Privalsky V., Yushkov V. Validation of CMIP5 models for the contiguous United States // Atmosph. Sci. Lett. 2015, Volume 16, Issue 4, pp. 461–464. DOI: 10.1002/asl.582

Abstract. Major statistical characteristics – trend rates, mean values, standard deviations, probability densities, autoregressive model orders, persistence criteria, and spectra – of annual surface temperature over the contiguous United States from 1889 through 2005 are compared with respective characteristics of 47 time series generated within the CMIP5 historical experiment. The observed and most simulated time series are Gaussian. Most autoregressive orders, persistence criteria, and spectra of simulated time series are close to what is found in nature. Although the multi-model mean value is not biased, individual models can err by almost 3.5° C. In addition, the models exaggerate linear trend rates and temperature variance is overestimated.