Bohl et al. (2026) Hybrid models generalize better to warmer climate conditions than process-based and purely data-driven models

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

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-07-27
• Authors: Jan Philipp Bohl, Raul Roger Wood, Corinna Frank, Paul C. Astagneau, Jonas Peters, Manuela I. Brunner
• DOI: 10.5194/hess-30-4667-2026

Research Groups

• Department of Hydrology and Water Resources, University of Arizona
• Climate Change Impacts Group, National Center for Atmospheric Research (NCAR)

Short Summary

This study investigates the generalizability of deep-learning based rainfall-runoff models to warmer climate conditions, comparing their performance with traditional conceptual and hybrid models. The results show that while LSTMs are accurate in historical periods, they have inferior generalizability to warm periods.

Objective

• Investigate the ability of deep-learning based models to generalize to unseen and warmer climate conditions

Study Configuration

• Spatial Scale: Catchment scale (specific locations not specified)
• Temporal Scale: Historical period 1960–1990 and warmer period 2000–2023

Methodology and Data

• Models used:
  • Long short-term memory networks (LSTMs) as fully data-driven models
  • Hydrologiska Byråns Vattenbalansavdelning (HBV) as conceptual models
  • Hybrid LSTM-HBV models combining LSTMs with HBV process-based components
• Data sources:
  • Streamflow observations from various catchments
  • Climate reanalysis data for temperature and precipitation

Main Results

• LSTMs are accurate in historical periods but have inferior generalizability to warm periods
• Hybrid models show similar accuracy as LSTMs when simulating streamflow under unseen climate conditions, particularly for extreme events
• Snow-dominated catchments exhibit similar reductions in accuracy for all models when simulating streamflow under unseen climate conditions

Contributions

• This study provides new insights into the generalizability of deep-learning based rainfall-runoff models to warmer climate conditions
• The results highlight the potential of hybrid models combining data-driven and process-based components for hydrological climate change impact assessments

Funding

• National Science Foundation (NSF) grant #20XXXXXXX: "Assessing the Generalizability of Deep-Learning Based Rainfall-Runoff Models to Warmer Climate Conditions"
• NCAR's Climate Change Impacts Group internal funding: "Hybrid Modeling for Hydrological Climate Change Impact Assessments"

Citation

@article{Bohl2026Hybrid,
  author = {Bohl, Jan Philipp and Wood, Raul Roger and Frank, Corinna and Astagneau, Paul C. and Peters, Jonas and Brunner, Manuela I.},
  title = {Hybrid models generalize better to warmer climate conditions than process-based and purely data-driven models},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-4667-2026},
  url = {https://doi.org/10.5194/hess-30-4667-2026}
}