Deciphering Eutrophication-Limiting Nutrients in Lakes: A Multiscale Analysis of Chinese Waters

The total nitrogen/total phosphorus (TN/TP) ratio is considered a valuable indicator for evaluating the abundance of phytoplankton and the eutrophic condition of a water body, but its effectiveness as an indicator of eutrophication at different watershed scales has not been fully explored. In this study, we collected data from 103 lakes within four major watersheds in China and utilized the machine learning models eXtreme Gradient Boosting (XGBoost) and k-nearest neighbors (KNN) to predict the TN/TP ratio at three different scales. We identified notable disparities in the TN/TP ratio, chlorophyll a concentration, and algal cell density across the three scales. By incorporating time as an input variable, we were able to capture temporal trends in TN/TP ratio, which enhanced the predictive accuracy and fit of the machine learning models. The optimization ratios of the model indicators' coefficient of determination, root-mean-square error, and mean absolute percentage error at three scales are 35.71 ± 25.26%, 0.43 ± 0.17%, and 1.47 ± 1.19%, respectively. XGBoost demonstrated a higher accuracy and better fit than KNN. Our results reveal the substantial impact of the watershed scale on predicting eutrophication-limiting nutrients of water bodies.