Temporal Evolution of Soil Moisture Maps of Areas at Risk of Floods Produced from Envisat/asar Images through Artificial Neur

In this paper, a technique that is able to retrieve soil moisture maps in areas at risk of flood from ENVISAT/ASAR images, based on a Neural Network approach, was tested. Some experimental trials were carried out in 2003 and 2004 in the agricultural area of the Scrivia watershed in northern Italy, simultaneously to the ASAR passes. This area is susceptible to flooding and has a dense hydrometric network. The potential of SAR images for land classification was confirmed from the analysis of an RGB image composition at HH and HV polarizations, from where information on five surface classes was obtained (urban areas, free water, forests, rough and smooth bare soils). The performances of an inversion algorithm based on Artificial Neural Networks (ANN) for the retrieval of several levels of soil moisture from backscattering data were tested. The results obtained were compared with ground data. This showed a satisfactory agreement and enabled us to generate multi-temporal maps with 4-6 levels of soil moisture.