the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Jun 2026
A novel database of Antarctic meteorological extremes over key ice shelves during 1995–2023
Abstract. The climate of Antarctica is showing increasing signs of being impacted by the warming trend in global temperatures, which has potential to result in accelerated break up of key ice shelves, which would contribute to global sea level rise. Here, we present a novel database of Antarctic extreme weather events over a selection of key ice shelves (Larsen, George VI, Wilkins, Abbot, Thwaites, Totten, Amery, Lazarev), using simulations from four regional climate models (RCMs: RACMO2, HCLIM, MetUM and MAR), driven by the ERA5 reanalysis, examining surface air temperature, precipitation, wind and surface pressure. In addition, we examine trends in the frequency of extreme events above or below specified thresholds (5th, 10th, 50th, 90th and 95th percentiles) and spatial atmospheric circulation and temperature anomaly patterns over Antarctica that are commonly associated with extreme events over key ice shelves. The RCM simulations have been compared with station observations close to the ice shelves, and we developed regressions to estimate simulated values during periods when only one or two of the RCMs were available.
Competing interests: At least one of the (co-)authors is a member of the editorial board of The Cryosphere. There are no other competing interests to declare.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(3858 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-2270', Anonymous Referee #1, 29 Jun 2026
-
AC1: 'Reply on RC1', Ian Simpson, 19 Aug 2026
- Lines 156–159: One of the strengths of this study is the use of high-resolution regional climate models. However, it has also been reported that ERA5 can reproduce these variables to some extent (Yamada et al., 2026). Therefore, I encourage the authors to include ERA5 in the analysis and demonstrate how much added value these regional models provide relative to the reanalysis.
Reference: Yamada, K., et al. (2026). Interannual variations of precipitation events at Dome Fuji station, Antarctica. Journal of Geophysical Research: Atmospheres, 131, e2025JD045296.
We will add ERA5 to the analysis for the George VI, Larsen and Amery ice shelves (covered in Table 2), and compare the extremes generated using ERA5 with the extremes generated using the four RCMs.
- Lines 207–212: The definition of extreme events includes both one-day exceedances of the 2nd, 5th, 95th, and 98th percentiles and three consecutive days exceeding the 10th or 90th percentiles. However, the physical rationale for using these specific thresholds is not fully explained. The authors should clarify why these thresholds were selected and whether the main conclusions are sensitive to this choice.
We will clarify that the "three consecutive days exceeding the 90th percentile" was chosen because it is a standard definition of a heatwave, and will cite a few papers that used this definition. We will add an explanation that the rationale for including single day events above p95 and p98 was to catch extreme events that feature one day spikes that may be lost when taking a 3-day average, while setting the thresholds higher for one day events.
- Lines 213–224: In this study, values for periods with missing model data are estimated using linear regression based on the available models. I assume that the regression coefficients were derived from the relationships among the models during the overlapping period, 2000–2020. If this understanding is correct, the relationships derived from 2000–2020 are applied to estimate values in the 2020s. However, because several extreme events occurred during the 2020s, these relationships may not necessarily hold for that period. At a minimum, the authors should compare the estimated values for the non-overlapping periods with available observations to evaluate the accuracy of this approach. In addition, the manuscript does not provide sufficient details on how the regression coefficients were calculated. Please explain the procedure more clearly.
We will add discussion of the a and b coefficients of the regressions, which is currently missing in the text, and we will compare estimated values for 2021-23 and 1995-99 with AWS observations for Larsen, George VI and Amery, plus Totten and Thwaites, which also have nearby AWS sites with reasonably long records.
- Lines 298–306: The authors appear to use the difference between the observed daily temperature and the observed 95th percentile temperature as the x-axis. However, it is unclear why this metric was chosen. I suggest that the authors explain the rationale for using this quantity in the Methods or Results section. If the objective is to evaluate model performance across different temperature conditions, would it not be more straightforward to use the observed daily temperature itself as the x-axis?
We chose this metric because we wanted to separate out events above the 95th percentile from events below, to determine if any biases in the RCMs were disproportionately affecting the high extremes. We will make this a lot clearer in the text that references the figure.
