These de Doctorat - LTHE

Abstract
Catchments of Southern France are regularly subject to quick oods, usually in autumn,
generated by intense rainfall events. Thus, ood risk is a major concern, necessitating a maximal
lead-time to issue early ood warning, as well as an estimation of future discharges. Firstly, the
elements required for hydrological forecasts and the related uncertainties are illustrated. Then, a
simple and modular approach adapted to ash ood catchments (having a time to peak of about
few hours) is proposed. Considering the targeted lead-time (24-48h), quantitative precipitations
forecasts are a key element of this approach. Two prediction systems are described and evaluated :
the EPS ensemble forecasts provided by ECMWF and the ANALOG probabilistic forecasts
issued from an analog sorting technique produced by LTHE. A statistical correction of the latter
is suggested to improve its reliability. The dierent forecasts are thereafter disaggregated by a
generator from a 12 or 24 hours time-step to hourly scenarios which respect the precipitation
forecasts and are climatologically consistent. Rainfall scenarios are then used as input to a simple
and robust hydrological model, to provide hydrological ensemble forecasts. These forecasts get
noticeably improved when sub-daily information about rainfall amounts is provided, either from
EPS at a 6 or 12h time-step, or from ANALOG applied at 12h, or from a combination of both
approaches, taking into account daily rainfall amount from ANALOG and a sub-daily chronology
from EPS at 6h.
Keywords :
Flash ood, Hydrologic forecasting, Hydro-meteorological forecasting, Ensemble prediction,
Probabilistic quantitative precipitation forecast, Analogues sorting, Uncertainties propagation,
Hydrologic modelling.