5. Conclusions

In this study, we evaluated the efficiency of the MLP-DA in an atmospheric data assimilation context with a 3D global model. The MLP-DA is able to emulate systems for known data assimilation scheme. For the present investigation, the MLP-DA approach is used to emulate the LETKF method, which is designed to improve the computational performance. The another experiments with the same methodology can be found in [7, 8].

The NN learned the whole process of the LETKF scheme of data assimilation through training process. The results for the MLP-DA analyses are very close to the results obtained from the LETKF data assimilation for initializing the SPEEDY model forecast, i.e., the analyses obtained with MLP-DA are similar to analyses computed by the LETKF. The difference between MLP-DA and LETKF analyses to surface pressure fields belongs to interval [5, 5] hPa. However, the computational performance of the set of 30 NN is better than LETKF scheme. The MLP-DA accelerates the LETKF data assimilation computation.

The application of the present NN data assimilation methodology is under investigation at the Center for Weather Prediction and Climate Studies (Centro de Previsão de Tempo e Estudos Climáticos-CPTEC/INPE) with operational numerical global model and real observations. After investigation with Florida State University model made in 2014, the results are found in [41, 42].
