An Experiment on the Sensitivity of a Global Circulation Model

Abstract

The growth of small errors in numerical models of the atmospheric circulation destroys the detailed predictive capability of those models within a few days. Despite the failure of the models to produce accurate local predictions, it was hypothesized that a change in the equator-to-pole temperature gradient would produce discernible effects in average conditions. This paper presents the results of an experiment to test this hypothesis. The Mintz-Arakawa model was started with a standard set of initial conditions and was run 60 days. The experiment was then replicated twice, with two independent sets of random temperature variations superposed on the temperatures at the σ = 0.25 and σ = 0.75 levels. At this point, the ice of the Arctic Ocean was replaced with water at the freezing temperature. Again the model was run, starting once with the standard initial conditions (other than the ice removal) and once each with the two sets of temperature “errors” added. For the four replications with temperature “errors,” the detailed predictive capability was lost after about 14 days. Yet an analysis of variance applied to the last 30-day zonal-average values of the three sets of ice-in/ice-out runs showed many significant changes in the general circulation.