大尺度跨境流域径流预测的迁移学习框架--敏感性分析及在数据稀缺流域的适用性

作     者：马凯 申朝鹏 许紫月 何大明 MA Kai;HEN Chaopeng;XU Ziyue;HE Daming

作者机构：Institute of International Rivers and Eco-securityYunnan UniversityKunming 650091China Yunnan Key Laboratory of International Rivers and Transboundary Eco-securityYunnan UniversityKunming 650091China Civil and Environmental EngineeringPennsylvania State UniversityUniversity ParkPAUnited States 

出 版 物：《Journal of Geographical Sciences》 (地理学报（英文版）)

年 卷 期：2024年第34卷第5期

页      面：963-984页

核心收录：

学科分类：08[工学] 081501[工学-水文学及水资源] 0815[工学-水利工程] 0705[理学-地理学] 

基　　金：National Key Research and Development Program of China,No.2022YFF1302405 National Natural Science Foundation of China,No.42201040 The National Key Research and Development Program of China,No.2016YFA0601601 The China Postdoctoral Science Foundation,No.2023M733006。 

主　　题：transfer learning streamflow prediction deep learning model sensitivity data scarcity international river 

摘      要：The imbalance in global streamflow gauge distribution and regional data scarcity,especially in large transboundary basins,challenge regional water resource management.Effectively utilizing these limited data to construct reliable models is of crucial practical im-portance.This study employs a transfer learning(TL)framework to simulate daily streamflow in the Dulong-lrrawaddy River Basin(DIRB),a less-studied transboundary basin shared by Myanmar,China,and India.Our results show that TL significantly improves streamflow pre-dictions:the optimal TL model achieves an average Nash-Sutcliffe efficiency of 0.872,showing a marked improvement in the Hkamti sub-basin.Despite data scarcity,TL achieves a mean NSE of 0.817,surpassing the 0.655 of the process-based model MIKE SHE.Addition-ally,our study reveals the importance of source model selection in TL,as different parts of the flow are affected by the diversity and similarity of data in the source model.Deep learning models,particularly TL,exhibit complex sensitivities to meteorological inputs,more accu-rately capturing non-linear relationships among multiple variables than the process-based model.Integrated gradients(IG)analysis furtherillustrates TL s ability to capture spatial het-erogeneity in upstream and downstream sub-basins and its adeptness in characterizing dif-ferent flow regimes.This study underscores the potential of TL in enhancing the under-standing of hydrological processes in large-scale catchments and highlights its value for wa-ter resource management in transboundary basins under data scarcity.