Evaluation of MARv3.14 over Greenland and the Impact of Model Resolution

Accurate simulation of the Greenland Ice Sheet (GrIS) surface mass balance (SMB) is essential for quantifying its contribution to sea-level rise. The regional atmospheric climate model MAR is widely used to study GrIS SMB changes and force ice-sheet models, highlighting the importance of assessing its performance and sensitivity to horizontal resolution. Here, we evaluate the latest MARv3.14 version at 5 km resolution and assess the impact of coarser resolutions (10–30 km) on simulated GrIS SMB and its components.

MAR outputs are evaluated against a range of independent observations, including in situ SMB measurements, automatic weather station records of near-surface meteorological variables, and satellite-derived melt extent and albedo products. At 5 km resolution, MAR reproduces observed SMB with a root-mean-square error of 0.53 m w.e. yr⁻¹. For near-surface meteorological variables and surface energy budget fluxes, model errors are smaller than the corresponding observed variability. The assimilation of bare-ice albedo, implemented in the latest MAR version, improves model performance in the ablation zone. Remaining biases in the accumulation zone suggest that improvements to the snow albedo scheme are further required. Simulated melt timing is consistent with satellite-based melt extent products.

Comparing simulations at different spatial resolutions, we find that SMB discrepancies mainly occur at the ice-sheet margins, characterized by strong topographic gradients. While integrated SMB differences generally remain within interannual variability, precipitation and runoff are highly sensitive to the model spatial resolution.

Corrections based on a vertical SMB gradient (as in Franco et al., 2012) or a sub-pixel methodology allowing the surface scheme (SISVAT) to be run at a higher resolution than the atmospheric model could deal in part with runoff discrepancies vs spatial resolution but precipitation anomalies remain a challenge. We conclude by discussing ongoing model developments, in particular the implementation of a sub-pixel methodology.