Tropical Inter-basin Interactions and Their Misrepresentation in Climate Models

Much of the potential for regional climate predictability around the globe rests on the predictability of tropical SSTs and their response to radiative forcing. Model misrepresentations of the magnitudes and patterns of tropical SST changes can thus have important consequences. We have estimated the sensitivities of global teleconnections to tropical SST changes (and equivalently, also SST errors) at different tropical locations from the global responses to 42 regularly spaced localized SST anomaly “patches” prescribed as anomalous boundary conditions in large ensembles of NCAR and ECHAM5 atmospheric model integrations. In effect, such integrations yield a “Fuzzy Green’s Function” of the global atmospheric response to tropical SST forcing. The dominant EOF patterns of the 42 responses, and the relative magnitudes with which they are excited by the individual patches, determine the dominant pairs of response and forcing (formally, the left and right singular vectors of the Green’s function operator). The dominant SST patterns can be interpreted both as major sources of regional climate predictability and as patterns of tropical SST changes and errors to which the global teleconnections are most sensitive.

The dominant SST sensitivity pattern (which is different in winter and summer) has a very different structure from the dominant ENSO pattern of observed interannual SST variability, and has the largest magnitude but opposite signs in the Western Pacific and Eastern Indian ocean basins. This makes it important for models to predict SSTs accurately in these basins, which they currently do not. Specifically, on seasonal scales, the National Multi-Model Ensemble (NMME) models used for seasonal predictions extend the predicted central Pacific SST anomalies during ENSO events too far west into the warm pool. On decadal and longer scales, the CMIP3 and CMIP5 models underestimate the magnitude and misrepresent the spatial variation of tropical SST changes in these sensitive areas of forcing global teleconnections.

To clarify the origin of these errors, we performed an extensive analysis of inter-basin SST feedbacks among 8 tropical ocean basins (two each in the Indian and Atlantic, four in the Pacific ocean) using six long-term observational SST datasets and all available CMIP3 and CMIP5 simulations of the 20^th century. The dominant model feedback error was found to be in the feedback of the western Pacific SSTs on the eastern Indian ocean SSTs.