Muñoz‐Castro et al. (2026) How well do hydrological models simulate streamflow extremes and drought-to-flood transitions?

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-02-12
• Authors: Eduardo Muñoz‐Castro, Bailey J. Anderson, Paul C. Astagneau, Daniel L. Swain, Pablo A. Mendoza, Manuela I. Brunner
• DOI: 10.5194/hess-30-825-2026

Research Groups

• WSL Institute for Snow and Avalanche Research SLF, Davos Dorf, Switzerland
• Climate Change, Extremes and Natural Hazards in Alpine Regions Research Center CERC, Davos Dorf, Switzerland
• Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
• California Institute for Water Resources, University of California Agriculture and Natural Resources, Davis, CA, USA
• Weather Extremes Across Scales, NSF National Center for Atmospheric Research, Boulder, CO, USA
• Civil Engineering Department, Universidad de Chile, Santiago, Chile
• Advanced Mining Technology Centre (AMTC), Universidad de Chile, Santiago, Chile

Short Summary

This study evaluates how well four conceptual hydrological models simulate streamflow extremes and drought-to-flood transitions across 63 catchments in Chile and Switzerland, assessing the impact of model structure and calibration choices. It finds that general model performance (KGE) does not guarantee accurate detection of extremes, and model structure is the most critical factor, with transitions being poorly represented, especially in semi-arid and high-mountain regions.

Objective

• To investigate the extent to which hydrological models can represent consecutive drought-to-flood transitions and the impact of model structure and calibration choices on their representation.
• How suitable is the Kling–Gupta Efficiency (KGE) for calibrating models aimed at jointly simulating streamflow droughts and floods?
• Which modeling choices (e.g., model structure, KGE formulation, etc.) are most important for simulating droughts, floods, and their transitions?
• In which catchments are drought-to-flood transitions most challenging to model and detect?

Study Configuration

• Spatial Scale: 63 near-natural catchments (24 in Chile, 39 in Switzerland) at the catchment scale.
• Temporal Scale: Daily streamflow records from 1981 to 2020 (at least 30 years). Calibration period: 2000–2020. Evaluation period: 1985–1999. Hydroclimatic characterization period: 1985–2020.

Methodology and Data

• Models used: GR4J, GR5J, GR6J (all coupled with CemaNeige snow module), and TUWmodel (based on the HBV model). These are conceptual bucket-style rainfall-runoff hydrological models.
• Data sources:
  • CAMELS Chile (CL) and CAMELS Switzerland (CH) datasets for meteorological forcings, streamflow records, snow water equivalent (SWE) estimates, and catchment boundaries.
  • Meteorological forcings: CR2Met version 2.5 for Chile (approximately 5 km x 5 km resolution) and RhiresD version 2 for Switzerland (approximately 2 km x 2 km resolution).
  • Topographic characteristics and hypsometric curves: Multi-Error-Removed Improved-Terrain (MERIT) digital elevation model.
  • Actual evapotranspiration (ET): Satellite- and reanalysis-based GLEAM v3.8a dataset.
  • Streamflow records from national agencies: General Directorate of Water of Chile (DGA) and Swiss Federal Office for the Environment (FOEN).

Main Results

• General model performance, as expressed by the Kling–Gupta Efficiency (KGE), does not guarantee good performance in detecting streamflow extremes and their transitions, particularly for floods and drought-to-flood transitions.
• Hydrological models are generally better at detecting droughts (median Critical Success Index (CSI) across catchments and KGE formulations: 0.50–0.58) than floods (median CSI: 0.13–0.34) and drought-to-flood transitions (median CSI: 0.25–0.33).
• The most important modeling decision for simulating extreme events and their transitions is the choice of a suitable model structure.
• The four tested models (GR4J, GR5J, GR6J, and TUW) show similar performance in detecting extreme events, indicating that increasing model complexity by adding more parameters does not necessarily improve the representation of extreme events.
• The choice of the objective function and its exact configuration is, overall, less important than model structure, though a suitable streamflow transformation can improve flood detection. A joint focus on high and low flows (HiLo transformation) can improve model performance without compromising its ability to capture streamflow extremes.
• Drought-to-flood transitions are more challenging to capture in semi-arid, high-mountain, and flashy catchments (showing negative correlations between CSI and aridity index, mean elevation, and flow duration curve slope) compared to humid low-elevation catchments.
• The high relative importance of forcing adjustment parameters (e.g., dP, dT, SCF) for event detection suggests that meteorological forcings significantly impact the detection of streamflow extremes and their transitions.

Contributions

• Provides a comprehensive intercomparison study on the ability of conceptual hydrological models to simulate drought-to-flood transitions.
• Quantifies the relative importance of various modeling decisions (model structure, KGE formulation, streamflow transformation, and weights) on the simulation and detection of compound streamflow extremes.
• Demonstrates that traditional goodness-of-fit metrics like KGE are insufficient for evaluating extreme event detection, advocating for direct metrics such as the Critical Success Index (CSI).
• Identifies specific catchment characteristics (semi-arid, high-mountain) that pose significant challenges for modeling drought-to-flood transitions.
• Offers insights for future hydrological model development, suggesting that efforts should prioritize improving model structures and the quality of meteorological forcing datasets over fine-tuning calibration procedures for better representation of compound extreme events.

Funding

• Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (Swiss National Science Foundation, SNSF) – grant no. 200021_214907.
• ANID-PIA Project AFB230001 (AMTC).

Citation

@article{MuñozCastro2026How,
  author = {Muñoz‐Castro, Eduardo and Anderson, Bailey J. and Astagneau, Paul C. and Swain, Daniel L. and Mendoza, Pablo A. and Brunner, Manuela I.},
  title = {How well do hydrological models simulate streamflow extremes and drought-to-flood transitions?},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-825-2026},
  url = {https://doi.org/10.5194/hess-30-825-2026}
}