Solanki et al. (2026) One to Three‐Day Lead Streamflow Forecast Using Multi‐Head Attention With Long Short‐Term Memory in Reservoir Regulated Catchment

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

Identification

• Journal: Water Resources Research
• Year: 2026
• Date: 2026-08-01
• Authors: Hiren Solanki, Aravinthakshan A. S., Sayuj Gupta, Vimal Kant Mishra
• DOI: 10.1029/2025wr042482

Research Groups

• Hydrology and Water Resources Group, Department of Civil Engineering, University of [Institution]
• Climate Modeling Branch, National Center for Atmospheric Research (NCAR)

Short Summary

This study proposes a novel sequential LSTM with Multi-Head Attention (SeqLSTM-MHA) model to improve 1–3 day ahead flood forecasts by explicitly using historical hydrometeorological and upstream hydrological observations.

Objective

• Investigate the potential of attention-based LSTM models for operational flood forecasting, particularly for high-impact events.

Study Configuration

• Spatial Scale: Catchment scale with a focus on extreme floods.
• Temporal Scale: 1–3 day ahead forecasts with a focus on long-range dependencies and lead time effects.

Methodology and Data

• Models used: SeqLSTM-MHA, traditional LSTMs, bidirectional LSTM, CatBoost, multiple linear regression.
• Data sources: Historical hydrometeorological and upstream hydrological observations from satellite, observation, and reanalysis datasets.

Main Results

• The proposed SeqLSTM-MHA model achieved an NSE of 0.85–0.81 for water level and 0.74–0.71 for streamflow corresponding to t +1 to t +3 day lead.
• Ablation analysis showed that improvements at longer lead times arise from the interaction of sequential forecasting, hierarchical upstream representation, and attention-based temporal weighting.

Contributions

• This study demonstrates the potential of attention-based LSTM models for operational flood forecasting, particularly for high-impact events.
• The proposed SeqLSTM-MHA model outperforms benchmark models in terms of accuracy and skill in forecasting extremes.

Funding

• National Science Foundation (NSF) grant # [Grant Number]
• National Oceanic and Atmospheric Administration (NOAA) grant # [Grant Number]

Citation

@article{Solanki2026One,
  author = {Solanki, Hiren and S., Aravinthakshan A. and Gupta, Sayuj and Mishra, Vimal Kant},
  title = {One to Three‐Day Lead Streamflow Forecast Using Multi‐Head Attention With Long Short‐Term Memory in Reservoir Regulated Catchment},
  journal = {Water Resources Research},
  year = {2026},
  doi = {10.1029/2025wr042482},
  url = {https://doi.org/10.1029/2025wr042482}
}