Sampling a North Atlantic Tipping Point: RareEvent Algorithms in UKESM on ARCHER2

Greenhouse gas emissions are warming the climate. A highly uncertain consequence is the crossing of climate tipping points, where parts of the Earth system undergo rapid, self-amplifying changes that are irreversible on societal 
timescales. Here we investigate tipping behavior in the North Atlantic subpolar gyre, a circulation feature south of Greenland that redistributes heat, freshwater, and nutrients. A sustained weakening would alter ocean stratification and deep-water formation, with impacts for marine ecosystems and European climate. 

A central bottleneck in quantifying tipping timescales, impacts and uncertainty is data scarcity: transitions are rare in observations and under-sampled in conventional climate-model ensembles. This limits data-hungry AI/ML approaches, while long high-resolution simulations needed to capture processes that trigger transitions are prohibitively expensive. We 
therefore employ rare-event algorithms, which oversample low-probability transitions by adaptively reweighting and branching ensemble members, generating statistically meaningful samples of rare outcomes at lower cost than direct 
simulation.

We are planning a programme of rare-event sampling experiments in UKESM, designed for execution on ARCHER2. Because the relevant transitions are rare, the approach requires many parallel multi-year coupled simulations with 
branching and extensive analysis, making it exceptionally demanding computationally. ARCHER2’s national-scale performance and ability to sustain large ensembles are central to the project, enabling us to move beyond under-sampled ensembles toward meaningful estimates of SPG transition likelihoods and pathways. We will describe the proposed simulation plan and report initial implementation and performance tests.