Early warning study of eutrophication in the Ulansuhai Lake based on Spearman and random forests

Timely and accurate prediction of water eutrophication is of great significance to lake management, but the traditional early warning model of water eutrophication is difficult to meet the demand for accurate monitoring of the eutrophication status of water bodies, and the prediction accuracy still needs to be improved. This paper collects the water quality data of Ulansuhai Lake from 2011 to 2020, including water temperature, chemical oxygen demand, total nitrogen, total phosphorus, and chlorophyll a concentration and other key indicators, in which chlorophyll a characterizes the eutrophication of the water body, explores the temporal and spatial patterns of chlorophyll a concentration, and constructs a water body eutrophication early warning model integrating the Spearman's correlation analysis and the Random Forest, and then provides an early warning model for the water body eutrophication of the lake. A model integrating Spearman's correlation analysis and random forest was constructed to predict the eutrophication status of water bodies in the Ulansuhai Lake, and the accuracy and stability of the model were verified by using various assessment indexes such as accuracy, recall, F1 value, ROC curve and confusion matrix. The results showed that the chlorophyll a concentration in the Ulansuhai Lake fluctuated significantly from 2011 to 2020, and the annual average value showed a trend of “decreasing-rising-decreasing”. The accuracy of the model in the training set was 95.89%, and it was able to accurately distinguish between eutrophication and non-eutrophication states (category 1 and category 0). A total of 80 category 0 judgments and 16 category 1 warnings occurred in the test set, of which 90 matched the actual water body status, with a test accuracy of 93.75%, a recall rate of 85.71%, and an F1 value of 0.8, which proved the model's ability to provide early warnings in real-world scenarios, and the confusion matrix and ROC curve analyses further verified the model's efficiency and reliability. The model was also tuned with hyper-parameters during the training of the training set, and the average accuracy of cross-validation of the model reached 91.32%. The feature importance analysis shows that the water quality indicators such as ammonia nitrogen and total phosphorus have an important influence in the prediction of eutrophication status of water bodies. This study provides an effective early warning tool for accurate monitoring of water body eutrophication and lake management, and has practical application value.