E.H. Shiguemori, J.D.S. da Silva, H.F. de Campos Velho (2008): Atmospheric Temperature Retrieval from Satellite Data: New Non-extensive Artificial Neural Network Approach, 23rd Annual ACM Symposium on Applied Computing, Track on Artificial Intelligence in Space Applications, March 16-20, Fortaleza (CE), Brazil.

Abstract: In this paper, vertical temperature profiles are inferred by neural networks based inverse procedure from satellite data, non-linear function estimation. A new approach to classical Radial Basis Function neural network is trained using data provided by the direct model characterized by the Radiative Transfer Equation (RTE). The neural network results are compared to the ones obtained from classical neural networks Radial Basis Function and traditional method to solve inverse problems, the regularization. In addition, real radiation data from the HIRS/2 - High Resolution Infrared Radiation Sounder - is used as input for the neural networks to generate temperature profiles that are compared to measured temperature profiles from radiosonde. Analysis of the new approach results reveals the generated profiles closely approximate the results obtained with classical neural networks and regularized inversions, [5] [15], thus showing adequacy of neural network based models in solving the inverse problem of temperature retrieval from satellite data. The advantages of using neural network based systems are related to their intrinsic features of parallelism; after trained, the networks are much faster than regularized approaches, and hardware implementation possibilities that may imply in very fast processing systems.