The weak heat exchanges at the northern border of the southern oceans in CM5_piCtrl are consistent with the strong cold anomalies in the southern subpolar area shown in Fig. Small molecule library 8 (top left). Fig. 11 (lower panel) shows the major differences between CM5_piStart and CM5_RETRO both in terms of heat transport (arrows) and of atmospheric heat flux (colours). Transport (flux) differences that are not significant at the 95% level according to a Student test are not plotted (dotted). If the oceanic drift is small or at least similar in the two simulations, the total

budget of the atmospheric flux and divergence of oceanic transport should be comparable. Fig. 1 (top panel) shows that it is indeed the case for the upper 300 m, and it can also be verified for the whole water column (not shown). Thus, in Fig. 11 (and similarly in Fig. 12), changes in oceanic heat transport can be interpreted in terms of changes in atmospheric heat fluxes and conversely. Regarding the heat transport, major differences are found again in the southern basins. The zonal heat transport in the Southern Ocean is weaker (by 2–10%) in CM5_piStart than in CM5_RETRO. Differences are largest at the longitude of the Cape of Good Hope. At 30°S in the South Atlantic, both the very weak northward transport in CM5_piStart (0.02 PW) and the very weak southward one in CM5_RETRO (0.01 PW) are unrealistic (0.35 PW northward in Ganachaud and Wunsch, 2000 and Talley, 2003).

Nevertheless, the weaker transport at Cape of

check details Good Hope in CM5_piStart could be explained by a weak northern loss in the southern Atlantic as compared to CM5_RETRO. This effect is however not strong enough to explain the whole difference. Variations of ACC heat transport are also explained by its meanders, as shown by Sun and Watts (2002): the ACC warms when it meanders equatorward, namely in the South Atlantic and Indian Oceans, mainly thanks to the Brazil and Agulhas western boundary currents, and cools in its poleward segments, primarily in the South Pacific. This feature in well reproduced in both simulations. The largest zonal changes in water mass heat content in CM5 RETRO Florfenicol is not associated with a strong change in mass transport (Fig. 13 below) and it could thus be due to stronger temperature gradients in the Brazil-Falkland confluence in this simulation compared to CM5_piStart (not shown). The northward heat transport entering the South Pacific is also weaker in CM5_piStart than in CM5_RETRO. This is consistent with the stronger oceanic heat uptake from the atmosphere between 15°S and 30°S. Reduced ITF in CM5_piStart compared to CM5_RETRO is also consistent with reduced northward (intensified southward) heat flux into the Arabian Sea. Again, this implies an excess of heat in the Arabian Sea, which is taken from the atmosphere. In the North Atlantic, the northward heat transport at 30°N is unchanged in the two simulations. The slight intensification (0.