%0 Conference Proceedings
%4 sid.inpe.br/mtc-m21b/2017/12.04.14.16
%2 sid.inpe.br/mtc-m21b/2017/12.04.14.16.30
%T Supervised neural network for data assimilation on atmospheric general circulation model
%D 2017
%A Cintra, Rosangela Saher,
%A Campos Velho, Haroldo Fraga de,
%A Cocke, Steven,
%@affiliation Instituto Nacional de Pesquisas Espaciais (INPE)
%@affiliation Instituto Nacional de Pesquisas Espaciais (INPE)
%@affiliation Florida State University (FSU)
%@electronicmailaddress
%@electronicmailaddress haroldo.camposvelho@inpe.br
%B International WMO Symposium on Data Assimilation, 7
%C Florianópolis, SC
%8 11-15 Sept.
%X Data assimilation (DA) is an essential process for the operational prediction centers, due to uncertainties associated to the forecasting model. Supervised artificial neural network (NN) is the DA method applied to an Atmospheric General Circulation Model (AGCM) used in Florida State University (FSU), USA. The NN is trained to have similar performance to the Local Ensemble Transform Kalman Filter (LETKF). The NN is self-configured, as a result of minimizing an optimization problem. There are three factors in the cost function: training error, generalization error, and NN complexity. The optimum solution for the NN configuration is found by using a new meta-heurisc named MCPA (Multi-Particle Collision Algorithm). The DA experiment was carried out on the FSU Global Spectral Model (FSUGSM), a multilevel spectral primitive equation model at resolution T63L27. Similar results for DA are obtained by NN and LETKF, but the NN scheme is dozens times faster than the ensemble method.
%@language en