pA novel method for hydrology and water quality simulation in karst regions using machine learning model

Abstract

Simulating hydrology and water quality in karst regions remains a significant challenge due to the complex and unique geological conditions. This study developed a novel conceptual model integrating baseflow segmentation and rainfall-stormflow analysis for hydrological and water quality simulation and implemented it using an interpretable machine learning approach. Two representative watersheds in the karst region of southwestern China—the Chishui river and Nanpan river—were selected as case studies. The results demonstrated that (1) The proposed method successfully achieved comprehensive simulations of streamflow, CODMn, TN, and TP. The maximum prediction time span was 3 days for streamflow (R2 > 0.60) and 2 days for CODMn, TN, and TP (R2 > 0.59). (2) Rainfall plays a significant positive driving role in hydrological and water quality simulations, effectively weakening the strong autocorrelation of these parameters in karst regions. The increase of baseflow and stormflow worsens water quality, which is most significant for CODMn and TP in Chishui river basin. (3) Water quality in ridge-furrow valleys, fractured basins, and fractured mountainous areas is more significantly influenced by rainfall due to the unique geological features. The proposed method can be effectively applied in karst areas with non-point source pollution and directly supports water environment prediction and identification of key pollution source areas.

Keywords

• Baseflow segmentation
• Rainfall-stormflow analysis
• Simulations
• Hydrology
• River water quality
• Machine learning models