Automatic selection by means of neural networks of GOME optimum spectral channels for the retrieval of ozone vertical profiles

A neural network procedure aiming at the retrieval of ozone concentration profiles from the radiance measurements provided by the instrument GOME on board of ESA satellite ERS-2 is described. The potentiality of neural networks has been exploited not only for the inversion purposes but also for sensitivity analysis. In fact, via electromagnetic modelling and pruning algorithms, a complete procedure has been designed to extract the GOME channels most crucial for the inversion. The effectiveness of the retrieval algorithm has been evaluated comparing its performance to that yielded by other instruments and inversion techniques.