Multiple Machine Learning Methods with Correlation Analysis for Short-Term River Water Quality Prediction

Water quality prediction and early warning are crucial aspects of water quality monitoring and management. These predictions play a pivotal role in river pollution control. In this study, we focus on a typical river watershed in Nanjing, China. We collect and process historical monitoring data from multiple automated water quality stations within the basin. To predict water quality, we employ three time series models: Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory Network (LSTM). The primary water quality parameters exceeding standards in the basin are ammonia nitrogen and total phosphorus. Notably, the predicted values of ammonia nitrogen and total phosphorus exhibit a strong correlation with their historical counterparts, while showing weaker associations with other water indicators. Among the three machine learning models, LSTM demonstrates superior predictive performance for total phosphorus, while RF excels in predicting ammonia nitrogen.