Status: this preprint has been withdrawn by the authors.
The implications of initial model drift for decadal climate predictability using EC-Earth
Abstract. The large heat capacity of the ocean as compared to the atmosphere provides a memory in the climate system that might have the potential for skilful climate predictions a few years ahead. However, experiments so far have only found limited predictability after accounting for the deterministic forcing signal provided by increased greenhouse gas concentrations. One of the problems is the drift that occurs when the model moves away from the initial conditions towards its own climate. This drift is often larger than the decadal signal to be predicted. In this paper we describe the drift occurring in the North Atlantic Ocean in the EC-Earth climate model and relate it to the lack of decadal predictability in that region. While this drift may be resolution dependent and disappear in higher resolution models, we identify a second reason for the low predictability. A subsurface heat content anomaly can only influence de atmosphere if (deep) convection couples it to the surface, but the occurrence of deep convection events is random and probably mainly determined by unpredictable atmospheric noise.
This preprint has been withdrawn.
How to cite. Sterl, A.: The implications of initial model drift for decadal climate predictability using EC-Earth, Ocean Sci. Discuss. [preprint], https://doi.org/10.5194/os-2016-27, 2016.
Received: 10 May 2016 – Discussion started: 30 May 2016
Attempts at decadal climate predictions have largely failed so far. We here investigate two reasons. The first is the large drift occurring after initialization, and which masks the signal to be predicted, and the second is the fact that a sub-surface heat signal has to be connected to the surface to influence the atmosphere. However, vertical mixing in the ocean to expose a sub-surface anomaly to the surface depends on unpredictable atmospheric variability.
Attempts at decadal climate predictions have largely failed so far. We here investigate two...