Weather types prediction at medium-range from ensemble forecasts

International audienceMedium-range weather forecasts can be ofhigh economic value in many fields: agriculture, renewableenergy production, maintenance operations planning.Such forecasts can be based on ensembles derivedfrom weather models, and the postprocessing of suchensembles is an active research problem in the statisticalweather community. In this work, we try to face theproblem of long forecasting horizons, and focus on themultivariate case where different meteorological variablesinteract. The prediction problem is simplified and definedas the prediction of a weather type, which is a categoricalvariable defined by the interaction of the meteorologicalvariables. We use machine learning techniques to predictthis weather type from the multivariate ensembleforecasts. The algorithms are applied to a 5 to 10 daysweather forecasting in the north-west of France, based onwind and precipitation data from the ECMWF ensemblesystem