GCM simulations of a future climate: How does the skill of GCM precipitation simulations compare to temperature simulations?

Which General circulation model (GCM) is more accurate? This is a question that has been addressed by many, using a range of assessment criteria and specified regions and time periods. The question we seek to address in this paper is not related to the relative skills of individual GCMs, but their collective skill at simulating a range of commonly used GCM hydroclimatic variable outputs. Hence, we seek to answer questions such as - Are GCM simulated temperatures more accurate than surface level pressures? How poor is the GCM skill at simulating rainfall compared to more stable variables such as temperature or wind speed? And how does this skill vary with region and distance from the coast? The Variable Convergence Score (VCS) was used to rank hydroclimatic variables based on the coefficient of variation of the ensemble of all models. The VCS is a simple methodology that allows a quantitative assessment of the performance of the models for different hydroclimatic variables. The skill score methodology has been applied to the outputs of multiple GCMs for a range of hydroclimatic variables and future emission scenarios to provide a relative ranking of the performance of the models over Australia. The methodology would be applicable for any region or any variable of interest available as a GCM output. The variation of model convergence with distance from the coast was examined. It was found for some variables such as temperature, specific humidity and precipitable water that the agreement of the GCMs in their future projections decreases for areas that are further inland. For other variables such as longwave radiation and wind speed, distance from the coast is not a good indicator of model agreement. For these variables there is a strong north-south gradient for model convergence. The effects of spatial averaging on model convergence were also assessed using the VCS. As expected, the spread of model projections lies closer to the multi-model ensemble mean for increasing levels of spatial averaging. This improvement in skill is more pronounced for variables such as wind speed that show pronounced regional variations. Variables for which the models consistently agree (e. g temperature, surface pressure) or disagree (precipitation) do not show as strong improvement in model convergence for larger spatial scales. The VCS has been shown to provide information to researchers and policy makers on how much agreement from GCMs we can expect in time and space.