Here using ensemble climate change simulations with three climate models subject to different initial conditions, we find a large spread in projected 20-year globally averaged surface air temperature trends. This spread is found to be linked to Pacific climate variability whereby the trade winds, a key element of tropical Pacific climate, intensify in association with reduced globally averaged surface warming, and weaken with accelerated global warming. This implies diminished predictive skill for global surface air temperature trends over decadal time scales, due to a large extent intrinsic Pacific Ocean variability. We show, however, that this uncertainty can be considerably reduced when the initial oceanic state is known and well represented in the models. In this case, the spatial patterns of the surface temperature trends depend largely on the initial state of the Pacific Ocean.
Our results indicate that when the model integration starts from a negative phase of the Interdecadal Pacific Oscillation (IPO), the subsequent decade is more likely to exhibit accelerated global warming. Similarly, starting from a positive IPO leads to a more likelihood for reduced near-term global warming. This suggests that as we have recently been in a negative IPO phase, an accelerated global warming is likely in the coming decade.