Bakke et al. (2026) The ability of LSTM to model snowmelt versus rainfall generated floods

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-10-06
• Authors: Sigrid Jørgensen Bakke, Danielle Marie Barna, Kolbjørn Engeland, Sjur Anders Kolberg, Sunniva Nordeide
• DOI: 10.5194/hess-30-6207-2026

Research Groups

• Department of Hydrology, Norwegian Water Resources and Energy Directorate (NVE), Oslo, Norway.

Short Summary

This study evaluates the ability of Long Short-Term Memory (LSTM) networks to model snowmelt versus rainfall generated floods in Norway. The results show that LSTM outperforms the operational hydrological model HBV in 70%–91% of catchments for flood peak timing and magnitude, with a 27 pp higher percentage of correctly simulated peak days for rainfall generated floods compared to snowmelt generated floods.

Objective

• Evaluate the ability of LSTM to simulate streamflow series with similar or exceeding overall performance as compared to the operational model used in the region.
• Investigate how well LSTM simulates peak timing and magnitude of all collected floods generated by (i) snowmelt, (ii) rainfall, and (iii) a mix of the two.

Study Configuration

• Spatial Scale: Regional scale, covering the mainland of Norway.
• Temporal Scale: Daily to annual time step, with 15 hydrological years used for model evaluation.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) networks and Hydrologiska Byråns Vattenbalansavdelning (HBV) model.
• Data sources: Gridded hydrometeorological data from SeNorge_2018, daily discharge data from 200 streamflow gauging stations.

Main Results

• LSTM outperforms HBV in 70%–91% of catchments for flood peak timing and magnitude.
• LSTM has a 27 pp higher percentage of correctly simulated peak days for rainfall generated floods compared to snowmelt generated floods.
• LSTM's ability to simulate flood peak magnitudes is similar for the different flood types, with percent errors within 20% for close to half of the events.

Contributions

• This study fills the gap in existing evaluations of LSTM's ability to simulate floods by evaluating events generated by different processes separately.
• The results bring confidence that LSTM can improve hydrological services in regions subject to both snowmelt and rainfall generated floods.

Funding

• Not specified.

Citation

@article{Bakke2026ability,
  author = {Bakke, Sigrid Jørgensen and Barna, Danielle Marie and Engeland, Kolbjørn and Kolberg, Sjur Anders and Nordeide, Sunniva},
  title = {The ability of LSTM to model snowmelt versus rainfall generated floods},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-6207-2026},
  url = {https://doi.org/10.5194/hess-30-6207-2026}
}