Abstract: To mitigate potential error rates in traditional CNN-LSTM models for water quality prediction, this study proposes an enhanced CNN-LSTM water quality prediction method incorporating the Prophet model and Principal Component Analysis (PCA). In the data preprocessing phase, the Prophet model is employed for outlier detection and handling of water quality monitoring data. PCA is then utilized to reduce the dimensionality of influencing variables and eliminate variable correlations. The processed results serve as input for the CNN-LSTM model to predict the total nitrogen index of water quality. Experimental validation of the proposed method demonstrates significant improvements over the standard CNN-LSTM model across three evaluation metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Squared Error (MSE). Specifically, the proposed method achieved a 13% reduction in MSE, a 6.7% decrease in RMSE, and a 5.6% improvement in MAE. These results highlight the effectiveness of integrating Prophet and PCA techniques with CNN-LSTM for enhancing water quality prediction accuracy and reliability.