Do Statistical Pattern Corrections Improve Seasonal Climate Predictions in NMME Models?

The North American Multimodel Ensemble (NMME) contains global predictions of SST, surface air temperature, precipitation and other variables from 8 or more state-of-the-art coupled general circulation models (Kirtman 2014). In phase one of the NMME project, hindcasts of monthly average climate extending to up to 12 months into the future were created, spanning the 1982-2010 period. Here we seek to determine whether a commonly used multivariate statistical methods—namely, CCA—can improve the temporal anomaly correlation skill of the individual models, with the goal of improving the predictions of the multimodel ensemble. The anomaly correlation is used as the main metric because it measures the ability to reproduce the interannual variability of the climate, regardless of the presence of systematic errors that can be treated locally using simpler statistical methods.