Estimating marine PM2.5 concentrations and exploring the drivers over the eastern China seas using multi-source data and machine learning

Fine particulate matter (PM2.5) plays an important role in marine ecosystems and regional climate processes. However, accurate monitoring of PM2.5 over marine environments remains challenging due to sparse ground observations and the complex interactions among meteorological conditions, atmospheric chemistry, and pollutant transport. To address this issue, a multi-source data framework for estimating PM2.5 concentrations over the eastern China seas was developed by integrating aerosol optical depth (AOD), meteorological variables, and atmospheric chemical parameters (NO2, CO, O3, and SO2). Three machine learning algorithms, including Back Propagation Neural Network (BPNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), were evaluated using sample-based, spatial-based, and temporal-based validation strategies. Results show that XGBoost demonstrated competitive performance among the three models across the different validation strategies, suggesting that the proposed framework is applicable to marine PM2.5 estimation. Based on the XGBoost-derived PM2.5 estimates, the spatio-temporal variations and dominant influencing factors of marine PM2.5 from 2019 to 2024 were further examined. The estimated PM2.5 concentrations showed a clear decreasing gradient from west to east and from north to south, with high-value areas mainly located in Bohai Bay and the Shandong coastal regions, indicating a strong influence from terrestrial emissions and coastal pollutant transport. Temporally, PM2.5 showed a fluctuating downward trend and a stable seasonal cycle, with winter maxima and summer minima. The marked decline in 2020 further reflected the response of marine PM2.5 to changes in inland anthropogenic emissions. SHapley Additive exPlanations (SHAP) analysis showed that AOD dominated PM2.5 variations in spring, summer, and autumn, whereas reduced boundary layer height, enhanced northwesterly transport, and increased pollution-related gases such as NO2 and CO jointly promoted wintertime PM2.5 accumulation. These findings demonstrate the value of integrating atmospheric chemical parameters into marine PM2.5 estimation and provide insights into air pollution mechanisms over the eastern China seas.