Jump to Main Content
Distributed hydrological modeling of floods in the Cévennes-Vivarais region, France: Impact of uncertainties related to precipitation estimation and model parameterization
- Navas, Rafael, Delrieu, Guy
- Journal of hydrology 2018 v.565 pp. 276-288
- autumn, floods, hydrograph, hydrologic models, kriging, model uncertainty, radar, rain, runoff, space and time, stream flow, watersheds, France
- The Cévennes-Vivarais (France) is a region prone to heavy precipitation events and flash floods occurring mainly during the autumn season. Due to this vulnerability, it is a well instrumented region to monitor rainfall (4 weather radars of the French ARAMIS radar network, 200 hourly raingauges) and river discharge (45 stations). A multiscale radar-raingauge rainfall re-analysis from 10 to 300 km2 and 1 h time step has been established for the period 2007–2014 by using the kriging with external drift (KED) technique. In the present work, a quantification of the isolated and joint impacts of precipitation and model uncertainties in distributed rainfall–runoff simulations is presented for 10 km2–1 h model resolution. For this purpose the following methodology is implemented: (1) Development of a distributed hydrological model based on the Curve Number and the Unit Hydrograph concepts for the Ardèche and Gardon catchments; (2) Generation of an ensemble of perturbed precipitation fields based on the KED error standard deviations and the space–time structure of the residuals to the drift. (3) Generation of so-called behavioral parameter sets for the hydrological model based on generalized sensitive analysis (GSA) and the use of discharge observations; (4) Implementation of the hydrological model for gauged and ungauged catchments with the radar-raingauge rainfall ensembles and the various parameter sets. Uncertainties in rainfall and runoff simulations are then quantified in terms of coefficient of variation. The main findings reveal that (i) the precipitation uncertainty is dampened in the hydrological simulation, especially as long as the size of the watershed increases; in the considered context, the model uncertainty dominates the precipitation uncertainty and it is shown to be independent on catchment size.