全球变化知识资源中心

Geologic archives constraining the variability of the Greenland ice sheet (GrIS) during the Holocene provide targets for ice sheet models to test sensitivities to variations in past climate and model formulation. Even as data-model comparisons are becoming more common, many models simulating the behavior of the GrIS during the past rely on meshes with coarse horizontal resolutions (>= 10 km). In this study, we explore the impact of model resolution on the simulated nature of retreat across southwestern Greenland during the Holocene. Four simulations are performed using the Ice Sheet System Model (ISSM): three that use a uniform mesh and horizontal mesh resolutions of 20, 10, and 5 km, and one that uses a nonuniform mesh with a resolution ranging from 2 to 15 km. We find that the simulated retreat can vary significantly between models with different horizontal resolutions based on how well the bed topography is resolved. In areas of low topographic relief, the horizontal resolution plays a negligible role in simulated differences in retreat, with each model instead responding similarly to retreat driven by surface mass balance (SMB). Conversely, in areas where the bed topography is complex and high in relief, such as fjords, the lower-resolution models (10 and 20 km) simulate unrealistic retreat that occurs as ice surface lowering intersects bumps in the bed topography that would otherwise be resolved as troughs using the higher-resolution grids. Our results highlight the important role that high-resolution grids play in simulating retreat in areas of complex bed topography, but also suggest that models using nonuniform grids can save computational resources through coarsening the mesh in areas of noncomplex bed topography where the SMB predominantly drives retreat. Additionally, these results emphasize that care must be taken with ice sheet models when tuning model parameters to match reconstructed margins, particularly for lower-resolution models in regions where complex bed topography is poorly resolved.