F.P. Harter, H.F. de Campos Velho (2001): Data Assimilation in Non-linear Dynamical Systems, Workshop of the Applied Computing Courses of the INPE ( WORKCAP-2001), 25 October, Sao Jose dos Campos (SP), Brasil, pp. 85-87.

Abstract: Traditional techniques of data assimilation consist to improve the prediction done by inaccurate mathematical model, combining ones with observational data. The Kalman filter (KF) is a technique employed in data assimilation, presenting an optimal solution for the linear problem with Gaussian statistics for the noise. The KF in its extended and adaptive versions are suboptimal solutions for non-linear problems, but this technique has a high computational cost. An alternative for KF in data assimilation are the neural networks (NN). In this work a comparison between radial base function NN and the KF is done. The chaotic Lorenz's model is applied in the comparison. Both techniques presented good results. However, after training step, the computational cost of NN is less than KF.