Computing vertical refractivity profiles by neural networks. Comparison with bulk model results

Abstract

As ducting situations considerably modify the radar coverages in maritime situations, it is of major importance to characterize the corresponding refractivity profiles. A promising solution could be to use Neural Networks methods. Once trained by physical models they could be computationally efficient. Actually, the errors introduced by these techniques in terms of propagation results lies between 3 and 5 dB for a classical radar scenario. So other NN implementations with more hidden layers will have to be tested in the future.