Classification of ensemble forecasts by means of an artificial neural network

The problem of clustering the members of an ensemble forecast is addressed. Fields of 500 hPa height are classified by a self-organising Kohonen artificial neural network, for which the learning process is described. The members of the ensemble are classified according to the learned situations. An objective measure of the predictability is proposed, using the concept of information entropy. This classification technique also provides a measure for comparing forecasts with their verifying analysis.