The Antarctic ice sheet (AIS) is experiencing mass loss, with ice discharge exceeding mass gain from surface processes. Surface mass balance (SMB), the difference between snow accumulation and ablation, significantly influences AIS mass change variability. Surface melt on floating ice shelves leads to surface lowering, firn pore space depletion, and meltwater ponding, potentially triggering ice shelf collapse and accelerating ice discharge. Climate models, typically operating at resolutions of 25–100 km, struggle to accurately represent these complex surface processes due to their inability to resolve fine-scale topography. This underestimation particularly affects regions with steep slopes and high snowfall/melt gradients. This research aims to improve the accuracy of SMB and surface melt estimations by utilizing statistical downscaling techniques to achieve 2 km resolution, allowing for better representation of the AIS's complex topography and surface processes. The improved resolution is expected to provide more realistic projections of future AIS mass change and contribute to a more accurate understanding of its role in global sea-level rise.

Previous studies have highlighted the importance of surface processes in Antarctic ice sheet mass balance. Research using regional climate models (RCMs) and Earth System Models (ESMs) often underestimates SMB and surface melt, particularly in topographically complex regions. Multi-model comparisons reveal substantial inter-model differences in annual SMB estimates. Model evaluation using in-situ measurements and satellite data (e.g., QuikSCAT) indicates a general underestimation of melt, especially in RCMs like RACMO2.3p. The limitations of coarser resolution models in capturing orographic precipitation and melt gradients near grounding lines have been widely recognized. This necessitates higher-resolution analyses to better understand and project future changes in Antarctic ice sheet mass balance.

This study generated daily SMB and surface melt products at 2 km resolution for the grounded AIS and floating ice shelves. The methodology involved two steps: First, contemporary climate data from ERA5 (1979–2021) and three CESM2 global climate projections (SSP1-2.6, SSP2-4.5, SSP5-8.5; 1950–2099) were used as lateral forcing for RACMO2.3p, simulating SMB components at 27 km resolution. Second, statistical downscaling corrected these SMB components for elevation biases between the 27 km RACMO2.3p grid and a 2 km high-resolution topography from REMA. Melt and runoff were further adjusted for local albedo biases using a 2 km MODIS albedo map (2000–2021). The statistical downscaling technique corrects individual SMB components (precipitation, sublimation, snow drift erosion, melt, runoff) for elevation biases using daily-specific vertical gradients estimated from the 27 km RACMO2.3p grid. Melt and runoff were additionally corrected for albedo in areas with blue ice or exposed rock. The downscaling procedure is not mass-conservative, allowing for adjustments that improve the representation of SMB components. In situ measurements from the AntSMB dataset and satellite data from QuikSCAT were used for model evaluation. The method builds upon previous work successfully applying statistical downscaling to other glaciated regions.

Statistical downscaling at 2 km resolution significantly improved the representation of SMB and surface melt compared to the native 27 km RACMO2.3p data. The 2 km product showed an overall 3% increase in Antarctic-wide SMB (72 Gt year⁻¹), mainly due to enhanced accumulation in high-elevation mountain ranges and differences in ice mask extent. This modest increase in SMB, when combined with ice discharge data, reconciled modelled and GRACE/GRACE-FO mass change estimates, eliminating the need for regional corrections previously applied to the 27 km data. The 2 km product demonstrated better agreement with both in-situ and remote sensing data. Surface melt increased by 46% (51 Gt year⁻¹) in the 2 km product compared to the 27 km product, particularly pronounced near the grounding line. This increase is attributable to elevation corrections, changes in ice mask extent, and albedo adjustments. Future melt projections show a consistent melt underestimate in the 27 km product (45%) across all warming scenarios. The 2 km downscaled projections reveal significantly higher future surface melt rates than previously estimated, with a substantial increase anticipated along grounding lines of floating ice shelves.

The findings highlight the significant impact of model resolution on Antarctic ice sheet mass balance estimates. The modest increase in SMB from the 2 km downscaling was sufficient to reconcile modeled and observed mass changes, demonstrating the importance of accurately resolving topographic complexities. The substantial increase in surface melt rates near grounding lines, validated by in-situ and remote sensing data, underscores the vulnerability of floating ice shelves to future warming. The persistence of melt underestimation in lower-resolution models suggests that previous estimates of future melt may have been underestimated. The study provides a more refined picture of Antarctic mass balance and future melt projections, which is crucial for improving sea-level rise predictions and assessing the stability of Antarctic ice shelves.

This study demonstrates the crucial role of high-resolution modeling in accurately representing Antarctic ice sheet mass balance and future melt projections. Statistical downscaling to 2 km resolution significantly improved SMB and surface melt estimates, leading to better agreement with observations and a more accurate assessment of AIS mass change and its contribution to sea level rise. Future research should focus on incorporating ice dynamics into high-resolution models to account for potential feedback mechanisms and further refine future projections. The approach presented here is a valuable tool for improving the understanding and prediction of Antarctic ice sheet behavior under future climate change scenarios.

The study's limitations include uncertainties associated with the high-resolution surface topography data, ice sheet and ice shelf masks, and the satellite albedo product. The RACMO2.3p model and the utilized CESM2 scenarios do not incorporate ice dynamics, thus future projections assume fixed present-day ice sheet geometry. This simplification might underestimate the potential for melt-elevation feedback and other dynamic interactions affecting future melt patterns. While the albedo correction accounted for blue ice, future surface darkening due to other factors (e.g., biological growth or impurity deposition) is not included, potentially making the melt projections conservative estimates.