Investigating the mechanisms of seasonal ENSO phase locking bias in the ACCESS coupled model

The mechanisms of coupled model bias in seasonal ENSO phase locking are investigated using versions 1.0 and 1.3 of the CSIRO-BOM ACCESS coupled model (hereafter, ACCESS1.0 and ACCESS1.3, respectively). The two ACCESS coupled models are mostly similar in construction except for some differences, the most notable of which are in the cloud and land surface schemes used in the models. ACCESS1.0 simulates a realistic seasonal phase locking, with the ENSO variability peaking in December as in observations. On the other hand, the simulated ENSO variability in ACCESS1.3 peaks in March, a bias shown to be shared by many other CMIP5 models. To explore the mechanisms of this model bias, we contrast the atmosphere-ocean feedbacks associated with ENSO in both ACCESS model simulations and also compare the key feedbacks with those in other CMIP5 models. We find evidence that the ENSO phase locking bias in ACCESS1.3 is primarily caused by incorrect simulations of the shortwave feedback and the thermocline feedback in this model. The bias in the shortwave feedback is brought about by unrealistic SST-cloud interactions leading to a positive cloud feedback bias that is largest around March, in contrast to the strongest negative cloud feedback found in ACCESS1.0 simulations and observations at that time. The positive cloud feedback bias in ACCESS1.3 is the result of a dominant role played by the low-level clouds in its modeled SST-cloud interactions in the tropical eastern Pacific. Two factors appear to contribute to the dominance of low-level clouds in ACCESS1.3: the occurrence of a stronger mean descending motion bias and, to a lesser extent, a larger mean SST cold bias during March-April in ACCESS1.3 than in ACCESS1.0. A similar association is found between the positive cloud feedback bias and the biases in spring-time mean descending motion and SST for a group of CMIP5 models that show a seasonal phase locking bias similar to ACCESS1.3. Significant differences are also found between the thermocline feedbacks simulated by ACCESS1.0 and ACCESS1.3 that appear to reinforce the seasonal ENSO phase locking bias in the latter model. We discuss a mechanism by which the thermocline feedback differences could arise from atmospheric forcing differences in the two models.