Assimilação de dados com redes neurais artificiais em modelo de circulação geral da atmosfera / Data assimilation with artificial neural networks in atmospheric general circulation model

AUTOR(ES)
DATA DE PUBLICAÇÃO

2010

RESUMO

Weather forecasting systems require a model for the time evolution and an estimate of the current state of the system. The numerical weather prediction (NWP) incorporates the equations of atmospheric dynamics with physical process and it can predict the future state of the atmosphere. Data assimilation provides such an initial estimate of the atmosphere where it combines information from observations and from a prior short-term forecast producing an current state estimate. This work investigated the approach of data assimilation with Artificial Neural Networks (ANN). The short-term predictions are from a global primitive equation model, the SPEEDY model extit{Simplified parameterizations, primitive-Equation Dynamics}, simplified physical processes of an atmospheric general circulation with resolution in tridimensional coordinates. Molteni (2003) showed that the SPEEDY model has similar characteristics to the state-of-art atmospheric models. For the data assimilation scheme, it applied a supervised ANN Multilayer Perceptron to emulate the analysis results for extit{Local Ensemble Transform Kalman Filter} (LETKF). LETKF is an approximation of Kalman filter, with Monte-Carlo ensemble of short-term forecasts to estimate the forecast model error covariances. The method using RNA in this work can be described as a process of data assimilation, where the ANN trained after obtaining the results, like a function of the state model SPEEDY and its synthetic observations. The strategy of ANN supervised training, the implementations of networks and the model and the observations are presented. The main emphasis of this technique is the computational speed in obtaining the initial condition for state model that accelerates the whole process of numerical weather prediction. The numerical results demonstrate the effectiveness of this ANN technique in atmospheric data assimilation because these have been very close to the results compared with LETKF data assimilation results. The simulations demonstrate the great advantage in using neural networks: the best computational performance.

ASSUNTO(S)

assimilação de dados redes neurais artificiais previsão numérica de tempo perceptron de multiplas camadas modelo de circulação geral data assimilation numerical weather predicition artificialneural networks multilayer perceptron general circulation model

ACESSO AO ARTIGO

Documentos Relacionados