Abstract: Artificial Neural Network (ANN) is a new and encouraging approach for data assimilation process. The performance of two feedforward (multilayer perceptron and radial basis function), and two recurrent (Elman and Jordan) ANN are analized in a data assimilation process. The DYNAMO model, a non-linear shallow model, is used as a test problem. These four ANNs were trainned for emulating a Kalman filter using cross validation scheme. One important issue is that the ANNs used for data assimilation do not use the forecats and observation data considering the whole domain (as employed in previous works [1-3]). This procedure reduces dramaticly the complexity of the algorithm used. Good results were obtained applying this scheme, encouraging the use in the operational primitive equations models.

References

1. H.F. Campos Velho, N.L. Vijaykumar, S. Stephany, A.J. Preto, A.G. Nowosad (2002): A Neural Network Implementation for Data Assimilation using MPI, Applications of High-Performance Computing in Engineering (Eds. C.A. Brebbia, P. Melli, A. Zanasi), WIT Press, Southampton (UK), Section 5, pp. 211-220.

2. A.G. Nowosad, H.F. Campos Velho (2003): New Learning Scheme for Multilayer Perceptron Neural Network Applied to Meteorological Data Assimilation, XXIV Iberian Latin-american Congress on Computational Methods in Engineering (CILAMCE-2003), 29-31 October, Ouro Preto (MG), Brazil - Proc. in CD-Rom: paper code cil262-28, Abstract Book pp. 231.

3. A.G. Nowosad, H.F. Campos Velho, A. Rios Neto (2000): Neural Network as a New Approach for Data Assimilation, Brazilian Congress on Meteorology, 16-20 October, Rio de Janeiro (RJ), Brasil, Proc. in CD-ROM (paper code PT00002), pp. 3078-3086.