- Lines 307–311: Although the authors note that a similar warm bias has been reported in previous studies, the underlying cause of this warm bias is not discussed. It would be helpful for readers if the authors briefly addressed the possible reasons for this bias, or referred to previous studies that have investigated its origin.
We will elaborate more on the Gilbert et al. (2025) reference, which goes into some detail on the underlying causes of the warm bias. We also plan to address another reviewer's points by adding a version of the database that does not include HCLIM.
- Lines 380–383: Similar findings have also been discussed in the following study. I recommend citing this reference.
References: Sato, K., and I. Simmonds, 2021: Antarctic skin temperature warming related to enhanced downward longwave radiation associated with increased atmospheric advection of moisture and temperature. Environ. Res. Lett., 16, 064059, https://doi.org/10.1088/1748-9326/ac0211.
We will add and cite this reference.
- Lines 410–426, Tables 5–8: The relationships between ENSO, SAM, sea-ice extent, and the meteorological variables appear to vary considerably among seasons. Therefore, I suggest presenting seasonal correlation maps between ENSO, SAM, and sea-ice extent and each meteorological variable (temperature, precipitation, surface pressure, and wind speed) for four seasons. Such figures would provide a more comprehensive view of the seasonal dependence of these relationships than the current tables alone.
We will add the seasonal results to the annual results, possibly in supplementary if it results in the main paper having too many tables.
- Table 2: I suggest adding the model surface elevations at each observation site. Since the model elevations differ from the actual station elevations and elevation corrections were applied, including these values would help readers better assess the model performance.
We will do this, since the model elevations are readily available.
- Table 2: It would also be useful to include the results for the model mean, as this is one of the primary products used throughout the study.
We will add this in, and there is sufficient space for this (plus the ERA5 results, next point)
- Table 2: I also recommend including the corresponding ERA5 results. This would provide a useful benchmark and allow readers to evaluate the added value of the regional climate models over the driving reanalysis.
We will add this in, which corresponds to earlier point about using ERA5 reanalysis.
- Table 2: In addition to the correlation coefficients and RMSE, I suggest reporting the mean bias for each variable. While the temperature bias can be inferred from Figure 2, the biases in surface pressure and wind speed cannot be readily assessed from the current presentation. Including the mean bias would provide a more complete evaluation of model performance.
We will add the model biases in as well.
- Figure 2: The caption states that the Larsen Ice Shelf data cover the period 1985–2016. Does this refer to the observation period? What is the actual overlapping period used for the comparison between the observations and the model simulations? In addition, were the same time periods used for all four models at the Amery G3 and Fossil Bluff sites? For example, in the case of Fossil Bluff, RACMO2 extends to 2023, whereas the other models are only available until 2020. It is not clear whether all four models were compared over the same common period. If different periods were used, the comparison would not be directly comparable, and the analysis should instead be restricted to the common period.
Yes, the Larsen Ice Shelf's observation period is 1985-2016. The comparison between the four RCMs and observations is limited to within the common period of 2001-2020. We will make this clearer in the text.
- Figure 2: The caption refers to a "vertical line." Which line does this refer to? Is it the line at 0°C on the x-axis? If so, does every point to the right of this line represent a temperature exceeding the 95th percentile? Please clarify this in the figure or caption.
Yes, every point to the right of the line represents a temperature exceeding the 95th percentile. We will clarify this in the caption and the text.
- Figure 4: I suggest adding the corresponding observational results to this figure. Including observations would allow readers to directly evaluate the realism of the model simulations.
Yes, we will add observational results from the Larsen C AWS.
- Figures 5 and 6: It would be helpful to provide the numerical values of the seasonal trends for each ice-shelf region to figures. This would allow readers to quantitatively compare the magnitude of the trends among regions and seasons.
We will add the numerical values as requested.
