Underlying Surface Heterogeneity Modulates Meteorological Drivers of PM2.5 and O3 Pollution in the Sichuan Basin: Insights from Explainable Machine Learning.

Air pollution in topographically complex basin regions presents major challenges to air quality management due to strong interactions among heterogeneous surfaces, local meteorology, and pollutant accumulation. Characterized by enclosed terrain and frequent occurrences of weak ventilation and temperature inversion conditions, the Sichuan Basin presents a unique environment where underlying surface heterogeneity critically modulates local pollution processes. This study established a dual-dimensional underlying surface classification system integrating geomorphology and land use, initially classifying monitoring sites into eight surface categories based on terrain and urbanization characteristics and further refining them into fifteen underlying surface types. The spatial patterns of pollution and their meteorological drivers were systematically investigated using the XGBoost-Deweather-SHAP framework. Results showed that urban-related underlying surfaces generally exhibited higher pollution frequencies, with the Hill-Urban (HU2) showing high frequencies of PM2.5 and O3 pollution events, exceeding 29% and 27%, respectively. River Valley-Urban (RVU1) experienced the highest PM2.5 concentrations whereas Plain-Urban (PU2) exhibited the most severe O3 pollution. Meteorological influences exhibited strong dependence on underlying surface conditions and pollutant-specific characteristics. PM2.5 variability was generally dominated by temperature across most underlying surface categories, while relative humidity and boundary layer height showed stronger influences in specific river valley and complex terrain environments. In contrast, O3 exhibited a distinct response pattern characterized by temperature-dominated conditions in plain areas and humidity-dominated conditions in river valley environments. These findings demonstrate that underlying surface heterogeneity plays an important role in modulating meteorological sensitivities of air pollutants and provide a scientific basis for terrain-specific air quality management strategies in complex basin regions.