Temporal and spatial feature extraction using graph neural networks for multi-point water quality prediction in river...

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Traditional water quality prediction often overlooks spatial pollutant dynamics, focusing on individual sites.
• Deep learning excels at capturing nonlinear relationships but struggles with distributed monitoring data.

Purpose of the Study:

• To develop a novel deep learning model that integrates spatial and temporal features for accurate water quality prediction.
• To address the limitations of existing methods in capturing pollutant spatial dynamics across monitoring stations.

Main Methods:

• Proposed a Spatio-Temporal Feature Graph Neural Network (STF-GNN) integrating graph convolutional networks (GCN), gated recurrent units (GRU), and self-attention.
• Represented monitoring stations as graph nodes to model multi-scale spatiotemporal dependencies.
• Utilized multivariate time series data for training and validation.

Main Results:

• Achieved superior performance in dissolved oxygen (DO) and total nitrogen (TN) prediction with RMSE values of 0.233 (DO) and 0.033 (TN).
• Demonstrated robust generalization capabilities through cross-basin validation, with maximum relative errors below 0.639 (DO) and 0.606 (TN).
• Showcased strong anti-interference ability with 88% peak-valley synchronization at untrained stations.

Conclusions:

• The STF-GNN model effectively captures spatiotemporal dependencies for improved water quality prediction.
• Both spatial and temporal feature extraction are critical for enhancing predictive performance.
• The study provides a framework for spatially-aware water quality prediction, supporting environmental monitoring.