Ensemble hydro-meteorological simulation for flash flood early detection in southern Switzerland

Ongoing changing climate has raised the attention towards weather driven natural hazards. Local floodings and debris flows following exceptional downpours often come without any adequate warning and cause heavy tolls to the human society. This work proposes a novel flood alert system for small catchments prone to flash flooding, capable of monitoring a large portion of the European domain. Operational streamflow simulations are produced through distributed hydrological modeling of ensemble weather forecasts. A long-term reforecast dataset is run through the same hydrological model to derive coherent warning thresholds. These are compared with operational discharge ensembles in a threshold exceedance analysis to produce early warnings. A case study in the southern Switzerland is tested over a 17-month period and system skills are evaluated by means of different quantitative and qualitative analyses. Results from three different predictors derived from the streamflow ensemble are shown and compared, also by accounting for the persistence of lagged forecasts. Significant improvements in predicting discharge thresholds exceedance are achieved by fitting gamma probability distributions to the raw ensemble. Further discussion underlines the limits of predictability of extreme events in small catchments due to the comparatively coarse space-time resolution of current weather forecasts.