Spatiotemporal graph neural networks for analyzing the influence mechanisms of river hydrodynamics on microplastic transport processes

Microplastic pollution in riverine systems has become a critical environmental concern, with complex hydrodynamic processes governing their transport and fate. This study presents a novel spatiotemporal graph neural network framework to elucidate the influence mechanisms of river hydrodynamics on microplastic transport processes. The methodology integrates graph-based river network representation with multi-scale temporal feature extraction, incorporating physics-informed constraints to ensure prediction consistency with fundamental transport principles. Field validation using microplastic concentration data from multiple monitoring stations demonstrates superior predictive performance, achieving correlation coefficients exceeding 0.89 compared to traditional numerical models (0.6-0.7). The model reduces computational time by approximately 92% while maintaining comparable accuracy. Sensitivity analysis reveals that flow velocity and bed shear stress constitute dominant controls, accounting for 62.9% of concentration variance. The framework effectively captures transport phenomena across multiple spatiotemporal scales, with optimal performance at 1-7 day forecasting horizons and 1-10 km spatial extents. This research provides significant contributions to environmental modeling methodologies and offers valuable capabilities for pollution source identification, real-time monitoring system design, and remediation strategy optimization in river environmental management.