Evaluation of the AR4 CMIP3 and the AR5 CMIP5 Model and Projections for Precipitation in Northeast Brazil

International audienceThis article compares the sensitivity of IPCC CMIP3-AR4 and CMIP5-AR5 models used on the latest reports from the Intergovernmental Panel on Climate Change (IPCC) in representing the annual average variations (austral summer and autumn) on three regions in Northeastern Brazil (NNEB) for the periods 1979–2000 using the CMAP (Climatology Merged Analysis of Precipitation) data as reference. The three areas of NNEB chosen for this analysis were the semiarid, eastern, and southern regions. The EOF analysis was performed to investigate how the coupled models resolve the temporal variability of the spatial modes in the Tropical Atlantic Sea Surface Temperature (SST), which drives the interannual variations of the rainfall in the Northeastern Brazil. CMIP3-AR4 and CMIP5-AR5 models presented a good representation of the annual cycle of precipitation. Results from correlation and mean absolute error analysis indicate that both CMIP3 and CMIP5 models produce large errors and barely capture the interannual rainfall variance during austral summer and autumn in Northeast Brazil, this features is closely related to the poor representation of the modes of SST variability in the Tropical Atlantic Ocean. For the summer and autumn rainfall projections (2040–2070) in the semiarid region, there was no convergence between the CMIP3 and CMIP5 models. During the summer and autumn in the eastern sector, both the CMIP3 and CMIP5 models projected rainfall above the mean for the 2040–2070 period