Multi-parameter watershed water quality level prediction based on integrated algorithms

Abstract In the context of global water scarcity and increasing water pollution, accurate water quality assessment and prediction is crucial for water resources management and pro-tection. To address the shortcomings of traditional water quality assessment and pre-diction methods, this study constructed a multi-parameter watershed water quality level prediction model based on an integrated algorithm, which utilized principal component analysis (PCA), C4.5 decision tree, BP neural network, convolutional neural network (CNN), and long-short-term memory network (LSTM) to predict water quality classification. A total of 31296 samples were collected from 23 monitoring stations in a region of China from May to October 2023, covering nine key water quality indicators. After PCA dimensionality reduction, the data were input into the prediction models. The results show that the PCA-C4.5 decision tree model has a classification prediction accuracy of 88.13\%; the PCA-BP neural network model has an overall accuracy of 94.53\%, with excellent precision, recall and F1 value in categories 3 and 4; the PCA-CNN model has an accuracy of 93.65\%, with high precision in categories 1 and 6; and the PCA-LSTM model is the best model, with an accuracy of 94.87\%. The PCA-LSTM model has the best performance with an accuracy of 94.87\%, and its recognition ability is outstanding as its precision and recall are over 94\% for categories 3 and 4. This study confirms the feasi-bility of integrating algorithms for water quality prediction and provides a new path for dynamic monitoring of water quality in watersheds. In the future, we can incorporate transfer learning or attention mechanism to optimize the recognition ability of the model for small-sample categories, and explore the synergy between multi-source re-mote sensing and ground monitoring data to improve the generalization and timeliness of the model.