Temperature Estimates from Models and Observations

Abstract

Seventy estimates of annual global mean near-surface temperature from models have been compared with an estimate from observations, for the period 1959–1999. Correlations coefficients ranged from 0.35 to 0.85. Only 3 were better than the correlation with the level of atmospheric carbon dioxide. That suggests that the fluctuations of model values do not mirror “natural variation”, a conclusion reinforced by the lack of association between one year's estimate and that for the previous year from observation, with that from models. Regression slopes of observation on model ranged from 0.26 to 1.16; that does not constitute strong support for the “green-house gas” explanation of global warming. Model outputs can be modified to predict global temperatures but then the model output is logically no different from the empirical relationship with atmospheric carbon dioxide, which is as good as the best of the models and is transparent.