Probabilistic evaluation of ensemble discharge nowcasts in two nested Alpine basins prone to flash floods

The quality of hydrological discharge simulations depends to a great extent on the uncertainties in the meteorological input and the model parameterization. To quantify these uncertainties, we adopt ensemble techniques in a 4-year nowcast experiment for two nested flash-flood-prone basins in the southern Swiss Alps. The spatiotemporal uncertainties in the weather radar quantitative precipitation estimates (QPE) were accounted for by applying an ensemble of 25 radar fields. To account for uncertainties in model parameterization, a Monte Carlo experiment was run to find 26 equifinal model realizations. The resulting parameter ensemble, consisting of 26 members, was run with precipitation input obtained from interpolated pluviometer data and with the deterministic operational weather radar QPE. To produce the discharge nowcast, the PREcipitation-Runoff-EVApotranspiration HRU Model (PREVAH) was used. PREVAH was calibrated for the main catchment Verzasca. The results for the sub-catchment Pincascia are an independent internal verification of the nowcasting system. The three ensemble nowcasts and the two deterministic nowcasts are evaluated for a 4-year time series and for two events included in that period. The event analysis shows no clear superiority for either pluviometer-based or radar-based nowcasts. The performance for single events depends heavily on the storm characteristics. However, the evaluation of the 4-year nowcast shows that pluviometer-based nowcasts outperform radar-based nowcasts in the gauged and calibrated catchment and that there is added value in the application of parameter ensembles. For the small, ungauged catchment, the results achieved by the radar-based nowcasts are superior to the pluviometer-based nowcasts. Especially the radar ensemble proves to be of significant advantage for flash flood nowcasts in such catchments. Copyright (C) 2012 John Wiley & Sons, Ltd.