Abstract: Data assimilation is a step for improving forecasting process by means of a weighted combination between observational and data from a mathetical model. This procedure is essential for operational prediction centers for weather, ocean circulation, and atmospheric pollution. The goal here is to compare four schemes for data assimilation: Kalman filter [1, 2, 4, 5], optimal interpolation [1, 2, 4], variational approach [1, 2, 4], and artificial neural networks (ANN) [3, 5]. The multilayer perceptron is the approach used to implement the ANN. The assimilation techniques are tested on the Lorenz dynamical system. It is imorportant to note that ANN is a method recently developed, and its application is under study. However, all thecniques tested presented a good performance, depending on the ratio of sampling of the observations. From computational point of view, the ANN presented a best perfoamance, considering only a trained ANN.

References:

[1] Daley, R., Atmospheric data analysis, Cambridge University Press, Cambridge (1991).

[2] Kalnay, E. (2003): Atmospheric modeling, data assimilation and predictability, Cambridge University Press, Cambridge.

[3] Harter, F. P., Campos Velho, H. F. (2005): Recurrent and Feedforward Neural Networks Trained with Cross Correlation Applied to the Data Assimilation in Chaotic Dynamic, Brazilian Journal of Meteorology, 20(3), 411-420.

[4] Lawless, A. S. (2002): Data assimilation with Lorenz equations, Tecnical Report, University of Reading (UK).

[5] Nowosad, A. G., Rios Neto, A., Campos Velho, H. F. (2000): Data assimilation using an adaptative Kalman filter and Laplace transform, Hybrid Methods in Engineering, 2(3), 291-310.

[6] Nowosad, A. G., Rios Neto, A., Campos Velho, H. F. (2000): Data Assimilation in Chaotic Dynamics Using Neural Networks, 3rd International Conference on Nonlinear Dynamics, Chaos, Control and Their Applications in Engineering Sciences, July 31 - August 4, Campos do Jordão (SP), Brazil, Vol. 6, Chapter 6: Control, pp. 212-221.