Multiyear initial-value predictability of subsurface temperature in the North Atlantic Ocean and the impact of ensemble size

Abstract

Apart from the well-documented relationships between North Atlantic sea surface temperature and local climate, there exists a close link connecting subsurface temperature in the North Atlantic Ocean and large-scale oceanic circulation and climate on multiple time scales. However relatively less is known about the variation and predictability of subsurface temperature due mainly to lack of long-term observations, which motivates this study. Using a hierarchy of coupled climate models, this thesis focuses on multiyear initial-value predictability of subsurface temperature in North Atlantic Ocean and impact of ensemble size on prediction skill and predictability assessment. Analysis of outputs from two GCMs with different resolutions reveals that, horizontal development of subsurface temperature prediction uncertainty, which is characterized by large ensemble variance, is propagating clockwise in the subtropical North Atlantic while anticlockwise in the subpolar North Atlantic. The path along the Gulf Stream and the North Atlantic Current is the principle site where large prediction uncertainty emerges first. A further examination of the latitude depth space reveals that temperature prediction uncertainty propagates downward mainly in the subpolar North Atlantic and Southern Ocean. The downward development in the North Atlantic is thought to be associated with mode water subduction and deep water formation process. It is also discovered that ensemble size casts a substantial impact on prediction skill and assessment of initial-value predictability of North Atlantic Oceanic climate. Increasing ensemble size leads to a significant reduction of biases of initial-value predictability assessment. Though there is slight difference, the minimal ensemble sizes required to make a ‘steady’ assessment of predictability for sea surface temperature, subsurface temperature and the Atlantic meridional overturning circulation in the North Atlantic in a coarse resolution (nominal resolution 2.75°-by-3° degree) model are all approximately 20. This minimal size reduces to about 7 in an eddy-permitting model (nominal resolution 1/4°-by-1/4°). This is probably caused by the difference in the potential uncertainty sources, as a consequence of the difference in spatial resolution, as well as the fact that the IPSL-CM5A-LR is a fully coupled climate model, while the eddy-permitting model is a forced ocean model. That is to say, for the case of North Atlantic Oceanic climate forecasting, increasing ensemble size could improve the prediction skill and credibility of predictability assessment; and minimal size are variable-independent and resolution dependent. We therefore do not recommend ensemble size far less than 20/7 and suggest that continuing increasing ensemble members if the size is already over 20/7 could be unnecessary when attempting to forecast North Atlantic Oceanic climate using non-eddy-permitting/eddy-permitting models.

More information

Identifiers

Catalogue record

Export record