Refinement of the daily precipitation simulated by the CMIP5 models over the north of the Northeast of Brazil

• Autor(a) principal: Silva, Gyrlene A. M. da
• Data de Publicação: 2015
• Outros Autores: Mendes, David
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UFRN
• Texto Completo: https://repositorio.ufrn.br/jspui/handle/123456789/29242
• topic: Artificial neural network; Sea surface temperature; Precipitation; CMIP5 models; Intertropical convergence zone; Multiple linear regression

The ability of the Artificial Neural Network (ANN) and the Multiple Linear Regression (MLR) in reproducing the area-average observed daily precipitation during the rainy season (Feb–Mar–Apr) over the north of the Northeast of Brazil (NEB) is examined. For the present climate of Dec-Jan-Feb from 1963 to 2003 period these statistical models are developed and validated using the observed daily precipitation and simulated from the historical outputs of four models of the fifth phase of the Coupled Model Intercomparison Project (CMIP5). The simulations from all the models during DJF and FMA seasons have an anomalous intensification of the ITCZ and southward displacement in comparison with the climatology. Correlations of 0.54, 0.66, and 0.66 are found between the simulated daily precipitation of the CCSM4, GFDL_ESM2M, and MIROC_ESM models during DJF season and the observed values during FMA season. Only the CCSM4 model displays a slightly reasonable agreement with the observations. A comparison between the statistical downscaling using the nonlinear (ANN) and linear model (MLR) to identify the one most suitable for the analysis of daily precipitation was made. The ANN technique provides more ability to predict the present climate when compared to MLR technique. Based on this result, we examined the accuracy of the ANN model in project the changes for the future climate period from 2055 to 2095 over the same study region. For instance, a comparison between the daily precipitations changes projected indirectly from the ANN during Feb–Mar–Apr with those projected directly from the CMIP5 models forced by RCP 8.5 scenario is made. The results suggest that ANN model weights the CMIP5 projections according to the each model ability in simulating the present climate (and its variability). In others, the ANN model is a potentially promising approach to use as a complementary tool to improvement of the seasonal numerical simulations