, the Random Forest (RF). Separate RF models were defined to: predict PM_2.5 and PM_2.5-10 concentrations in monitors where only PM₁₀ data were available (stage 1); impute missing satellite Aerosol Optical Depth (AOD) data using estimates from atmospheric ensemble models (stage 2); establish a relationship between measured PM and satellite, land use and meteorological parameters (stage 3); predict stage 3 model over each 1-km² grid cell of Italy (stage 4); and improve stage 3 predictions by using small-scale predictors computed at the monitor locations or within a small buffer (stage 5). Our models were able to capture most of PM variability, with mean cross-validation (CV) R² of 0.75 and 0.80 (stage 3) and 0.84 and 0.86 (stage 5) for PM₁₀ and PM_2.5, respectively. Model fitting was less optimal for PM_2.5-10, in summer months and in southern Italy. Finally, predictions were equally good in capturing annual and daily PM variability, therefore they can be used as reliable exposure estimates for investigating long-term and short-term health effects.

Machine Learning Approaches for Outdoor Air Quality Modelling: A Systematic Review

Current studies show that traditional deterministic models tend to struggle to capture the non-linear relationship between the concentration of air pollutants and their sources of emission and dispersion. To tackle such a limitation, the most promising approach is to use statistical models based on machine learning techniques. Nevertheless, it is puzzling why a certain algorithm is chosen over another for a given task. This systematic review intends to clarify this question by providing the reader with a comprehensive description of the principles underlying these algorithms and how they are applied to enhance prediction accuracy. A rigorous search that conforms to the PRISMA guideline is performed and results in the selection of the 46 most relevant journal papers in the area. Through a factorial analysis method these studies are synthetized and linked to each other. The main findings of this literature review show that: (i) machine learning is mainly applied in Eurasian and North American continents and (ii) estimation problems tend to implement Ensemble Learning and Regressions, whereas forecasting make use of Neural Networks and Support Vector Machines. The next challenges of this approach are to improve the prediction of pollution peaks and contaminants recently put in the spotlights (e.g., nanoparticles).