Cherry-picking rainfall events from ensemble reforecast archives: what flood events the current climate didn’t show us (yet)?

Abstract:

Stress tests in flood risk management require the definition and formulation of physically plausible rainfall scenarios. Quantifying the chance of extreme rainfall events and their spatiotemporal patterns is fundamentally constrained by the short length of the recent observational record resulting in turn in a small flood sample, representative only for a limited number of low probability - but high impact - hydro-meteorological configurations. While weather reforecast archives have been already exploited as a large sample of the present day atmosphere for expanding the knowledge about potential damage and loss produced by wind-storms, very little has been done in this sense to explore the possibility of using these data to “semi-empirically” estimate extremely rare and unprecedented flood events.

In this study we use two forecast products of ECMWF, the extended range forecasts (ENSext) and the seasonal forecasts (SEAS5). We extracted the largest precipitation events from the ensembles by considering different spatiotemporal patterns and used these events as input for a comprehensive modelling chain assessing losses in the Aare river basin, Switzerland. Each ensemble member was considered as a physically consistent realization of meteorological fields, thus the ensembles can be regarded as independent sets that cumulate to over 2000 (ENSext) and more than 6000 (SEAS5) years of data. We compare peak discharges and loss estimation results with previous studies making use of other methods for defining extreme events such as extracting extreme events from long-term GCM/RCM simulations or from Monte-Carlo-Simulations with a physically-consistent generation of spatiotemporal rainfall patterns of probable maximum precipitation scenarios. We find floods nearly equivalent to the probable maximum flood, what opens the door for a discussion on the length of a meteorological time series for representative flood sampling, and the search for black-swan events.

Besides providing accurate hazard footprints highlighting a high inner-catchment variability, and allowing for reducing uncertainties in probability estimation of up to at least 1000-year events, the results bear the potential of developing story-lines of extreme flood events and training tools for local civil protection authorities.