Water quality assessment in a large plateau lake in China from 2014 to 2021 with machine learning models: Implications for future water quality management.

Amid the global surge of eutrophication in lakes, investigating and analyzing water quality and trends of lakes becomes imperative for formulating effective lake management policies. Water quality index (WQI) is one of the most used tools to assess water quality by integrating data from multiple water quality parameters. In this study, we analyzed the spatio-temporal variations of 11 water quality parameters in one of the largest plateau lakes, Erhai Lake, based on surveys from January 2014 to December 2021. Leveraging machine learning models, we gauged the relative importance of different water quality parameters to the WQI and further utilized stepwise multiple linear regression to derive an optimal minimal water quality index (WQImin) that required the minimal number of water quality parameters without compromising the performance. Our results indicated that the water quality of Erhai Lake typically showed a trend towards improvement, as indicated by the positive Mann-Kendall test for WQI performance (Z = 2.89, p < 0.01). Among the five machine learning models, XGBoost emerged as the best performer (coefficient of determination R2 = 0.822, mean squared error = 3.430, and mean absolute error = 1.460). Among the 11 water quality parameters, only four (i.e., dissolved oxygen, ammonia nitrogen, total phosphorus, and total nitrogen) were needed for the optimal WQImin. The establishment of the WQImin helps reduce cost in future water quality monitoring in Erhai Lake, which may also serve as a valuable framework for efficient water quality monitoring in similar waters. In addition, the elucidation of spatio-temporal patterns and trends of Erhai Lake's water quality serves as a compass for authorities, offering insights to bolster lake management strategies in the future.