Yin et al. (2025) Reconstruction of Daily Runoff Series in Data-Scarce Areas Based on Physically Enhanced Seq-to-Seq-Attention-LSTM Model

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

Identification

• Journal: Water
• Year: 2025
• Date: 2025-11-28
• Authors: Zhaokai Yin, Tao Xu, H. Ye, Lin Wang, Li‐Li Liang
• DOI: 10.3390/w17233396

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study proposes a Physics-enhanced Seq-to-Seq Attention LSTM (PSAL) model to reconstruct high-accuracy daily streamflow from remote sensing data in data-scarce regions, demonstrating significant performance improvements over a baseline model on the Jinsha River.

Objective

• To develop and evaluate a physics-enhanced deep learning model for high-accuracy daily streamflow reconstruction from remote sensing data, addressing the sparsity of remote sensing inversions and discontinuity of discharge observations.

Study Configuration

• Spatial Scale: Five representative gauging sites on the Jinsha River (river basin/sub-basin scale).
• Temporal Scale: Daily streamflow reconstruction, with lagged output configurations from 1 day (T-1) to 7 days (T-7).

Methodology and Data

• Models used: Physics-enhanced Seq-to-Seq Attention LSTM (PSAL); Baseline: Seq-to-Seq Attention LSTM.
• Data sources: Remote sensing inversions (for river discharge monitoring), discontinuous discharge observations, and precipitation data (as a key hydrological driver).

Main Results

• PSAL achieved high streamflow reconstruction accuracy across five gauging sites on the Jinsha River (mean Nash-Sutcliffe Efficiency (NSE) = 0.81).
• The T-7 lagged output configuration yielded the best performance (mean NSE = 0.85).
• PSAL significantly improved reconstruction skill compared to the baseline Seq-to-Seq Attention LSTM model (mean ΔNSE = 0.76).
• Feature ablation analysis revealed that precipitation has a strong influence on model performance (mean ΔNSE = 0.32).

Contributions

• Presents a novel approach integrating physical knowledge with data-driven methods for streamflow reconstruction.
• Addresses the critical hydrological challenge of reconstructing daily streamflow from remote sensing data in data-scarce regions.
• Offers theoretical support and methodological guidance for digital twin watershed development and historical hydrological data infilling.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Yin2025Reconstruction,
  author = {Yin, Zhaokai and Xu, Tao and Ye, H. and Wang, Lin and Liang, Li‐Li},
  title = {Reconstruction of Daily Runoff Series in Data-Scarce Areas Based on Physically Enhanced Seq-to-Seq-Attention-LSTM Model},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17233396},
  url = {https://doi.org/10.3390/w17233396}
}