Estudo sobre a precipitação simulada no outono na região tropical da América do Sul através de downscaling dinâmico e previsão por conjunto

Abstract

This study presents an analysis of the performance of the method ensemble forecasting by Multiple Linear Regression Principal Component (RPC) to combine simulations with different configurations, performed with a regional climate model (downscaling dynamic technique) in order to simulate rainfall over tropical portion of South America. The analysis focused on two sub-regions of Brazil: Northeast Brazil (NEB) and Amazon (AMZ). The regional model used in the simulations was the RegCM4, forced by ERA-Interim data produced by the European Center for Medium- Range Weather Forecast (ECMWF). They analyzed 18 austral autumns (1991 until 2008). For the composition of a prediction ensemble, were used for training period 2/3 and 1/3 for validation, as holdout technique. The results were compared to daily precipitation data from the Global Precipitation Climatology Center (GPCC) and the average of the simulations (MA method), which is the method generally used for ensemble forecasting. The RegCM4 model showed potential to predict rainfall over the tropical region of AS. Moreover, different parameterization of the model can be modified to make it more effective. However, the good performance of the model combined with the RCP method of prediction ensemble increased accuracy compared to the MA method, with: less biased bias (average close to zero), while the MA method was biased (underestimated precipitation); greater capture the variability of GPCC data and moderate correlation with the GPCC data, while MA obtained weak correlation and greater mean square error. In addition, the reproduced PCR method improves the performance, the distribution of daily precipitation along the entire the period for both regions. Therefore, from the developed and implemented methodology yielded improvements in forecasting intensity of daily rainfall over the tropical region of South America.