Transferred Long Short-Term Memory Network for River Flow Forecasting in Data-Scarce Basins

Hydrological models have made significant advances in methodologies and applications in recent years. However, there is still a need to address the challenge of modeling in areas with limited or no data. This study proposes a transferred Long Short-Term Memory (T-LSTM) network based on transfer lear...

Full description

Saved in:
Bibliographic Details
Published in:Water resources management Vol. 39; no. 9; pp. 4493 - 4507
Main Authors: Xie, Zaichao, Xu, Wei, Zhu, Bing, Yin, Shiming, Yang, Yi, Li, Xiaojie, Wang, Sufan
Format: Journal Article
Language:English
Published: Dordrecht Springer Netherlands 01.07.2025
Springer Nature B.V
Subjects:
Algorithms
Artificial intelligence
Atmospheric Sciences
Basins
China
Civil Engineering
Earth and Environmental Science
Earth Sciences
Environment
Geotechnical Engineering & Applied Earth Sciences
Hydrogeology
Hydrologic data
Hydrologic models
Hydrology
Hydrology/Water Resources
Hydrometeorology
Interbasin transfers
K-nearest neighbors algorithm
Long short-term memory
Machine learning
Networks
Neural networks
Performance evaluation
Precipitation
River basins
River flow
River forecasting
Rivers
Soil and Water Assessment Tool model
Stream flow
Time series
Topography
Transfer learning
water
Water resources
China
SWAT
LSTM
Hydrological modeling
Transfer learning
K-nearest neighbor
ISSN:0920-4741, 1573-1650
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Hydrological models have made significant advances in methodologies and applications in recent years. However, there is still a need to address the challenge of modeling in areas with limited or no data. This study proposes a transferred Long Short-Term Memory (T-LSTM) network based on transfer learning and Long Short-Term Memory (LSTM) networks to address this issue. Firstly, the K-nearest neighbor (K-NN) algorithm is used to estimate precipitation data, while the Soil and Water Assessment Tool (SWAT) is applied to generate long series of flow data for training. Secondly, four transfer learning scenarios, classified into intra-basin transfer and inter-basin transfer, are constructed based on the simulated and observed data. Finally, T-LSTM networks are constructed with different transfer learning scenarios and the performance of the networks is evaluated in five river basins in China, Hunjiang, Jialingjiang, Wujiang, Minjiang and Jinshajiang. The results indicate that inter-basin T-LSTM networks perform exceptionally well in data-scarce basins, particularly those with similar hydrometeorological and basin characteristics.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
content type line 23
ISSN:0920-4741
1573-1650
DOI:10.1007/s11269-025-04165-y