Jon Robson : Understanding the performance of a decadal prediction system : UoR, Dept Of Meteorology

See also

Understanding the performance of a decadal prediction system

The UK Met Office has recently developed a Decadal Prediction System (DePreSys), which is initialised from the observed state of the climate system in order to constrain uncertainty in ``near term'' projections of climate change. This thesis aims to understand the decadal predictions made with DePreSys with a focus on the Atlantic sector.

First, an analysis of the observed decadal variability in the ocean is presented, and a significant warming of the North Atlantic in the 1990s is highlighted. By comparing the observed evolution with ocean model experiments, it is shown that the warming can be understood as a lagged response to the persistently positive North Atlantic Oscillation (NAO) forcing that preceded the warming.

An extended analysis of the mean skill of DePreSys, relative to predictions that do not assimilate information (NoAssim), is presented. It is shown that the mean bias of both DePreSys and NoAssim significantly affects the mean skill, and that the original published comparison of skill was in error. It is further shown that many different choices and processes affect the mean bias, including: sampling uncertainty, model drift, unbalanced initial conditions, and model errors, Moreover, it is demonstrated that much, but not all, of the increased skill of DePreSys at long lead times, relative to NoAssim, is due to a reduction of the bias, rather than prediction of the low frequency variability.

The rapid warming of the North Atlantic in the 1990s is analysed as a specific case study for decadal prediction. This analysis reveals specific limitations of the anomaly assimilation method: hindcasts can be sensitive to the model climatology, and errors can also arise from the non-linear equation of state.

Finally, based on the results, this thesis gives specific recommendations for the future development of DePreSys, and for decadal climate prediction in general.

Thesis - final version