Insights into the LGM-to-present evolution of the Greenland Ice Sheet from a data evaluated ensemble of numerical model simulations

Ice sheets have a memory that is stored within both the geometry and thermal properties of the ice. The current Greenland Ice Sheet is thus not in equilibrium with present-day climate, but is in fact affected by a complex product of past changes that occurred over millennial timescales. Therefore, simulating the late-Pleistocene evolution of the Greenland Ice Sheet accurately is important when running future projections using paleo model initialization procedures. Using a novel model-data comparison procedure, we ran an experiment that aimed to produce numerical model simulations that fit available empirical data on the extent and timing of the grounded margin evolution of the Greenland Ice Sheet from the global LGM (24 kyr BP) to 1850 AD. Given the numerous uncertain parameters and boundary conditions required by numerical ice sheet models, finding simulations which adequately replicate empirical data on past grounded ice extent is a challenging task. In an attempt to address this challenge, we ran a perturbed parameter ensemble of 100 ice-sheet-wide simulations at 5 km spatial resolution using the Parallel Ice Sheet Model. Our simulations are forced by the latest transient paleo-climate and ocean simulations of the isotope-enabled Community Earth System Model (iCESM 1.2 and 1.3). Using quantitative model-data comparison tools and the newly developed, Greenland-wide PaleoGrIS 1.0 isochrone reconstruction of former ice extent, each ensemble simulation’s fit with empirical data was assessed quantitatively across both space and time. Using our best-scoring simulations, we here present new insights into the former Greenland Ice Sheet’s likely response to transitional climatic phases throughout the last deglaciation. Secondly, our results suggest ice temperature, geometry, and glacial isostatic adjustment-induced mechanisms of centennial to millennial-scale inertia in ice-extent response to past climatic forcing, with potential implications for the future evolution of the ice sheet. Thirdly, our results show that different parameter combinations produce a better model-data fit during different time periods and for different regions of the ice sheet – i.e. parameter values that work well at one place or time, produce worse fit at others. We hypothesise that better paleo model initializations may be achieved using time- and space-dependent parameter configurations. Finally, after extending several past ensemble simulations to the end of the 21st century under CMIP6-derived forcing, we find that accounting for the past modifies projections of the future. Using a steady-state contemporary ice sheet as an initial state leads to vastly different projected sea level contributions when compared to simulations that perform well at recreating past glacial history.