Zhang et al. (2025) Long Short-Term Memory (LSTM) Based Runoff Simulation and Short-Term Forecasting for Alpine Regions: A Case Study in the Upper Jinsha River Basin

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-30
• Authors: Feng Zhang, Jiajia Yue, Chun Zhou, Xuan Shi, Biqiong Wu, Tianqi Ao
• DOI: 10.3390/w17213117

Research Groups

This study comparatively evaluates physics-based (BTOP) and data-driven deep learning models (LSTM, BiLSTM) for runoff simulation and short-term forecasting in the Upper Jinsha River, an alpine region. It finds that deep learning models achieve superior accuracy in daily runoff simulation and flood characteristic capture, while the physics-based model offers greater stability in water balance and long-term simulation.

Objective

• To comparatively evaluate the performance of a physics-based hydrological model (BTOP) and data-driven deep learning models (LSTM, BiLSTM) in runoff simulation and short-term forecasting in the Upper Jinsha River, an alpine region.

Study Configuration

• Spatial Scale: Upper Jinsha River basin (case study).
• Temporal Scale: Daily-scale runoff simulation; short-term forecasting (1–7 days).

Methodology and Data

• Models used: BTOP (physics-based hydrological model), LSTM (data-driven deep learning model), BiLSTM (data-driven deep learning model).
• Data sources:

Main Results

• For daily-scale runoff simulation, LSTM and BiLSTM models demonstrated superior capabilities, achieving Nash–Sutcliffe efficiency coefficients (NSE) of 0.82/0.81 (Zhimenda Station) and 0.87/0.86 (Gangtuo Station) during the test period.
• The BTOP model achieved lower validation NSE values of 0.57 at Zhimenda and 0.62 at Gangtuo for daily simulation.
• The BTOP model exhibited greater stability in water balance and long-term simulation due to its hydrology-based structure.
• In short-term forecasting (1–7 days), LSTM and BiLSTM performed comparably, with the bidirectional architecture of BiLSTM offering no significant advantage.
• Data-driven models excelled at capturing flood event peak timing and hydrograph shape.
• The physical BTOP model demonstrated superior stability in flood peak magnitude.
• Forecasts from the data-driven models lacked hydrological consistency between upstream and downstream stations.

Contributions

• Provides a comparative evaluation of physics-based versus deep learning models for runoff simulation and short-term forecasting specifically in complex alpine regions.
• Confirms the superior accuracy of deep learning models for runoff simulation and their effectiveness in capturing key flood characteristics in such challenging environments.
• Highlights the trade-offs between accuracy (data-driven models) and stability/consistency (physics-based models) for hydrological applications in alpine regions.

Funding

Citation

@article{Zhang2025Long,
  author = {Zhang, Feng and Yue, Jiajia and Zhou, Chun and Shi, Xuan and Wu, Biqiong and Ao, Tianqi},
  title = {Long Short-Term Memory (LSTM) Based Runoff Simulation and Short-Term Forecasting for Alpine Regions: A Case Study in the Upper Jinsha River Basin},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17213117},
  url = {https://doi.org/10.3390/w17213117}
}