Reducing uncertainty in arctic climate projections

Global climate models vary in both their ability to reproduce recent observations and climate variability and their projected responses to increasing atmospheric greenhouse gases. We present a quantified assessment of CMIP3 global climate model performance for primary atmospheric variables, seasonally integrated over several Arctic domains. Model biases based on root mean squared errors of temperature, sea level pressure and precipitation are determined from corresponding observational reanalyses for the most recent two decades. The performance analysis demonstrates consistency between variables in the model biases (i.e., models that tend to perform well for one variable are usually among the top performers for the other variables). There is general consistency across different Arctic sub-regions, as well. The skill shown by models in simulating the recent climate can be used to weight the various future model projections in an attempt to reduce the uncertainty in Arctic climate projections by augmenting the influence of the better performing models and limiting the influence of the weaker performing models. When this strategy is applied, weighted model projections for the 21st-century Arctic climate show greater Arctic warming, lower sea level pressures, and enhanced precipitation when compared to corresponding constant-weight composite mean projections. Similarly, weighting model projections according to their skill in reproducing observed cloud and surface radiation properties produces reduced summer downwelling surface shortwave solar flux and increased downwelling longwave flux at the surface when compared to corresponding unweighted composite mean projections.