CMIP5 Simulations of 20th-Century Pacific Northwest USA Climate
Rupp, D.E., Abatzoglou, J.T., Hegewisch, K.C. and Mote, P.W. 2013. Evaluation of CMIP5 20th century climate simulations for the Pacific Northwest USA. Journal of Geophysical Research: Atmospheres 118: 10,884-10,906.
In describing how work in this line of study has been progressing, Rupp et al. explain how they compared monthly temperature and precipitation projections from 41 CMIP5 GCMs with real-world observations for the 20th century, focusing on the United States Pacific Northwest (PNW) and surrounding regions.
Results indicated, first of all, that individually, "the models generated a wide range of values for many metrics," and that as a group they "performed less well as judged by the precipitation-based metrics. More specifically, they note (1) "a cursory comparison with 24 CMIP3 models [an earlier group of models] revealed few differences between the two generations of models with respect to the statistics analyzed," after which they go on to report (2) "the change in precipitation persistence as represented by the Hurst exponent was actually in the direction away from the observed value," (3) "the models generated a wide range of values for many metrics," (4) "the models as a group, performed less well as judged by the precipitation-based metrics," (5) "five observation data sets were 0.8°C warmer than the median of the simulated mean annual temperatures," (6) "all but one model generated more precipitation than observed," (7) "the amplitude of the seasonal cycle [of temperature] varied widely among models," (8) "the models generated a wide range of amplitudes of the seasonal precipitation cycle, with a handful of models severely under-simulating the strength of the seasonal variation," (9) "the multi-model mean generated a larger seasonal amplitude (by ~5°C) in southeastern Idaho than was evident in [the data]," (10) "simulated DTR [diurnal temperature range] tended to be about 2.5-3.5°C too low throughout the year compared to [the data]," (11) "with very few exceptions, the individual GCMs generated a DTR that was too small, irrespective of season," (12) "over much of western North America and over the ocean west of Mexico, the multi-model mean gave too much precipitation," (13) "there was no consistency in seasonal differences between simulations and observations, i.e., the seasons with greater observed warming were not those with greater simulated warming," (14) "overall, the CMIP5 models tended to produce too much interannual-to-decadal variability in PNW-averaged time series of temperature relative to the observations," (15) "in the case of precipitation, nearly all models generated less temporal variability than seen in observations," (16) "the simulated variances in general decreased too rapidly with increasing scale," and (17) "absent from the response ... was the tongue of observed positive (wetter) response that extends northward through eastern Oregon and Washington."
In terms of modeling both temperature and precipitation, which are the most basic of climatic parameters, it would appear that the CMIP5 GCMs are still a long, long way from satisfactorily doing what they were created to do.