Unraveling uncertainties of extreme-flood simulations over Switzerland based on different weather generator scenarios

Abstract

Floods are a high-impact natural hazard in Switzerland and are responsible for considerable damage to property values, infrastructure and agricultural land (Hilker et al., 2009). Hence, reliable flood estimates are critical in flood risk reduction, emergency preparedness and disaster management. The traditional flood estimation methods rely on statistical techniques based on observed streamflow and precipitation. However, the short length of observational records is often a limiting factor that leads to considerable uncertainties in flood estimates, especially for rare floods. An alternative approach that circumvents this limitation is the combination of stochastic weather generators with hydrological models using long continuous simulations. The advantage of this approach is that it avoids assumptions about antecedent catchment states (e.g., soil moisture, snowpack, storage levels of lakes and reservoirs) and simplified representations of the underlying physical flooding processes. Here, we use an elaborate framework based on continuous simulations with a hydrometeorological modeling chain (Viviroli et al., 2022) to estimate rare floods for large catchments in Switzerland (larger than ~450 km²). The modeling chain starts with a multi-site stochastic weather generator (GWEX), focusing on generating extremely high precipitation events. Then, a bucket-type hydrological model (HBV) is used to simulate discharge time series. Finally, the RS Minerve (RSM) model is employed to implement simplified representations of river channel hydraulics and floodplain inundations.We aim to investigate the uncertainties of derived flood estimates, with an experimental set-up focusing on the first component of the modeling chain, the weather generator. Aiming to explore the impact of precipitation inputs on flood estimates, GWEX is subject to different tests while the remaining components (HBV and RSM) remain unchanged. To achieve this, two experiments have been conducted: (a) parameterization of GWEX based on a bootstrap sampling of observed precipitation, from which we get an ensemble of 10 different synthetic time series (b) conditioning of the most relevant GWEX parameters to different weather types that describe intermediate, moderate and stronger precipitation intensities. A set of reference scenarios using the initial parameters of GWEX serves as a benchmark for comparison.Our experimental framework unravels the sensitivity of the catchments to changes in precipitation inputs. While bootstrapping shows a higher impact compared to weather-type conditioning, the latter seems to reduce the spread of uncertainty both in precipitation and simulated floods. These findings provide an essential basis for follow-up studies on hazard assessment, safety analyses and hydraulic engineering projects.