Acid rain prediction in the Guangdong-Hong Kong-Macao Greater Bay Area using an explainable machine learning framework

Abstract

Acid rain, characterized by pH values lower than 5.6, is a critical natural disturbance of ecosystems, which threatens the sustainability of ecosystems, agriculture, and human society worldwide. However, accurately quantifying the driving factors of acid rain remains challenging due to a changing environment of significant spatial heterogeneity. Here, we established an explainable machine-learning framework (MLF) using 19 meteorological, air pollutant, and land surface variables as model input to construct the pH values of acid rain across the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) during 2006–2021. The MLF includes Extreme Gradient Boosting (XGBoost) for acid deposition prediction and a SHapley Additive exPlanations method (SHAP) for interpreting factor importance. The results indicated that the observed increases in pH values of acid rain are predominantly controlled by the significant decreases in maximum daily sulfur dioxide (SO2) concentration of air across GBA, with its relative contribution ranging from 16.2% to 31.9% for each city. Changes in the urbanization rate and the proximity to the coast also play significant roles in predicting the pH values of acid rain. Meteorological variables typically have minimal impact on acid rain predictions, with their contribution generally being less than 5%, indicating the complex physical process of acid rain generation. This study enhanced our comprehension of the spatial variability of acid rain drivers across a highly developed region, providing valuable insights and case studies for regions worldwide that frequently experience acid rain.