Improved water quality assessment and prediction for small watersheds in human settlements of the Chengdu plain, southwestern China.

Water is a critical natural resource for maintaining the stability of Earth's ecosystems and the sustainable development of human society and economy. As the basic unit of terrestrial hydrological cycles, small watersheds have water quality conditions that significantly affect residents' drinking water safety, agricultural irrigation quality, and regional hydrological cycle systems. This study focused on the hydrochemical features and formation processes of water bodies, proposing an innovative framework that combined advanced evaluation with predictive modeling to assess drinking water quality in a small watershed within the human settlements of the Chengdu Plain. The results showed that the surface water samples were generally weakly alkaline fresh water, dominated by the Ca-HCO₃ hydrochemical type, with inapparent seasonal variation. The weathering and dissolution of rock minerals were the main controlling factors for the hydrochemical composition. In addition, NO₃- was also affected by human activities, such as agriculture. The improved water quality assessment results, which integrated the analytic hierarchy process (AHP) and entropy weight method (EWM) using the game theory (GT), indicated that the average drinking water quality index (WQI) values of the four phases of water samples were 23.77, 20.82, 16.87, and 23.08, respectively. All samples were of excellent water quality, suitable for domestic use and as drinking water sources. Furthermore, extending the paradigm from static assessment to dynamic prediction. Multiple machine learning methods were used to predict water quality. Among these, the multiple linear regression (MLR) model performed best, with a coefficient of determination (R2) of 0.996, a root mean square error (RMSE) of 0.184, and a mean absolute error (MAE) of 0.126 on the test set. Lasso and Ridge regression identified pH, TDS, and NO3- as the core indicators for prediction. These findings, derived from an integrated assessment and predicting framework, contribute to the scientific exploitation and utilization of watershed water resources and the formulation of related strategies.