Citation: https://doi.org/10.5194/egusphere-2026-2270-AC1
-
AC1: 'Reply on RC1', Ian Simpson, 19 Aug 2026
-
RC2: 'Comment on egusphere-2026-2270', Anonymous Referee #2, 13 Jul 2026
General comments
This study presents a novel dataset of extreme weather events over key Antarctic ice shelves based on a mini-ensemble of high resolution regional climate models. This combined PolarRES dataset has been previously used in an individual case study (Gilbert et al. 2025), but applying the PolarRES dataset to create a climatology of extreme events over select ice shelves represents an important and useful advancement. The authors present the statistics and trends from the PolarRES dataset in relation with large-scale modes of climate variability and sea-ice extent. The PolarRES dataset is then compared with previous case studies of previously documented Antarctic extreme events to demonstrate where those events fall within the distribution of the PolarRES climatology.
Unfortunately, I find the manuscript does not achieve in convincing the reader in the utility of the PolarRES dataset. The manuscript did not address why their methodology is an advancement in extreme event detection. The authors should demonstrate why their database in more useful than just using reanalysis data for a similar product. The high resolution of the models should answer this question, but this needs to be demonstrated in the manuscript.
Overall, I believe the PolarRES dataset will be a valuable tool for the Antarctic science community, which is why I feel the manuscript needs to be improved to better demonstrate its value. Thus my opinion this manuscript needs major revisions before it can be considered for publication. My specific and then minor comments are found below.
Specific comments
- Introduction: Overall, the introduction lacks focus and organization. Each paragraph jumps to a different topic related to Antarctic weather extremes and melt, but with no real connection to the previous paragraph or a narrative throughline. Some topics seem randomly placed without a clear reason. For instance, in lines 59-65, the atmospheric river is introduced and then uses the Brunt ice shelf iceberg calving example to demonstrate their impacts even though Francis et al. (2022) describes a combination of factors that caused the calving. Then a few paragraphs later, the authors include a case of calving on the Amery ice shelf which was caused by similar circumstances as the Brunt ice shelf example, but is now being discussed under the context of ice shelf calving. Also, teleconnections are discussed in lines 51-58 and then again in line 74-81. I suggest a rewrite of the Introduction to create a clear narrative that connects the existence of extreme weather to a lack of observation and high resolution modeling. Try to condense the introduction and remove unnecessary material so that the reader can quickly get to the point of the study. Like there are only 5 lines (121-125) that actually address the project motivation (i.e. the creation of an extreme event database). In this regard, the authors should discuss Antarctic extremes in the context of how a lack of high resolution modeling (Gilbert et al. 2025), observations, and connecting the extreme events inhibits our ability to holistically understand their impacts.
- Data: I understand the logic of not using ERA5 since its coarser resolution misses extreme weather activity in complex terrain, but comparing the PolarRES models with ERA5 would still be helpful to get a general understanding of their performance. Here the PolarRES models could be interpolated to the ERA5 resolution and then determine if there are any large systemic biases in temperature and wind. The other concern I have is the lack of AWS stations used for validation. Essentially, 5 out of the 8 ice shelves don’t have observations for comparison. If the authors use the 3-hourly AntAWS dataset, would they find more stations that could be used for validation? Since the data is being used to calculate daily means and not being analyzed at the hourly scale, is the downside from using 3-hourly data really that large? Having more observational and reanalysis data to compare against would give more confidence to the model extremes database.
- Methods: I understand the need for using the linear regression to fill gaps in the model records, but is there a way to test the potential errors this might introduce? What about preforming a data deprivation experiment where data from a model over a set of year is removed and replaced with the linear regression method? This could be done over a period where all four models have data so that the estimated mean value from the four models can be compared with the actual four model mean.
- Figure 2 and the corresponding analysis: If the x-axis is the difference between station daily temperature (please state daily mean if this is the case) and the 95th percentile of that station’s daily mean temperature, then shouldn’t the y-axis be the difference the model daily mean temperature and the 95th percentile of the corresponding station’s daily mean temperature? If the y-axis is comparing the model daily mean against the station daily mean, then the comparison does not make sense. Also, the description of the vertical line in the caption does not seem correct. Since the x-axis is just describing station data, 5% of the daily mean temperature from the station will cross the vertical line regardless of what the model simulated. Please check this figure and correct the analysis if changes are needed.
