The Role of Internal Variability and External Forcing in Shaping AMOC Tipping

The Atlantic Meridional Overturning Circulation (AMOC) is a critical component of the climate system, responsible for the large-scale redistribution of heat and freshwater, and potentially vulnerable to an abrupt transition to a collapsed state under anthropogenic warming. Its future stability arises from a complex interplay between external forcings, such as increasing greenhouse gas concentrations, and the system’s internal climate variability. While most modelling studies focus on identifying deterministic thresholds for AMOC collapse, the role of internal variability in shaping the timing, probability, and nature of such transitions remains insufficiently understood. This thesis addresses this gap by analysing AMOC dynamics under a range of forcing scenarios using the intermediate-complexity climate model PlaSim–LSG, combined with hysteresis experiments, ensemble Monte Carlo simulations, and rare-event algorithms. In Chapter 3, we demonstrate that internal variability can substantially modify the timing of tipping points and reshape the AMOC hysteresis curve, occasionally reversing the expected pattern of the hysteresis cycle. This challenges the concept of a unique deterministic bifurcation threshold: once intrinsic climate fluctuations are accounted for, no unique critical CO2 concentration exists. Simulations that differ only in their initial atmospheric state by small perturbations may evolve toward markedly different AMOC states. Furthermore, commonly proposed statistical early-warning indicators are shown to have limited predictive skills, as internal variability generates false positives and obscures the trends predicted by classical Critical Slowing Down theory. To investigate the physical mechanisms promoting AMOC collapse and to provide a probabilistic assessment of its risk under different emission scenarios, it is necessary to sample rare collapsing trajectories. Since standard Monte Carlo approaches are computationally prohibitive, two rare-event methods are employed. Using the Giardina–Kurchan–Tailleur– Lecomte (GKTL) algorithm in Chapter 4, we successfully sample spontaneous AMOC transitions at fixed 354 ppm CO2 over 125-year simulations. The associated climate responses— including anomalies in zonal atmospheric circulation and strong surface cooling in the North Atlantic—are in line with results from higher-complexity models forced through AMOC shutdown using external perturbations. In our simulations, the spontaneous transitions appear to be triggered by anomalous westerly wind stress, which activates a weakening of the surface meridional transport through Ekman dynamics. This is subsequently followed by a collapse of deep convection in the Labrador Sea. In Chapter 5, the Trajectory-Adaptive Multilevel Splitting (TAMS) method is applied to estimate AMOC collapse probabilities under constant and transient forcing scenarios. TAMS is shown to be the most efficient and accurate estimator, indicating that an AMOC collapse is very unlikely during the 21st century but becomes a plausible outcome in the 22nd century under sustained high-emission scenarios. While model-specific biases remain, these results highlight the central role of internal variability in AMOC evolution and underscore the necessity of probabilistic frameworks for assessing tipping risks and defining a safe operating space for the climate system