Using redundancy analysis to improve dynamical seasonal mean 500 hPa geopotential forecasts

In this study, we evaluate and compare 500 hPa geopotential height hindcast skill in two large dynamical hindcast experiments performed with the Canadian Climate Centre second generation general circulation model (GCM). In one hindcast experiment, seasonal hindcasts are made from lagged initial conditions observed at the beginning of each season. The sea-surface temperatures (SSTs) required by the model during each forecast period are forecast by persisting the SST anomalies observed during the month just prior to the forecast period. The second hindcast experiment consists of an ensemble of simulations in which continuously evolving observed SSTs are specified at the model's lower boundary. These hindcasts do not benefit from re-specification of the initial state at the beginning of each season, but they do enjoy the benefit of 'perfect' SST forecasts. We also demonstrate the use of a regression technique, called redundancy analysis (RA), for statistically improving the skill of both types of dynamical hindcast. The results indicate that specification of the initial state at the beginning of each season adds skill to the seasonal hindcasts, even though SSTs at the lower boundary are imperfectly specified. We also find that the model can predict the mean state of the North Atlantic Oscillation (NAO) with some skill in boreal winter and spring when the initial state is specified at the beginning of each season. The results also indicate that statistical post-processing with the RA technique improves the (cross-validated) skill of both types of dynamical hindcast. Copyright (C) 2001 Royal Meteorological Society.