Stringent Comparison and Evaluation of CMIP5 Model Sea Surface Temperatures

Control simulations of sea surface temperatures by 19 CMIP5 coupled ocean-atmosphere general circulation models (CGCMs) have been evaluated in comparison with the Extended Reconstructed Sea Surface Temperature (ERSST) version 4 monthly sea surface temperature data for the period of 1861-2005 based on CSEOF analysis.  Each CSEOF loading vector (CSLV) and principal component (PC) time series derived from individual CMIP5 models are compared with those of the ERSST data for the first four modes.  The first CSEOF mode (the seasonal cycle) is simulated accurately; average pattern correlation is 0.901 and temporal correlation is 0.999. The accuracy of CSEOF modes derived from models quickly degrades as the mode number increases.

In an effort to improve the CMIP5 sea surface temperature data for practical uses, regression analysis is conducted in CSEOF space for each mode.  Correlation between the CSEOF modes derived from the ERSST v.4 and the regressed CSEOF patterns is significantly improved beyond that of the best CMIP5 model.  Spectral contents of the PC time series of the regressed CSEOF modes are compared with those of the ERSST v.4 to assess the model accuracy in simulating SST variability.