Predicting AMOC collapse probabilities using trajectory-adaptive multilevel sampling (TAMS)

The Atlantic Meridional Overturning Circulation (AMOC) plays a major role in the global and local climates by redistributing heat through the global ocean. Current climate projections indicate a consistent weakening of the AMOC strength which could be sign of a forthcoming collapse. There is strong evidence that such events occurred in the geological past. This work aims at evaluating the risk of AMOC transitioning to a collapsed state by 2100 using a global ocean model. To estimate the transition probability using such a complex model, the trajectory-adaptive multilevel splitting (TAMS) method is employed to greatly reduce the computational cost associated with classical Monte-Carlo approach when tackling rare events. We developed an implementation of the TAMS method amendable to expensive simulations, relying on the Omuse framework to interface the TAMS algorithm with the POP high fidelity ocean model. Early results indicate the viability of the proposed approach and ongoing effort aim at further improving the method by selecting a better score function, a key component of the TAMS method.