Forecasting high-resolution PM2.5 concentrations in southeastern China by combining high-resolution satellite data and numerical simulation with machine learning

PM_2.5 is a significant air contaminant that presents a serious risk to human health. Accurate PM_2.5 forecasts with high spatial resolution are essential for decision-makers to implement effective mitigation strategies and prevent harmful public exposure to PM_2.5. Current methods often rely on spatial data of limited precision, like outputs from spatial interpolation and chemical transport models (CTMs), resulting in PM_2.5 forecasts that either have inaccurate spatial patterns or completely omit spatial details. For this research, we developed a novel approach to demonstrate the feasibility of employing 1 km satellite AOD data to generate 1 km resolution PM_2.5 forecasts in southeastern China up to five days in advance by integrating machine learning models, CTM simulations, and 1 km resolution satellite AOD measurements. Our forecast framework integrated with satellite AOD data demonstrated superior performance, surpassing the precision of the original CTM forecast data, as evidenced by both spatial cross-validation and overall validation results. In addition, incorporating satellite AOD into the forecasting model could enhance the spatial resolution of PM_2.5 forecasts. The model enables the production of PM_2.5 forecasts featuring both accurate spatial representation and high spatial resolution.