Multimodel Assessment of Regional Surface Temperature Trends: CMIP3 and CMIP5 Twentieth-Century Simulations

Regional surface temperature trends from phase 3 of the Coupled Model Intercomparison Project (CMIP3) and CMIP5 twentieth-century runs are compared with observationsat spatial scales ranging from global averages to individual grid pointsusing simulated intrinsic climate variability from preindustrial control runs to assess whether observed trends are detectable and/or consistent with the models' historical run trends. The CMIP5 models are also used to detect anthropogenic components of the observed trends, by assessing alternative hypotheses based on scenarios driven with either anthropogenic plus natural forcings combined, or with natural forcings only. Modeled variability is assessed via inspection of control run time series, standard deviation maps, spectral analyses, and low-frequency variance consistency tests. The models are found to provide plausible representations of internal climate variability, although there is room for improvement. The influence of observational uncertainty on the trends is assessed and is found to be generally small in comparison with intrinsic climate variability. Observed temperature trends over 1901-2010 are found to contain detectable anthropogenic warming components over a large fraction (about 80%) of the analyzed global area. In about 70% of the analyzed area, the modeled warming is consistent with the observed trends; in about 10% it is significantly greater than simulated. Regions without detectable warming include the high-latitude North Atlantic Ocean, the eastern United States, and parts of the eastern and northern Pacific Ocean. For 1981-2010, the observed warming trends over only about 30% of the globe are found to contain a detectable anthropogenic warming: this includes a number of regions within about 40 degrees-45 degrees of the equator, particularly in the Indian Ocean, western Pacific, South Asia, and tropical Atlantic.