Improving the characterization of initial conditions for streamflow prediction using a precipitation reconstruction algorithm

Abstract

Hydrologic forecasts derive their skill from knowledge of initial conditions at the forecast date, climate forecast, model structure and parameters. Uncertainty on the initial conditions has as much influence as uncertainty on the weather forecasts on the hydrologic forecasts for some watersheds. Initial conditions depend on several parameters: evapotranspiration, soil composition and mainly former rain events which are measured by rain gauges or radars. Precipitation measures often show uncertainties or even data gaps, and thus, the evolution of the soil states is unknown. The initial conditions can only be determined by following up the evolution of the variable states. The measured discharge runoff is the only available reliable data and thus, that information can be used to determine the variable states, by the inversion of the rainfall-runoff model. This study proposes a post-processing method that adjust the initial conditions using measured discharge runoff at the outlet of a watershed. The heuristic is applied on the Echez watershed, and the effectiveness of the method is illustrated thanks to a comparison of the results obtained with the measured observation during an analysis period falling out after the forecast date.