Mao, J. and alan Robock 1998: Surface air temperature simulations by AMIP general circulation models: volcanic and ENSO signals and systematic errors.  Journal of Climate, 11, 1538–1552.
Thirty surface air temperature simulations for 1979–88 by 29 atmospheric general circulation models are analyzed and compared with the observations over land. These models were run as part of the Atmospheric Model Intercomparison Project (AMIP). Several simulations showed serious systematic errors, up to 4°–5°C, in globally averaged land air temperature. The 16 best simulations gave rather realistic reproductions of the mean climate and seasonal cycle of global land air temperature, with an average error of 0.9°C for the 10-yr period. The general coldness of the model simulations is consistent with previous intercomparison studies. The regional systematic errors showed very large cold biases in areas with topography and permanent ice, which implies a common deficiency in the representation of snow-ice albedo in the diverse models. The SST and sea ice specification of climatology rather than observations at high latitudes for the first three years (1979–81) caused a noticeable drift in the neighboring land air temperature simulations, compared to the rest of the years (1982–88). Unsuccessful simulation of the extreme warm (1981) and cold (1984–85) periods implies that some variations are chaotic or unpredictable, produced by internal atmospheric dynamics and not forced by global SST patterns.

Among the 16 best simulations, 8 reproduced the dominant El Niño–Southern Oscillation (ENSO) mode in the 10-yr period, which includes the 1982–83 and 1986–87 warm episodes and the 1988 cold episode. On the average, the ENSO mode explains about 30% of the total variance in surface air temperature fluctuation and has a 2-month lag from the Southern Oscillation index. In this mode, North America displays a Pacific–North American–like anomaly pattern, but Eurasia gave little response to warm SSTs in the eastern equatorial Pacific, in good agreement with results based on historical data.

The special design of the AMIP experiment provides a unique opportunity to estimate the effects of the El Chichón volcanic eruption in spring 1982, which was not included in the model forcing. Comparison of the simulations with data delineated a visible global cooling in the first months following the El Chichón eruption, in addition to the cooling from the volcanic eruption of Nyamuragira in December 1981, due to the reduction of incoming solar radiation by volcanic aerosols. However, the mean climate shift in the AMIP experiment due to the forcing data discontinuity at the end of 1981 made the quantitative estimate of El Chichón global cooling influence impossible. The contrast between the simulated ENSO signal and observations shows that the major warming over the northern continents during the 1982/83 winter (DJF) is not an ENSO-like signal. Instead it is most likely a pattern resulting from the enhanced polar vortex produced by a larger pole-to-equator temperature gradient. This gradient was due to the larger absorption of radiation in low latitudes by the El Chichón volcanic sulfate aerosols in the stratosphere. These results suggest that during the Northern Hemisphere wintertime, the stratospheric polar vortex has substantial influence on surface air temperature fluctuations through its effects on vertically propagating planetary waves of the troposphere, and imply that current GCMs are deficient in simulation of stratospheric processes and their coupling with the troposphere.