Water environment risk prediction method based on convolutional neural network-random forest

The accelerated processes of urbanization and industrialization globally have resulted in an increased risk to aquatic environments, posing a significant threat to the sustainable management of water resources and the health of ecosystems. Accurate prediction of water environmental risks is crucial for the prompt identification of potential pollution sources, the safeguarding of water resources, the maintenance of ecological balance, and the support of environmental policy formulation. This study introduces an innovative prediction methodology that integrates the spatial feature extraction capabilities of Convolutional Neural Networks (CNN) with the multivariate data analysis strengths of Random Forest (RF), aiming to enhance the accuracy and applicability of water environmental risk predictions. The results demonstrate that the proposed prediction method enhances the coefficient of determination (R2) performance by 5.8 %, reduces the Mean Absolute Error (MAE) by 21.5 %, decreases the Mean Bias Error (MBE) by 41.5 %, and lowers the Root Mean Square Error (RMSE) by 56.82 %. Furthermore, this study incorporates surface water data from Henan Province for practical application, merging the prediction results with satellite imagery to facilitate intuitive visualization of water environmental risks, thereby enhancing decision-makers' comprehension and response capabilities regarding complex environmental data. This research not only presents a novel methodology for predicting water environmental risks but also elucidates the evolving trends in such risks.