On the cold start problem in transient simulations with coupled atmosphere-ocean models

Abstract

Finite computer resources force compromises in the design of transient numerical experiments with coupled atmosphere-ocean general circulation models which, in the case of global warming simulations, normally preclude a full integration from the undisturbed pre-industrial state. The start of the integration at a later time from a climate state which, in contrast to the true climate, is initially in equilibrium then induces a cold start error. Using linear response theory a general expression for the cold start error is derived. The theory is applied to the Hamburg CO₂ scenario simulations. An attempt to estimate the global-mean-temperature response function of the coupled model from the response of the model to a CO₂ doubling was unsuccessful because of the non-linearity of the system. However, an alternative derivation, based on the transient simulation itself, yielded a cold start error which explained the initial retardation of the Hamburg global warming curve relative to the IPCC results obtained with a simple box-diffusion-upwelling model. In the case of the sea level the behaviour of the model is apparently more linear. The cold start error estimations based on a CO₂ doubling experiment and on an experiment with gradually increasing CO₂ (scenario A) are very similar and explain about two thirds of the coupled model retardation relative to the IPCC results.