Machine learning driving water suitability assessment of the Yarlungzangbo River, southern Tibetan Plateau.

The accessibility of surface water renders it vulnerable to contamination, particularly in alpine rivers characterized by high environmental sensitivity and eco-fragility. The Yarlungzangbo River (YR) is the largest-scale regional catchment in the southern Tibetan Plateau. River water quality is becoming an imperative issue that needs to be ascertained with increased human activities in recent decades. In this study, natural background levels (NBLs), novel water quality indices, and machine learning approaches were used to comprehensively evaluate the suitability of alpine surface water. The results of NBLs have revealed that the discrepancy of each hydrochemical indicator resulted from both natural processes and human activities. The novel irrigation water quality index (IWQI) has demonstrated that 64% in the upper reach (UR), 85% in the middle reach (MR), and 48% in the lower reach (LR) are deemed suitable for agricultural irrigation. Notably, the concentration of Na+ holds a predominant influence on the IWQI in both UR (7.82%) and LR (10.01%). Conversely, in MR, the Sodium Adsorption Ratio emerges as the primary factor, contributing 4.53% to the IWQI. The novel drinking water quality index (DWQI) demonstrated that 95% of samples are excellent for drinking. NO3- concentration is the most influential factor determining the DWQI in UR (9.63%), MR (8.25%), and LR (8.96%). Among four machine learning models, the convolutional neural network (CNN) was recognized as the robust algorithm to predict the IWQI (R2=0.9358) and DWQI (R2=0.9993) in the YR catchment. The research findings indicate that anthropogenic activities significantly contribute to water quality degradation in MR. This research presents an extensive suitability assessment of a typical alpine catchment at a large scale, serving as valuable information for water management in the Tibetan Plateau (Asian water tower).