An hourly and localized optimization method for soil fugitive dust emission inventory based on machine learning

Abstract

Soil fugitive dust (SFD) is characterized by a variety of sources and considerable spatial-temporal variability, exerting a significant impact on environmental air quality and ecological systems in cities across northern China. Multiple factors can shape SFD emission. Nevertheless, the current comprehension of its critical impact factors and quantitative methodologies remains constrained. This study utilizes interpretable machine learning techniques to identify the principal impact factors of SFD and their interactions while delineating their action thresholds. The findings reveal seasonal variations in impact factors and emphasize the substantial effect of bare soil source strength on SFD, including parameters such as bare soil area and soil moisture. Consequently, the Wind Erosion Equation model is optimized following these findings to localize its parameters and improve its capability to calculate hourly SFD emissions. The case application is validated using observational data, demonstrating the reliability and precision of the optimized methodology. This study provides insights and solutions for the local optimization of SFD parameterization schemes and further supports the formulation of precise prevention and control policies for SFD.