Ashcroft et al. (2026) Evaluation of AI-based seasonal weather ensembles as input for fluvial flood risk estimation: a case study over the Elbe basin

Identification

• Journal: Natural hazards and earth system sciences
• Year: 2026
• Date: 2026-07-07
• Authors: Alison Poulston, Marius K. Koch, Georg Ertl, Kirsty Brown, James Butler, Anthony Hammond, Owen Jordan, Sarah Warren, Rob Lamb, Paul J. Young, David Wood
• DOI: 10.5194/nhess-26-3129-2026

Research Groups

• JBA Risk Management, Skipton, BD23 3FD, United Kingdom
• Nvidia Corporation, 2788 San Tomas Expressway, Santa Clara, CA 95051, USA
• Lancaster Environment Centre, Lancaster University, Lancaster, LA1 4YQ, United Kingdom
• School of Engineering, Newcastle University, Newcastle, NE1 7RU, United Kingdom

Short Summary

This study evaluates the use of AI-based seasonal weather ensembles as input for fluvial flood risk estimation in a case study over the Elbe basin. The results show that the PrecipHENS framework can produce physically plausible and diverse weather scenarios suitable for flood risk analysis.

Objective

• Evaluate the suitability of AI-generated seasonal weather ensembles for fluvial flood risk estimation.
• Assess the performance of the PrecipHENS framework in reproducing present-day climatology, preserving spatial and temporal dependence, representing extreme events, and extrapolating beyond the historical record.

Study Configuration

• Spatial Scale: Continental-scale river basin (Elbe River basin)
• Temporal Scale: Seasonal scale (winter seasons)

Methodology and Data

• Models used: PrecipHENS framework combining SFNO-based atmospheric model with an AFNO-based diagnostic precipitation model.
• Data sources: ERA5 reanalysis data for initial conditions, historical reference, and river flow response calibration.

Main Results

• The PrecipHENS framework reproduces key features of precipitation and temperature climatology over the Elbe basin.
• The generated ensembles preserve spatial dependence, including the decay of extremal co-occurrence with distance.
• The PrecipHENS framework generates a substantially broader diversity of extreme precipitation events than an industry-standard conditional multivariate extreme-value benchmark.

Contributions

• This study provides a proof-of-concept assessment of AI weather generation for flood risk event sets under present-day conditions.
• The results demonstrate the potential of AI-based seasonal ensembles to support event set construction for fluvial flood risk applications.

Funding

• This research was funded by [insert funding sources and reference codes].

Citation

@article{Ashcroft2026Evaluation,
  author = {Ashcroft, John and Poulston, Alison and Koch, Marius K. and Ertl, Georg and Brown, Kirsty and Butler, James and Hammond, Anthony and Jordan, Owen and Warren, Sarah and Lamb, Rob and Young, Paul J. and Wood, David},
  title = {Evaluation of AI-based seasonal weather ensembles as input for fluvial flood risk estimation: a case study over the Elbe basin},
  journal = {Natural hazards and earth system sciences},
  year = {2026},
  doi = {10.5194/nhess-26-3129-2026},
  url = {https://doi.org/10.5194/nhess-26-3129-2026}
}