| 26 Feb 2026
Evaluation of climatic predictors of surface ponding on Antarctic ice shelves
Abstract. Ponding of surface meltwater on Antarctic Peninsula ice shelves has led to hydrofracture-driven calving and grounding line retreat, and other regions could become vulnerable to increased ponding as the climate continues to warm. Theory and qualitative observations suggest that ponding initiates when the meltwater-over-accumulation ratio (MOA) reaches 0.7. Here, we use present-day satellite-derived Antarctic meltwater products and RACMO climate model outputs to calibrate predictive thresholds of surface ponding based on air temperature, MOA, and a weighted combination of MOA ana grounding line proximity index (GLPI).
We tested three RACMO resolutions (27 km, 11 km, and 2 km) and three surface meltwater products. The meltwater product that is best aligned with MOA identifies ponding locations using aggregate meltwater depths at 27 km resolution. For this product, the calibrated MOA and GLPI threshold predicts present-day ponding with an F1 score over twice as high as the theoretical threshold of MOA ≥ 0.7 (F1 = 0.587 vs. 0.261). Under emissions pathway SSP1-2.6, the empirical threshold predicts 2.3 times more lake coverage by 2100 than the theoretical threshold, underscoring the importance of calibrating climatic ponding thresholds. If MOA-based thresholds are to be used in future ponding projections, we recommend they be applied at relatively coarse spatial scales, calibrated against present-day, depth-based meltwater products, and combined with grounding line proximal processes.
Summary
Glazer and Tinto evaluate climate predictors for surface ponding on Antarctic ice shelves, testing the theoretical meltwater-over-accumulation (MOA) threshold of 0.7 against a calibrated threshold based on observational surface melt products, air temperature and RACMO across three spatial resolutions (2, 11 and 27 km). They further apply the best empirical MOA value and a grounding line proximity index against the theoretical one to project surface melt ponding until 2100 under three climate scenarios, demonstrating a substantial underestimation of surface ponding when relying solely on the theoretical threshold. Depending on the emission scenario, the theoretical threshold fails to capture a significant amount ponded pixels relative to the empirical threshold. This underscores the necessity of calibrating the MOA threshold to avoid systematic and potentially drastic underestimation of surface ponding on Antarctic ice shelves in future projections. The manuscript is well written and the findings are of considerable scientific significance. The following comments are intended to assist the authors in clarifying certain methodological choices and strengthening the overall presentation.
General Comments
Title: The title is currently quite broad. Consider refining it to explicitly reference MOA, grounding line proximity and air temperature thresholds, so that the scope and contribution of the study are immediately apparent to the reader.
Temporal aggregation of datasets and used variables: It is not entirely clear over which time spans the averages were calculated. For example, Morris & Vaughan report the −9°C isotherm based on the mean annual air temperature for the year 2000. It would be helpful to specify whether air temperature was calculated for each lake dataset over its respective observation period or over the full RACMO time span. The selection of air temperature over skin temperature would also benefit from explicit justification, and it may be worth considering whether a parallel analysis using skin temperature could be conducted to assess sensitivity to this choice.
Taking into account strong intra-annual variability of surface ponding on Antarctic ice shelves: As understood, lake ponding was aggregated based on the number of observed lakes on a pixel basis for the Dell and Tuckett datasets. However, e.g. RACMO covers the more recent year 2023, during which a pronounced peak in ponded surface meltwater occurred along the Antarctic Peninsula that is not captured by the observational products. Given that these datasets span different time periods, it would be important to clarify how this temporal inconsistency was accounted for, as it may substantially influence the results and how the strong intra- and inter-annual variability of surface ponding was accounted for.
Specific Comments
L22: It should be noted that not all parts of an ice shelf exert buttressing forces. See Fürst et al. 2016.
L28: Consider acknowledging the role of blue ice areas and rock outcrops in meltwater formation, given their characteristically lower albedo relative to surrounding snow and ice surfaces.
L37: The causal interpretation for Larsen B warrants more careful treatment. Leeson et al. 2020 raise the question of whether lake drainage constitutes a cause or a consequence of ice shelf collapse, which should be acknowledged here.
