Using ensemble prediction methods to examine regional climate variation under global warming scenarios

The fate of the North Atlantic thermohaline circulation (THC) is of great significance for regional climate prediction. Research based on both numerical modelling and paleoclimate data has suggested that the THC might be intrinsically bistable, and could have the potential to switch rapidly between its stable modes. Using a low-resolution intermediate complexity model, we investigate the predictability of the response of the THC to anthropogenic forcing in the medium (100 years) and longer term. Using an ensemble Kalman filter we can efficiently tune the climate of ensemble members by varying multiple parameters simultaneously, and flux adjustments are not required to prevent unreasonable model drift. However, some biases remain, and we demonstrate that the common approach of subtracting the bias from a model forecast can result in substantial errors when the model state is close to a nonlinear threshold. Over 100 years of 1% per annum atmospheric CO2 enrichment, the THC drops significantly but steadily by about 4 or 5 Sv, a result that appears robust over a wide range of scenarios. In the longer term, the THC can collapse entirely, or recover to its original state, and small changes in the present uncertainties can have a large effect on the future outcomes. We conclude that generating reliable forecasts over the next century should be achievable, but the long term behaviour remains highly unpredictable. (c) 2005 Elsevier Ltd. All rights reserved.