Verification of hydrological ensemble forecasts
Emmanuel ROULIN
Royal Meteorological Institute of Belgium
Abstract
Methods suited for the verification of probabilistic forecasts will be presented. These methods mainly provide from the verification of meteorological forecasts. More attention will be drawn to the distribution-oriented approach and the relative economic value. The joint distribution of probability forecasts and the observations of an event may be factorized in conditional distributions in order to estimate measures of different aspects of the forecast quality. Using a simple static decision model, the value of the forecasts is compared with a perfect deterministic forecast and depends not only upon the hit rate and the false alarm rate of the forecast system but also the cost-loss situation itself as well as the observed relative frequency of the event. Examples will be taken from hydrological hindcasts performed by using archived operational medium-range precipitation predictions from the ensemble prediction system (EPS) of the ECMWF in a hydrological model.