We propose to generate a probabilistic decadal prediction of North American climate for the period 2011- 2020. The methodology will involve large ensemble integrations from multiple atmospheric general circulation models (AGCMs) driven by various, plausible trajectories of global sea surface temperature (SST) over the next decade. The latter will be derived from both uninitialized and initialized coupled climate model experiments. We are motivated by evidence that initial state information from the oceans is a key skill source in nascent attempts at decadal prediction. Furthermore, attribution studies have established that key features of observed regional decadal climate variability have been largely driven by variations in global SST. Multimodel large ensemble methods are proposed in order to generate meaningful statistics of regional climate change on decadal timescales, thereby overcoming current limitations of coupled model prediction efforts resulting from small ensemble size. A focus of the project will be to derive an estimate of the potential skill for predicting the evolution of North American decadal climate. We will perform a comprehensive analysis of North American decadal variability from 1900 to present using existing large ensembles of AGCMs driven by observed SST variability and coupled models driven by estimates of observed changes in greenhouse gas, aerosol, solar and volcanic forcing. These will be diagnosed to quantify the variances of North American climate driven by external forcing, internal coupled ocean-atmosphere variations, and internal atmospheric variations alone. As a prelude to our proposed 2011-2020 predictions, we will analyze “decadal hindcasts” over the past twenty years using initialized coupled models and AGCMs to assess perfect-model skill and sources of uncertainty in decadal predictions and predictability.