Statistical Properties of Downscaled CMIP3 Global Climate Model Simulations

Spatial downscaling of global climate model projections adds physically meaningful spatial detail, and brings the results down to a scale that is more relevant to human and ecological systems. Statistical/empirical downscaling methods are computationally inexpensive, and thus can be applied to large ensembles of global climate model projections. Here we examine some of the statistical properties of a large ensemble of empirically downscale global climate projections. The projections are the CMIP3 global climate model projections that were performed by modeling groups around the world and archived by the Program for Climate Model Diagnosis and Intercomparison at Lawrence Livermore National Laboratory. Downscaled versions of 112 of these simulations were created on 2007 and are archived at http://gdo-dcp.ucllnl.org/downscaled_cmip3_projections/dcpInterface.html. The downscaling methodology employed, “Bias Correction/Spatial Downscaling” (BCSD), includes a correction of GCM biases relative to observations during a historical reference period, as well as empirical downscaling to grid scale of ~12 km. We analyzed these downscaled projections and some of the original global model results to assess effects of the bias correction and downscaling on the statistical properties of the ensemble. We also assessed uncertainty in the climate response to increased greenhouse gases from initial conditions relative to the uncertainty introduced by choice of global climate model.