Answers are grounded in this paper and cite their sources.
River networks contribute to the spatial organisation of water, sediment, and ecological connectivity across landscapes. As global hydrographic datasets remain incomplete and basin monitoring needs to grow, there is increasing demand for methods that can perform network reconstruction. This study evaluates graph neural networks (GNNs) for two reach-level tasks on four global river basins, Amur, Ganga, Mekong, and Mississippi: directed link prediction for network reconstruction and Strahler order classification. Four architectures were compared: GraphSAGE, graph attention network (GAT), heterogeneous GAT (Hetero-GAT), and a proposed weighted directed MixHop model. The proposed model outperforms the baselines in most link prediction settings and in all basins for Strahler classification, achieving balanced accuracy of approximately 92.8–94.8% at low sparsity (10% missing edges) for link prediction and macro-F1 of approximately 90.6–93.6% for Strahler classification across basins. Overall, the results demonstrate that topology-aware GNNs can extract hydrologically meaningful patterns from river networks and offer a promising approach to reconstructing incomplete hydrographic connectivity and supporting basin-scale stream classification.{River network Graph Neural Network Network reconstruction Hetero-GAT GraphSAGE}
No takes yet. Share an insight, caveat, or question.
V et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: