Statistically optimising the input parameter space of a numerical ice sheet model to improve the model fit to observations of palaeo-ice flow direction

Both the Greenland and Antarctic ice sheets are experiencing increased levels of melt, contributing to potentially devastating sea level rise. Quantifying their future changes is imperative in order to understand and mitigate the risks associated with their demise. Projections of future ice sheet change due to climate change are highly uncertain. Palaeo-ice sheets left behind a wealth of information on past ice extents, timing and flow directions. By looking to the past and using such data to validate and constrain numerical ice sheet model simulations, the formulation of model approaches can be improved, and the uncertainty within projections of ice mass loss and sea level rise can be reduced. 

Here we simulate the last Eurasian Ice Sheet complex (EISC) between 40 and 5 thousand years ago, to find a model input parameter space that is optimised to fit the available flow geometry as revealed by observations of former ice flow direction such as from drumlins. We present a new Bayesian framework that takes an initial perturbed parameter ensemble for the EISC, compares each ensemble member to past observed flow directions and identifies an updated parameter sampling routine on a reduced parameter space to improve the overall model-data match of further simulations. To quantitatively compare and score observed flow geometry from glacial landforms with model simulations in a statistically rigorous way, a new model-data comparison tool is utilised: the Likelihood of Accordant Lineations Analysis (LALA) tool. This work could not only be used further to develop a robust simulation of the EISC, as well as other palaeo-ice sheets, optimised to flow geometry, but also to simulate data-driven spin-ups for use in future ice sheet projections.