Practical use of ensemble meteorological forecast for streamflow prediction : Examining and correcting the bias in the ensemble meteorological forecast.

The main source of uncertainty when predicting streamflows using a hydrologic model is meteorological forecasts. It is therefore of great importance to take into account uncertainty due to meteorological events. For this purpose, Hydro-Quebec intends to use ensemble meteorological forecasts. After some preliminary analysis, several problems persist concerning the practical use of ensemble meteorological forecasts for streamflow prediction. For instance : \begin{enumerate}

The Canadian meteorological forecasts are given on a coarse grid (1.2 deg in latitude and longitude). Therefore, at best only a few grid points fall within the limits of a typical watershed.

There is an obvious bias in temperature and precipitation ensemble forecasts.

The ensemble meteorological forecasts are not equiprobable. Therefore, we should eventually assign unequal weights to each member of the ensemble. In this talk we will first propose several statistical methods to analyse and correct the bias in the meteorological ensemble forecasts. These approaches will be applied and compared on real data on a basin managed by Hydro-Quebec. Then the problem of dowscaling will be discussed. In this work, we consider both classical and Bayesian methods.