Neural-network based stateless ice detection in ERS scatterometer data

A new scheme to perform stateless ice/sea discrimination in ERS scatterometer data is proposed. This method consists in combining several methods proposed in the literature using a Bayesian framework. Each of the combined method is first reviewed in a consistent framework. In particular, the ice/sea probability according to each individual criterion is extracted using a neural network. The proposed method is shown to provide acceptable results even without taking into account historic data, i.e. without performing temporal averaging.