A zone-specific water quality modelling framework for medium and large lakes using interpretable machine learning

The significant spatial heterogeneity of lake characteristics poses a persistent challenge for accurate prediction and effective management of medium and large lake water quality. In this study, we developed an interpretable machine learning (ML)-based zone-specific modelling framework for medium and large lakes to improve the prediction of key water quality parameters and the understanding of their dynamic changes. The proposed framework was applied to predict total nitrogen (TN) and total phosphorus (TP) concentration distribution and identify their change drivers in a large eutrophic lake (i.e., Yangcheng Lake, China), using two distinct modelling strategies (i.e., aggregated and segregated) and three ML algorithms (i.e., Extreme Gradient Boosting (XGBoost), Random Forest (RF) and Support Vector Regression (SVR)). Our results suggested that XGBoost algorithm applied with the aggregated modelling strategy effectively captured cross-zone pollutant transport and achieved optimal water quality prediction performance across all lake zones. Air temperature was identified as the most important variable for TN prediction, while water level was the key variable for TP prediction, across all lake zones. The inputs from three tributaries (i.e., Jiejing, Beihejing, and Nanxiaojing) and endogenous load were also found as the primary influence factors of nutrients in the western and middle lake zones. In the eastern zone, the influence of tributary inputs on lake nutrients was significantly weakened, with endogenous load playing a more prominent role. The proposed framework is applicable to water quality prediction and pollution management in other medium and large lakes, providing a promising approach on modelling spatial heterogeneity in lake water quality.