Bias in modeled Greenland Ice Sheet melt revealed by ASCAT

Abstract

The runoff of surface melt is the primary driver of mass loss over the Greenland Ice Sheet. An accurate representation of surface melt is crucial for understanding the surface mass balance and, ultimately, the ice sheet's total contribution to sea level rise. Regional climate models (RCMs) model ice-sheet-wide melt volume but exhibit large variability in estimates among models, requiring validation with observed melt. Here, we explore a novel data processing method from the Advanced SCATterometer (ASCAT) instrument on board the EUMETSAT MetOp satellites, which provides estimates of the spatiotemporal variability of melt extent over the Greenland Ice Sheet between 2007 and 2020. We apply ASCAT wet-snow maps to pinpoint differences in the melt products from three distinct RCMs, where one is forced at the boundary with two different reanalyses. Using automatic weather station (AWS) air temperature observations, we assess how well RCM-modeled melt volume aligns with in situ temperatures. With this assessment, we establish a threshold for the RCMs to identify how much meltwater is in the models before it is observed at the AWSs and ultimately infer the melt extent in the RCMs. We show that applying thresholds, informed by in situ measurements, reduces the differences between ASCAT and RCMs and minimizes the discrepancies between different RCMs. Differences between modeled melt extent and melt extent observed by ASCAT are used to pinpoint (i) biases in the RCMs, which include variability in their albedo schemes, snowfall, turbulent heat fluxes, and temperature, as well as differences in radiation schemes, and (ii) limitations of the liquid water detection by ASCAT, including misclassification in the ablation zone and a temporal melt onset bias. Overall, we find that the RCMs tend to have a later melt onset than ASCAT and an earlier end to the melt season, with a similar but slightly smaller melt area compared to that identified in ASCAT. Biases, however, vary spatially between models and with compensating errors in different regions, suggesting that one RCM can sometimes represent the present-day surface across the entire ice sheet more effectively than the ensemble mean.