In the latest model intercomparison, most coupled ocean-atmosphere general circulation models (GCMs) continue to suffer serious errors in their simulations of tropical Atlantic climate. Two errors common to all the models are 1) the failure to develop an eastern cold tongue on the equator, associated with a westerly surface wind bias and 2) an erroneous southward shift of the intertropical convergence zone (ITCZ) associated with a warm bias south of the equator. Such errors in the mean state seriously limit the models’ skills in seasonal prediction and future climate projection. Recent analyses of simulations in the IPCC Fourth Assessment Report (AR4) data archive hint that tropical Atlantic biases in coupled models originate from their atmospheric component. Specifically, the westerly wind error on the equator and the double ITCZ bias are already present during boreal spring in atmospheric simulations forced by observed SST. The spring westerly error depresses the thermocline and prevents the cold tongue from developing in the equatorial Atlantic in the subsequent season. Furthermore, studies show that simulated spring rainfall is deficient and excessive over equatorial South America and Africa, respectively, suggesting that continental precipitation biases are key to the westerly wind error over the equatorial Atlantic. The PIs propose to identify the sources of tropical Atlantic biases and investigate how they develop in coupled GCMs using a suite of diagnostic and modeling studies. First they will develop metrics to evaluate coupled model simulations of tropical Atlantic climate and identify common sources of error. They will apply them to the AR4 output as well as the upcoming AR5 simulations as the latter become available by early 2010. They will use the NOAA/GFDL Climate Model CM 2.1 and its atmospheric component to test the hypothesis that continental rainfall biases cause the models failure to develop the equatorial cold tongue, by perturbing convective heating over tropical South America.