- Case studies: One great justification of this research is to find extreme events that have not been previously discussed in the literature. As currently written, the case studies section mostly compares the PolarRES dataset against known extreme events while occasionally pointing out extremes identified in the database that were previously unknown. Also the analysis of the case studies focuses almost purely on warm extremes and not any potential cold or snowy extremes. I suggest a rewrite of the case studies section that quickly highlights the database picking up the well known extreme events, perhaps discussing ways the database expands our understanding of those events, and then giving examples of previously unknown events and discussing some of their impacts that may relate to cold extremes and precipitation. For instance, the authors mention an extreme event in June 2006 that I do not believe has been previously discussed. Here it would be nice to see some images of the snowfall anomalies that may have resulted from this extreme. Also the authors can take a previously documented extreme event and show how the high resolution of the PolarRES models reveals a greater detail about the temperature extremes over complex terrain than previously understood.
- Discussion and Conclusion: I feel that the benefits of having an extreme event database are not fully discussed or realized in these sections. Here it would be interesting if the authors can state how their database can provide greater insight into already documented extremes or uncover previously undocumented extreme events. Finally, as the PolarRES dataset is a product that should benefit the Antarctic scientific community, the authors can do a better job “selling” their product in the Conclusion by mentioning examples of other disciplines that would be interested in this dataset like ecologists studying greening on the Antarctic Peninsula (Roland et al. 2024) or biologists studying microbe response to extreme weather (Barrett et al. 2024).
Minor comments
Line 21: Having two “which”s in the opening sentence is clunky. Would recommend phrasing the end of the sentence
Line 41: Please clarify here that the increased vulnerability to calving results from ocean waves adding stress to the ice shelf front when sea ice is absent.
Line 40-42: Recommend splitting this sentence into two; first focusing on the fohn wind that brings warm air over the ice shelf then discussing the sea ice disintegration and exposure to ocean swells when the strong wind reaches the ice shelf front.
Line 42: Massom et al. (2018) only discusses the ocean swell stress on ice shelves. You need citations for the fohn wind. For example, (Turton et al. 2020; Wille et al. 2022; Elvidge et al. 2020).
Line 44: Yes extreme precipitation events add to the surface mass balance and offset loses from melting, but does precipitation actually reduce surface meltwater ponding? van Wessem et al. (2023) only discusses melt pond formation and nothing about precipitation. You can look at (Maclennan et al. 2022; Turner et al. 2019; Wille et al. 2021; González-Herrero et al. 2023; Adusumilli et al. 2021) for papers about extreme precipitation.
Line 45-50: The authors should also mention that the mass balance has been slightly negative to neutral over the past five years or so due to an increase in extreme snowfall events. See (Kolbe et al. 2026; Wang et al. 2023, 2025)
Line 52: Add an oxford comma
Line 58: Specify that this is basal melt.
Line 59-60: Please add a citation for atmospheric rivers (Zhu and Newell 1998).
Line: 213-216: Please rephrase this sentence. Using the word “for” three times feels awkward.
Figure 3: Please clarify the months of the seasons in the caption.
Line 307-310: Check grammar. It feels like a run-on sentence.
Line 337: “over Amery”. Check grammar.
Table 3 and 4: Recommend including observational values where available. It would help the reader understand the model biases.
Figure 4: It would be helpful if there was a panel for the station observations as well. Then the simulated temperature spikes in the 2006 can be validated with observations.
Figure 5: Is the annual count just the sum of the four seasons? Or is it the number of days that exceed the annually defined 95th quantile? Please ensure it is the former and specify in the caption.
Line 372-373: Please clarify if this statement refers to all ice shelves? Also calculating the trend for annual extremes and listing it in the upper right hand corners would be interesting.
Line 416: Double periods.
