Ice sheets could collapse faster than current models predict, and a UK research team is using artificial intelligence to learn from a prehistoric meltdown 14,500 years ago—when sea levels rose 18 metres in just 340 years—to improve forecasts of future sea level rise. This matters because today’s ice sheet records are too short to capture rare but catastrophic instabilities. The most extreme scenarios project over 1 metre of sea level rise by 2100, which would submerge land now home to 100 million people and worsen storm surges, flooding, and saltwater contamination of coastal aquifers. Planners need reliable worst-case numbers to design long-lived infrastructure such as nuclear power plants, the Thames Barrier, and transport networks. The team has already built AI tools and statistical methods that simulate past ice sheets far more realistically than before. In the second phase, they will apply these techniques to the UK’s flagship Earth system model, UKESM, to investigate abrupt ice sheet changes during past rapid sea level rises. If successful, the work will directly reduce uncertainty in projections of Greenland ice sheet behaviour and future sea level change, giving policymakers and engineers concrete numbers to work with rather than vague warnings.
View original technical description
The most significant threat of future rapid sea level rise is the collapse of ice sheets due to instability and runaway ice loss. It could lead to more than 1 m of sea level rise by 2100, submerging land currently home to 100 million people and causing further destruction in higher-elevation coastal regions through enhanced storm and flood risk and aquifer salinification. Predicting the future possibility of such instabilities and the resulting plausible 'worst case' sea level change is critical for adequately planning coastal defences and long-term infrastructures for which a rare event could have devastating consequences (e.g. nuclear power plants, the Thames Barrier, transport networks). However, this is highly challenging because ice sheet instabilities have not occurred since we started measuring ice sheet changes (the record is too short, and ice sheets have been stable in the recent past), and they depend on poorly understood mechanisms (e.g. sliding of ice) that occur in inaccessible areas (e.g. under kilometres of ice). There is a solution: ice sheet instabilities have occurred in the geological past, for example, 14,500 years ago (the time of mammoths and modern humans) when the collapse of ice sheets around the world produced up to 18 m of sea level rise in 340 years (more than five times the rates expected for the end of the century). Geological records of past ice sheet evolution provide an untapped goldmine of data that can be used to test and improve numerical models, informing future projections. However, in order to reliably translate knowledge from the past into confident future projections, the largest and most complex source of uncertainty in modelling past ice sheets needs to be accounted for: the climate. Tackling this problem requires new statistical methods and a unique combination of expertise in statistics, climate and ice sheet instabilities. The first phase of this fellowship has developed Artificial Intelligence tools and statistical (Bayesian) Uncertainty Quantification techniques that have transformed our ability to simulate realistic past ice sheets using a fast yet complex coupled climate-ice sheet model (FAMOUS-ice). These advances include tools and techniques for sampling through uncertain multidimensional model inputs and correcting model biases. We have also created an ultra-fast emulator (i.e. a statistical regression model) of the surface mass balance in FAMOUS-ice that adjusts as the ice sheet advances or retreats. Thanks to this work, my team has demonstrated that simulating the coldest part of the last ice age (~20,000 years ago) is a powerful approach for ensuring the model is able to flexibly predict climate and ice sheet behaviour different to today, thus reducing uncertainty and improving confidence in future projections of the Greenland ice sheet. We have learned that the largest source of uncertainty is how we model the albedo (i.e. brightness) of snow and ice. The project's second phase will apply our artificial intelligence tools to improve simulations of ice sheet instabilities with the higher resolution and high complexity UKESM, the flagship UK Earth System Model that can simulate how ice sheets flow and interact with the climate. We will investigate the abrupt ice sheet changes that took place during past rapid sea level rises, and will use these to improve projections of future ice sheet and sea level changes.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know