Report on assessment of decadal predictability of THC.

In THOR, new multimodel decadal hindcasts of the North Atlantic have been performed with state-of-the-art modelling systems, following the protocol of the international CMIP5 project. Decadal hindcasts performed in an earlier EU program, indicate predictability up to 9 years in the subpolar Atlantic. Predictability originates from the external forcing by greenhouse gasses and aerosols and from the state of the climate as represented in the initial state of the decadal hindcasts. In particular, the future state of the subpolar gyre in the Atlantic may depend on the current state of the ocean. Control simulations with state-of-the-art climate models show that ocean dynamics play an important role in generating decadal climate variability. Together with the observed variability this indicates that there is scope for predictions up to decadal time scales in the North Atlantic region. The new THOR results show that the major patterns of variability, such as the Atlantic Multidecadal Oscillation, and integrated quantities in the North Atlantic, such as heat content, are predictable with lead times up to 9 years. Also, the interhemispheric contrast of sea surface temperature in the Atlantic is predictable. The predictability in the North Atlantic leads to enhanced predictability of tropical storms and possibly translates to predictability of climate parameters over Europe. It is likely that this predictability originates from ocean dynamics associated with the Atlantic Meridional Overturning Circulation (AMOC). However, the mechanisms controlling the variability of the AMOC and related atmospheric variables, and the impact of initialisation on forecast skill, require further investigation. Finally, new techniques of perturbing the ocean component have been developed in order to create ensembles that represent the uncertainties in models and observations. For all participating models singular vectors have been developed that maximize grow in the AMOC. Tests have been performed to study the impact on predictability. The spread of the predictions, which was too small, increases, but the skill is not strongly affected. The new multimodel decadal hindcasts will also be submitted to the CMIP5 database, which will be assessed in the IPCC AR5 assessment report.