The sediment transport, involving the movement of the bedload and suspended sediment in the basins, is a critical environmental concern that worsens water scarcity and leads to degradation of land and its ecosystems. Machine learning (ML) algorithms have emerged as powerful tools for predicting sediment yield. However, their use by decision-makers can be attributed to concerns regarding their consistency with the involved physical processes. In light of this issue, this study aims to develop a physics-informed ML approach for predicting sediment yield. To achieve this objective, Gaussian, Center, Regular, and Direct Copulas were employed to generate virtual combinations of physical of the sub-basins and hydrological datasets. These datasets were then utilized to train deep neural network (DNN), conventional neural network (CNN), Extra Tree, and XGBoost (XGB) models. The performance of these models was compared with the modified universal soil loss equation (MUSLE), which serves as a process-based model. The results demonstrated that the ML models outperformed the MUSLE model, exhibiting improvements in Nash–Sutcliffe efficiency (NSE) of approximately 10%, 18%, 32%, and 41% for the DNN, CNN, Extra Tree, and XGB models, respectively. Furthermore, through Sobol sensitivity and Shapley additive explanation–based interpretability analyses, it was revealed that the Extra Tree model displayed greater consistency with the physical processes underlying sediment transport as modeled by MUSLE. The proposed framework provides new insights into enhancing the accuracy and applicability of ML models in forecasting sediment yield while maintaining consistency with natural processes. Consequently, it can prove valuable in simulating process-related strategies aimed at mitigating sediment transport at watershed scales, such as the implementation of best management practices.

Evaluating ozone levels at high resolutions and accuracy is crucial for understanding the spatiotemporal characteristics of ozone distribution and assessing ozone exposure levels in epidemiological studies. The national models with high spatiotemporal resolutions to predict ground ozone concentrations are limited in China so far. In this study, we aimed to develop a random forest model by combining ground ozone measurements from fixed stations, ozone simulations from the Community Multiscale Air Quality (CMAQ) modeling system, meteorological parameters, population density, road length, and elevation to predict ground maximum daily 8-h average (MDA8) ozone concentrations at a daily level and 1 km × 1 km spatial resolution. The model cross-validation R² and root mean squared error (RMSE) were 0.80 and 20.93 μg/m³ at daily level in 2013–2019, respectively. CMAQ ozone simulations and near-surface temperature played vital roles in predicting ozone concentrations among all predictors. The population-weighted median concentrations of predicted MDA8 ozone were 89.34 μg/m³ in mainland China in 2013, and reached 100.96 μg/m³ in 2019. However, the long-term temporal variations among regions were heterogeneous. Central and Eastern China, as well as the Southeast Coastal Area, suffered higher ozone pollution and higher increased rates of ozone concentrations from 2013 to 2019. The seasonal pattern of ozone pollution varied spatially. The peak-season ozone pollution with the highest 6-month ozone concentrations occurred in different months among regions, with more than half domain in April–September. The predictions showed that not only the annual mean concentrations but also the percentages of grid-days with MDA8 ozone concentrations higher than 100/160 μg/m³ have been increasing in the past few years in China; meanwhile, majority areas in mainland China suffered peak-season ozone concentrations higher than the air quality guidelines launched by the World Health Organization in September 2021. The proposed model and ozone predictions with high spatiotemporal resolution and full coverage could provide health studies with flexible choices to evaluate ozone exposure levels at multiple spatiotemporal scales in the future.

The adverse health effects associated with the inhalation and ingestion of naturally occurring radon gas produced during the uranium decay chain mean that there is a need to identify high-risk areas. This study detected radon-prone areas using a geographic information system (GIS)-based probabilistic and machine learning methods, including the frequency ratio (FR) model and a convolutional neural network (CNN). Ten influencing factors, namely elevation, slope, the topographic wetness index (TWI), valley depth, fault density, lithology, and the average soil copper (Cu), calcium oxide (Cao), ferric oxide (Fe₂O₃), and lead (Pb) concentrations, were analyzed. In total, 27 rock samples with high activity concentration index values were divided randomly into training and validation datasets (70:30 ratio) to train the models. Areas were categorized as very high, high, moderate, low, and very low radon areas. According to the models, approximately 40% of the study area was classified as very high or high risk. Finally, the radon potential maps were validated using the area under the receiver operating characteristic curve (AUC) analysis. This showed that the CNN algorithm was superior to the FR method; for the former, AUC values of 0.844 and 0.840 were obtained using the training and validation datasets, respectively. However, both algorithms had high predictive power. Slope, lithology, and TWI were the best predictors of radon-affected areas. These results provide new information regarding the spatial distribution of radon, and could inform the development of new residential areas. Radon screening is important to reduce public exposure to high levels of naturally occurring radiation.