Line 451: The storm that triggered the final collapse of the Larsen A was identified as an atmospheric river which is worth mentioning here (Scambos et al. 2000; Wille et al. 2022)
Line 504-508: This event was extensively documented in Bozkurt et al. (2018) which discussed an atmospheric river and resulting foehn wind over the Antarctic Peninsula.
References
Adusumilli, S., M. A. Fish, H. A. Fricker, and B. Medley, 2021: Atmospheric River Precipitation Contributed to Rapid Increases in Surface Height of the West Antarctic Ice Sheet in 2019. Geophysical Research Letters, 48, e2020GL091076, https://doi.org/10.1029/2020GL091076.
Barrett, J. E., and Coauthors, 2024: Response of a Terrestrial Polar Ecosystem to the March 2022 Antarctic Weather Anomaly. Earth’s Future, 12, e2023EF004306, https://doi.org/10.1029/2023EF004306.
Bozkurt, D., R. Rondanelli, J. C. Marín, and R. Garreaud, 2018: Foehn Event Triggered by an Atmospheric River Underlies Record-Setting Temperature Along Continental Antarctica. Journal of Geophysical Research: Atmospheres, 123, 3871–3892, https://doi.org/10.1002/2017JD027796.
Elvidge, A. D., P. Kuipers Munneke, J. C. King, I. A. Renfrew, and E. Gilbert, 2020: Atmospheric Drivers of Melt on Larsen C Ice Shelf: Surface Energy Budget Regimes and the Impact of Foehn. Journal of Geophysical Research: Atmospheres, 125, e2020JD032463, https://doi.org/10.1029/2020JD032463.
Francis, D., R. Fonseca, K. S. Mattingly, O. J. Marsh, S. Lhermitte, and C. Cherif, 2022: Atmospheric Triggers of the Brunt Ice Shelf Calving in February 2021. Journal of Geophysical Research: Atmospheres, 127, e2021JD036424, https://doi.org/10.1029/2021JD036424.
Gilbert, E., D. Pishniak, J. A. Torres, A. Orr, M. Maclennan, N. Wever, and K. Verro, 2025: Extreme precipitation associated with atmospheric rivers over West Antarctic ice shelves: insights from kilometre-scale regional climate modelling. The Cryosphere, 19, 597–618, https://doi.org/10.5194/tc-19-597-2025.
González-Herrero, S., F. Vasallo, J. Bech, I. Gorodetskaya, B. Elvira, and A. Justel, 2023: Extreme precipitation records in Antarctica. International Journal of Climatology, 43, 3125–3138, https://doi.org/10.1002/joc.8020.
Kolbe, M., J. A. Torres Alavez, R. Mottram, M. Katurji, R. Bintanja, and E. C. van der Linden, 2026: Atmospheric rivers and winter sea ice drive recent reversal in Antarctic ice mass loss. Communications Earth & Environment, 7, 255, https://doi.org/10.1038/s43247-026-03242-3.
Maclennan, M. L., J. T. M. Lenaerts, C. Shields, and J. D. Wille, 2022: Contribution of Atmospheric Rivers to Antarctic Precipitation. Geophysical Research Letters, 49, e2022GL100585, https://doi.org/10.1029/2022GL100585.
Massom, R. A., T. A. Scambos, L. G. Bennetts, P. Reid, V. A. Squire, and S. E. Stammerjohn, 2018: Antarctic ice shelf disintegration triggered by sea ice loss and ocean swell. Nature, 558, 383–389, https://doi.org/10.1038/s41586-018-0212-1.
Roland, T. P., and Coauthors, 2024: Sustained greening of the Antarctic Peninsula observed from satellites. Nature Geoscience, 17, 1121–1126, https://doi.org/10.1038/s41561-024-01564-5.
Scambos, T. A., C. Hulbe, M. Fahnestock, and J. Bohlander, 2000: The link between climate warming and break-up of ice shelves in the Antarctic Peninsula. J. Glaciol., 46, 516–530, https://doi.org/10.3189/172756500781833043.
Turner, J., and Coauthors, 2019: The Dominant Role of Extreme Precipitation Events in Antarctic Snowfall Variability. Geophysical Research Letters, https://doi.org/10.1029/2018GL081517.
