Hydrological controls on deep learning streamflow transferability across contrasting basins

Study region The Yuan River basin, a major tributary of the middle Yangtze River in China, is used as the source basin for model development, and the model's transferability is further evaluated across global hydrological zones. Study focus Accurate streamflow hindcasting using deep learning (DL) is essential for water resources management, yet transferring DL models from local to global scales remains a critical challenge. This study developed a multi-strategy ensemble DL framework that integrates feature selection, time-varying filtering-based empirical mode decomposition, Newton-Raphson-based optimization, and the CNN–BiLSTM–attention network to hindcast daily streamflow and diagnose cross-basin transferability under contrasting hydroclimatic conditions. New hydrological insights for the region In the source basin, the proposed framework achieved robust and statistically consistent improvements over the ML/DL benchmarks. Transfer experiments across 3486 basins showed that the modeling workflow performed most effectively in hydrologically compatible regions, including saturation-excess basins, snowmelt- or permafrost-influenced regimes, and catchments with coherent seasonal runoff patterns. Poor transferability to arid infiltration-excess basins and dryland systems with transmission losses highlights the heterogeneous runoff mechanisms that disrupt rainfall-runoff relationships between source and target basins. These results indicate that cross-basin DL transfer is associated with runoff generation similarity, rainfall spatial organization, regime regularity, and anthropogenic influences, rather than neural architecture alone, thereby informing water resources management across diverse global basins.

Authors

Institutions

Publication Details

Journal
Journal of Hydrology Regional Studies
Published
2026-11-13
DOI
https://doi.org/10.1016/j.ejrh.2026.104210
Primary Topic
Hydrological Forecasting Using AI
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Hydrological controls on deep learning streamflow transferability across contrasting basins

Yangxiaoyue Liu, Dianchang Wang, Guangrong Cao, Hou Jiang et al.
Journal of Hydrology Regional Studies
Hydrological Forecasting Using AI
article

Hydrological controls on deep learning streamflow transferability across contrasting basins

Yangxiaoyue Liu, Dianchang Wang, Guangrong Cao, Hou Jiang, Chaowei Xu, Boyang Jia, Yunpeng Zhao, Yating Zhu, Yukun Hou, Tengfei Huang, Yasong Chen
article en

Abstract

Study region The Yuan River basin, a major tributary of the middle Yangtze River in China, is used as the source basin for model development, and the model's transferability is further evaluated across global hydrological zones. Study focus Accurate streamflow hindcasting using deep learning (DL) is essential for water resources management, yet transferring DL models from local to global scales remains a critical challenge. This study developed a multi-strategy ensemble DL framework that integrates feature selection, time-varying filtering-based empirical mode decomposition, Newton-Raphson-based optimization, and the CNN–BiLSTM–attention network to hindcast daily streamflow and diagnose cross-basin transferability under contrasting hydroclimatic conditions. New hydrological insights for the region In the source basin, the proposed framework achieved robust and statistically consistent improvements over the ML/DL benchmarks. Transfer experiments across 3486 basins showed that the modeling workflow performed most effectively in hydrologically compatible regions, including saturation-excess basins, snowmelt- or permafrost-influenced regimes, and catchments with coherent seasonal runoff patterns. Poor transferability to arid infiltration-excess basins and dryland systems with transmission losses highlights the heterogeneous runoff mechanisms that disrupt rainfall-runoff relationships between source and target basins. These results indicate that cross-basin DL transfer is associated with runoff generation similarity, rainfall spatial organization, regime regularity, and anthropogenic influences, rather than neural architecture alone, thereby informing water resources management across diverse global basins.

Journal of Hydrology Regional StudiesVol. 68
China Three Gorges Corporation (China) (CN), Chinese Academy of Sciences (CN), Peking University (CN), Institute of Geographic Sciences and Natural Resources Research (CN), State Key Laboratory of Resources and Environmental Information System (CN)
China Three Gorges Corporation, National Natural Science Foundation of China, Natural Science Foundation of Hubei Province
Clean water and sanitation
Hydrological Forecasting Using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon