Multisource and multitemporal data in land cover classification tasks: the advantage offered by neural networks

Addresses the problem, within the MARS (Monitoring Agriculture with Remote Sensing) project, of land cover classification and acreage assessment based on remotely sensed images for the case of lack of optical input data due to cloud cover. An alternative strategy, based on the exploitation of multi-source and multi-temporal data by means of a feedforward neural network (NN) is proposed and discussed. The results reported show that NNs not only provide a useful tool for data fusion but also an extremely powerful means for early and reliable acreage assessment.