Abstract: In this paper a non-linear formulation of Hopfield neural network is applied to the problem of retrieval of vertical temperature profiles in the atmosphere from satellite data. This estimation is a key issue in meteorology, since it provides an important input for numerical weather prediction codes. This is a fundamental procedure in the Southern Hemisphere, where there are large areas uncovered by data collecting ground stations. This inversion process is a non-linear reconstruction, therefore a non-linear version of the standard Hopfield neural network is used to approach the problem. We employ a neural network implementation from a previous work [1]. The advantage of using Hopfield neural network is in the fact there is no need of set target data for training purposes. Finally, the results are compared with inversions obtained from regularized solutions [2, 3].

References

[1] R.C.O. Sebastiao, N.H.T. Lemes, L. S. Virtuoso e J.P. Braga (2003): Nonlinear global inversion of potential energy surfaces from the experimentally determined second virial coefficients, Chem. Phys. Lett., 378(3-4), 406–409

[2] F.M. Ramos, H.F. de Campos Velho, J.C. Carvalho, N.J. Ferreira (1999): Novel Approaches on Entropic Regularization, Inverse Problems, 15(5), 1139-1148.

[3] J.C. Carvalho, F.M. Ramos, N.J. Ferreira H.F. de Campos Velho (1999): Retrieval of Vertical Temperature Profiles in the Atmosphere, 3rd International Conference on Inverse Problems in Engineering (3ICIPE), Proceedings in CD-ROM, under paper code HT02 (see also in the internet: http://www.me.ua.edu/3icipe/fin3prog.htm) - Proc. Book: pp. 235-238, Port Ludlow, Washington, USA, June 13-18, UEF-ASME (2000).