Assessing inter- to multi-decadal variability in climate model simulations

Climate model simulations of surface temperature over the continental US are evaluated using data from the Coupled Model Intercomparison Project 3 (CMIP3). Data is analyzed for the period 1951-1999 to rank and quantify model performance. Models are assessed using the Mann-Whitney Z (MWZ) analysis method that identifies significant warm and cool periods within a time series. The MWZ method emphasizes a model's skill in reproducing inter- to multi-decadal variability while simultaneously discounting unforced interannual noise in the climate system. Through comparisons with observations, this technique allows for a single score to be produced associated with each model's ability to replicate annual or seasonal temperature variations. Rankings of each of the CMIP3 simulations include uncertainty estimates informed by using multiple ensemble members, multi-model averages and unforced simulations. Our results demonstrate that the MWZ method is an effective diagnostic for studying inter- to decadal-scale variability and is much preferable to the calculation of standard linear trends. These results also suggest that similar analysis could be an important tool in the upcoming evaluation of the CMIP5 simulations.