QUANTIFYING UNCERTAINTIES IN GLOBAL AND NORTH AMERICAN REGIONAL CLIMATE CHANGE PROJECTIONS USING A MULTI-THOUSAND MEMBER GLOBAL CLIMATE MODEL PERTURBED PHYSICS ENSEMBLE

Abstract

Information on the uncertainties in projections of future climate change from global climate models (GCMs) is vital for their effective use across a wide range of applications, including their increasing role in driving regionally downscaled models for higher resolution output useful to local impacts studies (e.g., hydrologic, ecosystems, agricultural). To better estimate GCM uncertainties, a multi-thousand member perturbed-physics ensemble (PPE) of climate simulations was assessed to quantify uncertainties in future climate change projections for the globe and North American region. The simulations were generated through the distributed computing project Climateprediction.net (CPDN), a joint effort between the UK Met Office Hadley Centre and Oxford University, where thousands of simulations were run on PCs across the globe, each running a different version of the Hadley Centre-based HadCM3L coupled atmosphere-ocean GCM with variations to their model physics parameters.

The large PPE was able to model many observed features in the Earth’s climate system and climate indices were found to be sensitive to changes in the model’s physics parameters with cloud physics parameters being of most importance. The PPE was constrained using observational performance and parameter sensitivity assessments and it was found that the constrained ensembles reduced both the ensemble mean and uncertainty range of the initial ensemble. Results were compared to CMIP3 and CMIP5 ensembles and the CMIP ensembles were found to underestimate the full range of uncertainties in physics parameters, thus indicating the usefulness of large PPEs to inform users of GCM output of the full range of model parameter and structural uncertainty.