Kritidou et al. (2025) Partitioning uncertainties of extreme flood estimates using long continuous simulations

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-12-18
• Authors: Eleni Kritidou, Martina Kauzlaric, Maria Staudinger, Guillaume Évin, Benoît Hingray, Marc Vis, Daniel Viviroli
• DOI: 10.1016/j.jhydrol.2025.134804

Research Groups

• Department of Geography, University of Zurich, Switzerland
• Institute of Geography and Oeschger Centre for Climate Change Research, University of Bern, Switzerland
• Univ. Grenoble Alpes, CNRS, INRAE, IRD, Grenoble INP, IGE, Grenoble, France

Short Summary

This study partitions the uncertainties in extreme flood estimates from a hydrometeorological modelling chain, revealing that uncertainty increases with return period and its dominant source varies with catchment characteristics. It highlights the critical role of hydrological model parameters in high-elevation catchments and weather generator components in lower-elevation, rainfall-dominated areas.

Objective

• To quantify and partition the contribution of different components (weather generator parameterizations, weather generator stochasticity, hydrological model structures, and hydrological model parameters) to the total uncertainty of simulated floods for return periods ranging from 1 to 1000 years.
• To assess how hydrological model structure affects the magnitude and uncertainty of floods for return periods from 1 to 1000 years.
• To investigate how hydrological model structure influences the estimation of the threshold return period (TRP), beyond which precipitation becomes the main driver of floods.

Study Configuration

• Spatial Scale: Nine large Swiss catchments, ranging from 478 square kilometers to 2965 square kilometers, representing diverse physiographic characteristics (Plateau, pre-Alps, Alps, Southern Alps).
• Temporal Scale: Flood return periods of 1–1000 years, estimated from 30,000 years of synthetic hourly precipitation and temperature time series (generated as 30 scenarios of 1000 years each). Observational data for model parameterization and calibration covered 1930–2019 (daily) and 1990–2019 (hourly).

Methodology and Data

• Models used:
  • Stochastic Weather Generator: GWEX (Evin et al., 2018, 2019) – used with two parameterizations (unconditioned and weather-type conditioned) and an ensemble of 10 bootstrapped members for stochastic variability.
  • Hydrological Model: HBV (Bergström, 1972; Seibert and Vis, 2012) – specifically two structural variants (HBV-72 and HBV-96) differing in their groundwater response routines, run with three representative parameter sets (lower, median, upper) for each structure.
  • Hydrological Routing Model: RS Minerve (Hernández et al., 2014) – used for routing discharge and incorporating hydraulic infrastructure in four catchments.
  • Uncertainty Partitioning: Mixed Analysis of Variance (ANOVA) framework with one random effect (stochastic variability, SV) and three fixed effects (weather generator parameterization, WGpar; hydrological model structure, HMstr; and hydrological model parameters, HMpar, nested within HMstr).
• Data sources:
  • Meteorological Observations: Precipitation data from 1176 stations (Switzerland and neighboring countries) and temperature data from 26 stations, covering 1930–2019 (daily) and 1990–2019 (hourly).
  • Hydrological Observations: Hourly discharge records from the Federal Office for the Environment (FOEN) gauging stations for each study catchment, with record lengths ranging from 6 to 96 years within the period 1923–2019.

Main Results

• The coefficient of variation (CV) of estimated flood return levels increases with increasing return period across all catchments, indicating a general increase in uncertainty for rarer events.
• The dominant source of uncertainty varies significantly with catchment characteristics and return period:
  • In higher elevation catchments (mean elevation > 1500 meters above sea level), hydrological model parameters (HMpar) were the main contributor to uncertainty, accounting for 80% to 92% of the total variance for 100-year return periods and 50% to 67% for 1000-year return periods.
  • In lower-elevation and rainfall-dominated catchments, weather generator parameterizations (WGpar) and stochastic variability (SV) were the primary sources of uncertainty. SV alone accounted for 15% to 65% of the total variance for 1000-year return periods across all catchments.
• The contribution of hydrological model structural uncertainty (HMstr) was relatively small (up to 13% for 1000-year return periods) and comparable between the two HBV model structures (HBV-72 and HBV-96).
• Threshold Return Periods (TRPs), beyond which precipitation becomes the main driver of floods, were found to be very similar for both HM configurations, ranging from approximately 3 to 10 years across the study catchments. This suggests that the influence of the hydrological model diminishes for very large return periods.
• Physiographic characteristics significantly affect the identification of the TRP and the contribution of different components to flood estimate uncertainty, challenging a priori generalizations.

Contributions

• This study is the first to apply a mixed and nested ANOVA framework in a continuous simulation approach under present climate conditions to comprehensively partition parametric and structural uncertainty of hydrological models and weather generators, as well as stochastic variability.
• It provides a quantitative framework for detecting and attributing uncertainties in extreme flood estimates across a wide range of return periods (1–1000 years) for diverse Swiss catchments.
• It offers novel insights into the impact of hydrological model structure on the magnitude and uncertainty of simulated extreme floods and on the estimation of the threshold return period.
• The work underscores the importance of considering diversity in modelling options and performing uncertainty decomposition for informed decision-making in flood risk management.

Funding

• Federal Office for the Environment (FOEN)
• Swiss Federal Office of Energy (SFOE)
• Project: “Extreme Floods in Switzerland” (EXCH)

Citation

@article{Kritidou2025Partitioning,
  author = {Kritidou, Eleni and Kauzlaric, Martina and Staudinger, Maria and Évin, Guillaume and Hingray, Benoît and Vis, Marc and Viviroli, Daniel},
  title = {Partitioning uncertainties of extreme flood estimates using long continuous simulations},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134804},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134804}
}