Finding an appropriate satellite image as simultaneous as possible with the sampling time campaigns is challenging. Fusion can be considered as a method of integrating images and obtaining more pixels with higher spatial, spectral and temporal resolutions. This paper investigated the impact of Landsat 8-OLI and Sentinel-2A data fusion on prediction of several toxic elements at a mine waste dump. The 30 m spatial resolution Landsat 8-OLI bands were fused with the 10 m Sentinel-2A bands using various fusion techniques namely hue-saturation-value (HSV), Brovey, principal component analysis (PCA), Gram-Schmidt (GS), wavelet, and area-to-point regression kriging (ATPRK). ATPRK was the best method preserving both spectral and spatial features of Landsat 8-OLI and Sentinel-2A after fusion. Furthermore, the partial least squares regression (PLSR) model developed on genetic algorithm (GA)-selected laboratory visible-near infrared-shortwave infrared (VNIR–SWIR) spectra yielded more accurate prediction results compared to the PLSR model calibrated on the entire spectra. It was hence, applied to both individual sensors and their ATPRK-fused image. In case of the individual sensors, except for As, Sentinel-2A provided more robust prediction models than Landsat 8-OLI. However, the best performances were obtained using the fused images, highlighting the potential of data fusion to enhance the toxic elements’ prediction models.

Cyclists can be easily exposed to traffic-related pollutants due to riding on or close to the road during commuting in cities. PM₂.₅ has been identified as one of the major pollutants emitted by vehicles and associated with cardiopulmonary and respiratory diseases. As routing has been suggested to reduce the exposures for cyclists, in this study, PM₂.₅ was monitored with low-cost sensors during commuting periods to develop models for identifying low exposure routes in three Asian cities: Taipei, Osaka, and Seoul. The models for mapping the PM₂.₅ in the cities were developed by employing the random forest algorithm in a two-stage modeling approach. The land use features to explain spatial variation of PM₂.₅ were obtained from the open-source land use database, OpenStreetMap. The total length of the monitoring routes ranged from 101.36 to 148.22 km and the average PM₂.₅ ranged from 13.51 to 15.40 μg/m³ among the cities. The two-stage models had the standard k-fold cross-validation (CV) R² of 0.93, 0.74, and 0.84 in Taipei, Osaka, and Seoul, respectively. To address spatial autocorrelation, a spatial cross-validation approach applying a distance restriction of 100 m between the model training and testing data was employed. The over-optimistic estimates on the predictions were thus prevented, showing model CV-R² of 0.91, 0.67, and 0.78 respectively in Taipei, Osaka, and Seoul. The comparisons between the shortest-distance and lowest-exposure routes showed that the largest percentage of reduced averaged PM₂.₅ exposure could reach 32.1% with the distance increases by 37.8%. Given the findings in this study, routing behavior should be encouraged. With the daily commuting trips expanded, the cumulative effect may become significant on the chronic exposures over time. Therefore, a route planning tool for reducing the exposures shall be developed and promoted to the public.

Traffic contributes to fine particulate matter (PM₂.₅) in the atmosphere through engine exhaust emissions and road dust generation. However, the evolution of traffic related PM₂.₅ emission over recent years remains unclear, especially when various efforts to reduce emission e.g., aftertreatment technologies and high emission standards from China IV to China V, have been implemented. In this study, hourly elemental carbon (EC), a marker of primary engine exhaust emissions, and trace element of calcium (Ca), a marker of road dust, were measured at a nearby highway sampling site in Shanghai from 2016 to 2019. A random forest-based machine learning algorithm was applied to decouple the influences of meteorological variables on the measured EC and Ca, revealing the deweathered trend in exhaust emissions and road dust. After meteorological normalization, we showed that non-exhaust emissions, i.e., road dust from traffic, increased their fractional contribution to PM₂.₅ over recent years. In particular, road dust was found to be more important, as revealed by the deweathered trend of Ca fraction in PM₂.₅, increasing at 6.1% year⁻¹, more than twice that of EC (2.9% year⁻¹). This study suggests that while various efforts have been successful in reducing vehicular exhaust emissions, road dust will not abate at a similar rate. The results of this study provide insights into the trend of traffic-related emissions over recent years based on high temporal resolution monitoring data, with important implications for policymaking.

