Regional climate predictability from multiple global GCMs in simulations of the 20th century and multimodel regional forecasts of future climate change

Under the auspices of the IPCC fourth annual report, simulations of the 20th century and several scenarios for future climate change have been produced by a set of GCMs developed by many modeling groups. Though the ability of these models to reproduce the large-scale circulation of the last century generally has been analyzed, the ability of these climate models to produce reasonable simulations of regional climates and predictions of future regional climate variability needs further study (IPCC, 2001). Krishnamurti et al. (2000) found that use of multiple models in generation of seasonal climate forecasts increases the predictability relative to any individual model. Following this philosophy, we use a multimodel ensemble of simulations of the 20th century to assess regional predictability from global GCM runs. Using the conditional entropy of model-generated probability distributions to observational and reanalysis data probability distributions from several sources (VASClimO precipitation, Reynolds SST and ECMWF ERA40 reanalysis atmospheric circulation), we calculate the information content of GCM simulations of the climate variability of various regional variables. Using wavelet and wavelet coherence transforms for the model and observational data, we consider the predictability of regional climate variability on timescales from interannual to multidecadal, so that predictability of climate change is considered in both the mean and variability. Metrics of the predictability of climate change in the 20th century are used to create multimodel ensemble forecasts of future regional climate change.