Spatiotemporal Prediction Algorithm for Groundwater Quality Under Multi-Indicator Coupling Constraints

Abstract

Spatiotemporal prediction of groundwater quality is of great significance for regional water environmental safety assessment, pollution risk identification, and urban groundwater resource management. To address the difficulty of existing methods in simultaneously characterizing multi-indicator coupling relationships, temporal evolution processes, and spatial heterogeneity, this study proposes a spatiotemporal groundwater quality prediction model under multi-indicator coupling constraints. First, indicators including dissolved oxygen, total nitrogen, electrical conductivity, dissolved organic carbon, pH, permanganate index, and total phosphorus are uniformly mapped into a risk space to construct an integrated groundwater quality risk index. Then, based on monthly groundwater monitoring data from Yiyang City during 2000–2023, continuous regional grid sequences are generated. In terms of model design, the Temporal Difference Interaction Module (TDIM) is introduced to enhance multi-scale temporal variation modeling, Region-Guided Feature Modulation (RGFM) is used to strengthen regional heterogeneity representation, and Spatiotemporal Boundary-Aware Loss (STB Loss) is adopted to maintain spatiotemporal boundary consistency. The experimental results show that the proposed method achieves a Structural Similarity Index Measure (SSIM) of 0.9814±0.0085, a Peak Signal-to-Noise Ratio (PSNR) of 40.47±2.19, a Mean Absolute Error (MAE) of 2.80×10−3±1.50×10−3, and a Root Mean Square Error (RMSE) of 9.70×10−3±2.69×10−3, outperforming comparison models overall and providing effective support for dynamic groundwater quality prediction and water environmental safety assessment.