The relative performance of Australian CMIP5 models based on rainfall and ENSO metrics

We assess the performance of 30 CMIP5 and two CMIP3 models using metrics based on an all-Australia average rainfall and NINO3.4 sea surface temperatures (SSTs). The assessment provides an insight into the relative performance of the models at simulating long-term average monthly mean values, interannual variability and the seasonal cycles. It also includes a measure of the ability to capture observed rainfall-NINO3.4 SST correlations. In general, the rainfall features are reasonably simulated and there is relatively little difference amongst the models but the NINO3.4 SST features appear more difficult to simulate as evidenced by the greater range in metric scores. We find little evidence of consistency in the sense that a relatively good metric score for one feature does not imply a relatively good score for another related (but independent) feature. The assessment indicates that more recent models perform slightly better than their predecessors, especially with regard to the NINO3.4 metrics. We also focus on the ability of models to reproduce the observed seasonal cycle of rainfall-SST correlations since this is a direct indicator of a model's potential utility for seasonal forecasting over Australia. This indicates some relatively good models (CNRM, HadGEM2-ESM, MPI-ESM-LR and MPI-ESM-MR) and some relatively poor models (CSIRO-Mk3.5, FGOALS, GISSE2- HP1 and INMCM4). We find that the ACCESS1.3 and CSIRO-Mk3.6 models rank as near median performers on this metric and represent improvements over their predecessors (ACCESS1.0, CSIRO-Mk3.0 and CSIRO-Mk3.5).