Neural network-based model for estimating the wind vector using ERS scatterometer data

Abstract

A technique based on artificial neural networks is developed for describing the inversion of the CMOD4 model for estimating wind speed and direction from the scatterometers aboard the ERS satellites. Multi-layer perceptrons are trained using simulated data from the CMOD4 model. The normalized radar cross-sections (NRCS) and the respective incidence angles of the three beams are used as inputs. Separate networks are trained for the wind speed and wind direction. It is shown that the neural networks are able to learn the inverse mapping process accurately. The networks are tested with actual scatterometer measurements from the ERS-1 scatterometer. For these data sets, the output of the network appear to be more accurate than the corresponding wind vector estimates provided by the European Space Agency (ESA). It is also shown that the network can also be easily modified to include the effects of extraneous sources such as swells.