Fine particulate matter (PM₂.₅) is associated with various adverse health outcomes and poses serious concerns for public health. However, ground monitoring stations for PM₂.₅ measurements are mostly installed in population-dense or urban areas. Thus, satellite retrieved aerosol optical depth (AOD) data, which provide spatial and temporal surrogates of exposure, have become an important tool for PM₂.₅ estimates in a study area. In this study, we used AOD estimates of surface PM₂.₅ together with meteorological and land use variables to estimate monthly PM₂.₅ concentrations at a spatial resolution of 3 km² over Taiwan Island from 2015 to 2019. An ensemble two-stage estimation procedure was proposed, with a generalized additive model (GAM) for temporal-trend removal in the first stage and a random forest model used to assess residual spatiotemporal variations in the second stage. We obtained a model-fitting R² of 0.98 with a root mean square error (RMSE) of 1.40 μg/m3. The leave-one-out cross-validation (LOOCV) R² with seasonal stratification was 0.82, and the RMSE was 3.85 μg/m3, whereas the R² and RMSE obtained by using the pure random forest approach produced R² and RMSE values of 0.74 and 4.60 μg/m3, respectively. The results indicated that the ensemble modeling approach had a higher predictive ability than the pure machine learning method and could provide reliable PM₂.₅ estimates over the entire island, which has complex terrain in terms of land use and topography.

With increasingly stringent regulations on emission criteria and environment pollution concerns, marine fuel oils (particularly heavy fuel oils) that are commonly used today for powering ships will no longer be allowed in the future. Various maritime energy strategies are now needed for the long-term upgrade that might span decades, and quantitative predictions are necessary to assess the outcomes of their implementation for decision support purpose. To address the technical need, a novel approach is developed in this study that can incorporate the strategic implementation of fuel choices and quantify their adequacy in meeting future environmental pollution legislations for ship emissions. The core algorithm in this approach is based on probabilistic simulations with a large sample size of ship movement in the designated port area, derived using a Bayesian ship traffic generator from existing real activity data. Its usefulness with scenario modelling is demonstrated with application examples at five major ports, namely the Ports of Shanghai, Singapore, Tokyo, Long Beach, and Hamburg, for assessment at Years 2020, 2030, and 2050 with three economic scenarios. The included fuel choices in the application examples are comprehensive, including heavy fuel oils, distillates, low sulphur fuel oils, ultra-low sulphur fuel oils, liquefied natural gas, hydrogen, biofuel, methanol, and electricity (battery). Various features are fine-tuned to reflect micro-level changes on the fuel choices, terminal location, and/or ship technology. Future atmospheric pollutant emissions with various maritime energy strategies implemented at these ports are then discussed comprehensively in details to demonstrate the usefulness of the approach.

Satellite-derived aerosol optical depth (AOD) has been widely used to predict ground-level fine particulate matter (PM₂.₅) concentrations, although its utility can be limited due to missing values. Despite recent attempts to address this issue by imputing missing satellite AOD values, the uncertainty associated with the AOD imputation and its impacts on PM₂.₅ predictions have been understudied. To fill this gap, we developed a missing data imputation model for the AOD derived from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) and PM₂.₅ prediction models using several machine learning methods. We also examined how the uncertainty associated with the imputed AOD and a choice of machine learning algorithm were propagated to PM₂.₅ predictions. The application of the proposed imputation model to the data from New York State in the U.S. achieved a superior performance than those related studies, with a cross-validated R² of 0.94 and a Root Mean Square Error of 0.017. We also found that there was considerable uncertainty in PM₂.₅ predictions associated with the use of imputed AOD values, although it was not as high as the uncertainty from the machine learning algorithms used in PM₂.₅ prediction models. We concluded that the quantification of uncertainties for both AOD imputation and its propagation to AOD-based PM₂.₅ prediction is necessary for accurate and reliable PM₂.₅ predictions.

High delicate particulate matter (PM₂.₅) concentration can seriously reduce air quality, destroy the environment, and even jeopardize human health. Accordingly, accurate prediction for PM₂.₅ plays a vital role in taking precautions against upcoming air ambient pollution incidents. However, due to the disturbance of seasonal and nonlinear characteristics in the raw series, pronounced forecasts are confronted with tremendous handicaps, even though for seasonal grey prediction models in the preceding researches. A novel seasonal nonlinear grey model is initially designed to address such issues by integrating the seasonal adjustment factor, the conventional Weibull Bernoulli grey model, and the cultural algorithm, simultaneously depicting the seasonality and nonlinearity of the original data. Experimental results from PM₂.₅ forecasting of four major cities (Shanghai, Nanjing, Hangzhou, and Hefei) in the YRD validate that the proposed model can obtain more accurate predictive results and stronger robustness, in comparison with grey prediction models (SNGBM(1,1) and SGM(1,1)), conventional econometric technology (SARIMA), and machine learning methods (LSSVM and BPNN) by employing accuracy levels. Finally, the future PM₂.₅ concentration is forecasted from 2020 to 2022 using the proposed model, which provides early warning information for policy-makers to develop PM₂.₅ alleviation strategies.