系统区分“异质性、位置、河网拓扑与水文状态交互”的独立贡献

主模型LSTM参考《From RNNs to Transformers: benchmarking deep learning architectures for hydrologic prediction》HESS【文献分享|01】从RNN到Transformer深度学习架构在水文预测中的基准测试数据大样本SPAT分布式分析CAMELS-SPAT的流域属性与气象北美水文研究的溪流观测、强迫数据和地理空间数据 |FRDR-DFDRUSGS日径流作为观测目标NLDAS-2日气象作为动态强迫NLDAS-2 Noah土壤湿度、ET和SWE作为状态信息CAMELS静态属性表征流域差异DEM及河网计算坡度、TWI、水文距离和方向关系仅保留日尺度不重新引入小时径流建模。空间异质性研究[1]HUANG Y F, TSANG Y. Unveiling how rainfall spatial structure and displacement drive peak flow magnitude and onset in small mountainous watersheds[J]. Journal of Hydrology, 2025, 662: 134069.[2]WU Y, YIN X, ZHOU G, 等. Rising rainfall intensity induces spatially divergent hydrological changes within a large river basin[J]. Nature Communications, 2024, 15(1): 823-837.