A scatterometer neural network sensor model with input noise

The ERS-2 satellite carries a scatterometer which measures the amount of ra diation scattered back toward the satellite by the ocean's surface. These m easurements can be used to infer wind vectors. The implementation of a neur al network-based forward model which maps wind vectors to radar backscatter is addressed. Input noise cannot be neglected as it significantly degrades the performance of the model. To account for this noise, a Bayesian framew ork is adopted. Gradient information is used with a non-linear optimisation algorithm to find the maximum a posteriori probability values of the unkno wn variables. The resulting models are shown to compare well with the curre nt operational model both by objective measures and when visualised in the target space. (C) 2000 Elsevier Science B.V. All rights reserved.