Turton, J. V., A. Kirchgaessner, A. N. Ross, J. C. King, and P. Kuipers Munneke, 2020: The influence of föhn winds on annual and seasonal surface melt on the Larsen C Ice Shelf, Antarctica. The Cryosphere Discussions, 2020, 1–25, https://doi.org/10.5194/tc-2020-72.
Wang, W., Y. Shen, Q. Chen, F. Wang, and Y. Yu, 2025: Spatiotemporal mass change rate analysis from 2002 to 2023 over the Antarctic Ice Sheet and four glacier basins in Wilkes-Queen Mary Land. Science China Earth Sciences, 68, 1086–1099, https://doi.org/10.1007/s11430-024-1517-1.
Wang, Y., Q. Wu, X. Zhang, and Z. Zhai, 2023: Record-breaking Antarctic snowfall in 2022 delays global sea level rise. Science Bulletin, https://doi.org/10.1016/j.scib.2023.08.055.
van Wessem, J. M., M. R. van den Broeke, B. Wouters, and S. Lhermitte, 2023: Variable temperature thresholds of melt pond formation on Antarctic ice shelves. Nature Climate Change, 13, 161–166, https://doi.org/10.1038/s41558-022-01577-1.
Wille, J. D., and Coauthors, 2021: Antarctic Atmospheric River Climatology and Precipitation Impacts. Journal of Geophysical Research: Atmospheres, 126, e2020JD033788, https://doi.org/https://doi.org/10.1029/2020JD033788.
——, and Coauthors, 2022: Intense atmospheric rivers can weaken ice shelf stability at the Antarctic Peninsula. Communications Earth & Environment, 3, 90, https://doi.org/10.1038/s43247-022-00422-9.
Zhu, Y., and R. E. Newell, 1998: A Proposed Algorithm for Moisture Fluxes from Atmospheric Rivers. Monthly Weather Review, 126, 725–735, https://doi.org/10.1175/1520-0493(1998)126%3C0725:APAFMF%3E2.0.CO;2.
Citation: https://doi.org/10.5194/egusphere-2026-2270-RC2 -
AC2: 'Reply on RC2', Ian Simpson, 19 Aug 2026
Reviewer 2
General comments
This study presents a novel dataset of extreme weather events over key Antarctic ice shelves based on a mini-ensemble of high resolution regional climate models. This combined PolarRES dataset has been previously used in an individual case study (Gilbert et al. 2025), but applying the PolarRES dataset to create a climatology of extreme events over select ice shelves represents an important and useful advancement. The authors present the statistics and trends from the PolarRES dataset in relation with large-scale modes of climate variability and sea-ice extent. The PolarRES dataset is then compared with previous case studies of previously documented Antarctic extreme events to demonstrate where those events fall within the distribution of the PolarRES climatology.
Unfortunately, I find the manuscript does not achieve in convincing the reader in the utility of the PolarRES dataset. The manuscript did not address why their methodology is an advancement in extreme event detection. The authors should demonstrate why their database in more useful than just using reanalysis data for a similar product. The high resolution of the models should answer this question, but this needs to be demonstrated in the manuscript.
Overall, I believe the PolarRES dataset will be a valuable tool for the Antarctic science community, which is why I feel the manuscript needs to be improved to better demonstrate its value. Thus my opinion this manuscript needs major revisions before it can be considered for publication. My specific and then minor comments are found below.