Figure 1: Please indicate the time period over which the average air temperature was calculated, and specify in the legend that 2 m air temperature is shown. It would also be helpful to clarify why the MOA color bar ends at 0.5 while the theoretical threshold is 0.7. Consider enlarging the maps for Dronning Maud Land, as the current size makes it difficult to resolve the gridded datasets.
L112: Please clarify which year's ice shelf shapefiles were used for clipping. E.g. whether they predate or postdate the collapsed ice shelves when clipping RACMO. Depending on the observed time span some very important pixels might be missing.
L118: A sensitivity analysis systematically testing a range of thresholds and their respective effects on the results would help quantify the uncertainty associated with what is acknowledged to be a somewhat subjective choice.
Figure 4: The maps for Dronning Maud Land should be enlarged, as spatial differences in the gridded datasets are currently very difficult to see.
Figure 5: Please indicate the time span over which the average air temperature shown was calculated.
Section 3.1. Average Air Temperature
L194: The characterisation "vastly underpredicts" may require revision in light of Table 2, which shows that the F1 score for the Tuckett-2k dataset is second highest at F1-score of 0.454, corresponding to an optimal temperature threshold of −9.2°C.
L197: The finding that no single temperature threshold constitutes an effective predictor merits further discussion. It would be worthwhile to examine whether skin temperature, maximum temperatures, or mean temperatures over the summer months might offer greater predictive utility.
Discussion: A notable gap in the discussion is the absence of a section addressing the need for region-specific thresholds. A more physically motivated distinction between the Antarctic Peninsula and the East Antarctic Ice Sheet would likely yield more robust and actionable recommendations than the current AP versus non-AP split.
Section 4.1: It should be noted in the discussion that the Dell et al. 2024 dataset includes areas of slush in addition to open water. Could this be a reason why in Figure 5 the Dell dataset is slightly different from the other datasets? Furthermore, the implications of the differing temporal coverages of the respective datasets for the comparability of results should be addressed.
L375: I am not a modelling expert but I wonder if you can actually do a proper comparison over different spatial scales when not all gridded datasets are based on the same RACMO version.
L414: The discussion would benefit from incorporating recent findings on the role of foehn winds in Dronning Maud Land from Mahagaonkar et al. 2025, rather than restricting this aspect of the discussion to the Antarctic Peninsula.
Conclusion: The manuscript presents scientifically valuable findings that merit a more confident presentation in the conclusion. In particular, the key result from the future projections section, namely that substantially greater ponding is projected under the empirical threshold relative to the theoretical one, is currently underrepresented and should be stated more prominently. I would suggest to provide explicit guidance on region-specific thresholds for the Antarctic Peninsula and EAIS/non-AP regions, and to state the optimum threshold values directly so that readers can readily reference them.
L585: The data aggregation procedure should be described in greater methodological detail. Downscaling from 30 m to 27 km resolution can introduce substantial errors depending on the approach taken, and a transparent account of the methodology is essential for reproducibility.
References mentioned:
Fürst, J. J., Durand, G., Gillet-Chaulet, F., Tavard, L., Rankl, M., Braun, M., and Gagliardini, O.: The safety band of Antarctic ice shelves, Nature Climate Change, 6, 479–482, 2016.
Leeson, A. A., Forster, E., Rice, A., Gourmelen, N., and Van Wessem, J. M.: Evolution of Supraglacial Lakes on the Larsen B Ice Shelf in the Decades Before it Collapsed, Geophysical Research Letters, 47, e2019GL085591, https://doi.org/10.1029/2019GL085591, 2020.
Mahagaonkar, A., Moholdt, G., Glaude, Q., and Schuler, T. V.: Katabatic and foehn winds control the distribution of supraglacial lakes in Dronning Maud Land, Antarctica, Earth and Planetary Science Letters, 666, 119482, https://doi.org/10.1016/j.epsl.2025.119482, 2025.
Morris, E. M. and Vaughan, D. G.: Spatial and Temporal Variation of Surface Temperature on the Antarctic Peninsula And The Limit of Viability of Ice Shelves, in: Antarctic Peninsula Climate Variability: Historical and Paleoenvironmental Perspectives, American Geophysical Union (AGU), 61–68, https://doi.org/10.1029/AR079p0061, 2003.