Spatial-Temporal Modeling of Urban Water Quality Using Gradient Boosting and H2O AutoML Techniques

Considering the increasingly serious water pollution problem, this study focuses on the monitoring and prediction of urban water environment pollution. The study analyzed a set of daily water quality sample data released by the United States Geological Survey and proposed a series of urban water pollution prediction models that integrate temporal and spatial characteristics. Experimental results show that H2O AutoML achieved the lowest prediction error performance on the test set. The Gradient Boosting model significantly improves the prediction accuracy after adding the time dimension information. This study explored the significant effects of dissolved oxygen, water temperature and electrical conductivity on water quality indicators through feature association analysis, spatial distribution analysis and comparison of models in different dimensions, and proposed a method to incorporate time dimension information into the model to improve prediction accuracy. The research results provide an accurate early warning tool for urban water quality management and a scientific basis for optimizing resource allocation in water pollution management, which has important practical value and guiding significance.