Improving short-term forecasting of surface dust concentration in Northern China: Integrating machine learning with multi-numerical models

Abstract

Sand and dust storms (SDSs) are among the most significant extreme weather and climate events impacting northern China, exerting substantial influences on global climate change, the ecological environment, socioeconomic systems, and public health. Traditional dust numerical models often exhibit considerable uncertainty due to the chaotic nature of the atmosphere, inaccuracies in initial conditions, and challenges in parameterizing physical processes. Therefore, a multi-model ensemble forecasting model (ML-SDC) for surface dust concentration was developed based on machine learning with multi-numerical models across northern China. The results demonstrate that the ML-SDC model exhibits significant improvements over single dust numerical models, traditional ensemble methods, and individual machine learning models during the 0–72 h forecast period with the average correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) for surface dust concentration reached 0.78, 91.49 μg/m³ and 36.91 μg/m³ respectively. Additionally, the ML-SDC model has a strong spatiotemporal correction ability for dust concentration, dispersion, and transport. This finding enhances the accuracy of short-term forecasts for extreme weather, offering a valuable tool for the identification and quantitative forecasting of dust weather, while supporting improved preparedness and mitigation strategies for SDS-related impacts and advancing research in climate modeling, air quality management, and environmental sustainability.