Evaluation of the CMIP3 and CMIP5 climate models' simulated precipitation over South America

CAMILLONI, INÉS; GULIZIA, CARLA

Exeter

Workshop; 4th WGNE Workshop on systematic errors in weather and climate models; 2013

Met Office

The main objective of the present study is to compare the ability of 19 Global Climate Models (GCMs) from the CMIP3 and CMIP5 intercomparison projects to represent the summer, winter and annual precipitation mean fields in South America for the period 1960-1999. Moreover, three sub-regions are particularly evaluated which correspond to the continental core of the monsoon region, the continental area of the South Atlantic Convergence Zone and the southern sector of SESA region. Observed precipitation data used for the validation come from the Climate Research Unit (CRU) monthly dataset TS3.0. Various statistical analysis are presented to identify the more adequate GCMs to represent present climate in the study region, as well as to detect if there have been any improvements in the new generation of GCMs relative to the previous one. The comparison between the biases of the CMIP3 and CMIP5 ensembles at the continental scale shows as the most distinctive feature that summer precipitation exhibits a better performance for the CMIP5 ensemble mean than the CMIP3 one. However, this new generation of GCMs still has difficulties to represent the precipitation in northeastern Brazil, the regions over and near the Andes Mountains and the south end of the continent south of 40ºS. Moreover, the median of the normalized root-mean square error (RMSE) of CMIP5 models is higher for summer, winter and the annual mean than for the CMIP3 models not showing a necessarily improvement in time of rainfall representation for the analyzed region considering at least these subsets of GCMs. However, the error of the ensemble mean is lower for the new generation of models for the winter and annual cases. Furthermore, for both CMIP3 and CMIP5, the models’ ensembles agree with observations better than the corresponding median of all the calculated normalized RMSEs. Additionally, the MRI-CGCM2.3.2 model in the CMIP3 analysis is superior to the multimodel ensemble while there is no individual model in the set of CMIP5 GCMs analyzed exhibiting the same superiority. Furthermore, results from the regional analysis show that the distance between the median of CMIP3 GCMs and the mean observed precipitation differs in the three studied regions and among the different seasons although in all cases GCMs underestimate precipitation. Results from the same analysis but for CMIP5 models show considerable improvements for both summer and annual precipitation.