Specific comments
1. Introduction: Overall, the introduction lacks focus and organization. Each paragraph jumps to a different topic related to Antarctic weather extremes and melt, but with no real connection to the previous paragraph or a narrative throughline. Some topics seem randomly placed without a clear reason. For instance, in lines 59-65, the atmospheric river is introduced and then uses the Brunt ice shelf iceberg calving example to demonstrate their impacts even though Francis et al. (2022) describes a combination of factors that caused the calving. Then a few paragraphs later, the authors include a case of calving on the Amery ice shelf which was caused by similar circumstances as the Brunt ice shelf example, but is now being discussed under the context of ice shelf calving. Also, teleconnections are discussed in lines 51-58 and then again in line 74-81. I suggest a rewrite of the Introduction to create a clear narrative that connects the existence of extreme weather to a lack of observation and high resolution modeling. Try to condense the introduction and remove unnecessary material so that the reader can quickly get to the point of the study. Like there are only 5 lines (121-125) that actually address the project motivation (i.e. the creation of an extreme event database). In this regard, the authors should discuss Antarctic extremes in the context of how a lack of high resolution modeling (Gilbert et al. 2025), observations, and connecting the extreme events inhibits our ability to holistically understand their impacts.
We will rewrite the introduction accordingly, and we agree that it could be made more coherent and more to the point.
2. Data: I understand the logic of not using ERA5 since its coarser resolution misses extreme weather activity in complex terrain, but comparing the PolarRES models with ERA5 would still be helpful to get a general understanding of their performance. Here the PolarRES models could be interpolated to the ERA5 resolution and then determine if there are any large systemic biases in temperature and wind. The other concern I have is the lack of AWS stations used for validation. Essentially, 5 out of the 8 ice shelves don’t have observations for comparison. If the authors use the 3-hourly AntAWS dataset, would they find more stations that could be used for validation? Since the data is being used to calculate daily means and not being analyzed at the hourly scale, is the downside from using 3-hourly data really that large? Having more observational and reanalysis data to compare against would give more confidence to the model extremes database.
Three of the ice shelves do not have long running AWS sites nearby, but there are nearby sites at Thwaites and Totten, which will allow us to increase to five of the eight ice shelves having observations for comparison. We plan to add ERA5 to the analysis for three of the ice shelves (George VI, Largen and Amery) as given in Table 2, and to add the ERA5 results to Table 2 as suggested by one of the other reviewers. However, we feel that interpolating the PolarRES models to the ERA5 resolution will be a large undertaking for relatively limited reward, as each of the PolarRES models has differences in gridding methods.
3. Methods: I understand the need for using the linear regression to fill gaps in the model records, but is there a way to test the potential errors this might introduce? What about preforming a data deprivation experiment where data from a model over a set of year is removed and replaced with the linear regression method? This could be done over a period where all four models have data so that the estimated mean value from the four models can be compared with the actual four model mean.
Yes, we will select a year within the common period and test the potential errors by removing individual models, applying the regression and comparing with the actual four model mean.
4. Figure 2 and the corresponding analysis: If the x-axis is the difference between station daily temperature (please state daily mean if this is the case) and the 95th percentile of that station’s daily mean temperature, then shouldn’t the y-axis be the difference the model daily mean temperature and the 95th percentile of the corresponding station’s daily mean temperature? If the y-axis is comparing the model daily mean against the station daily mean, then the comparison does not make sense. Also, the description of the vertical line in the caption does not seem correct. Since the x-axis is just describing station data, 5% of the daily mean temperature from the station will cross the vertical line regardless of what the model simulated. Please check this figure and correct the analysis if changes are needed.
We agree that the explanation of Figure 2 could be much better. The idea is to compare (model - station) against (station - 95th percentile) to determine whether the model biases change depending on how high up the percentile distribution the station observations were, and to pick out the model bias when observations exceeded the 95th percentile. We will make this much clearer in the figure and the text referencing it.
5. Case studies: One great justification of this research is to find extreme events that have not been previously discussed in the literature. As currently written, the case studies section mostly compares the PolarRES dataset against known extreme events while occasionally pointing out extremes identified in the database that were previously unknown. Also the analysis of the case studies focuses almost purely on warm extremes and not any potential cold or snowy extremes. I suggest a rewrite of the case studies section that quickly highlights the database picking up the well known extreme events, perhaps discussing ways the database expands our understanding of those events, and then giving examples of previously unknown events and discussing some of their impacts that may relate to cold extremes and precipitation. For instance, the authors mention an extreme event in June 2006 that I do not believe has been previously discussed. Here it would be nice to see some images of the snowfall anomalies that may have resulted from this extreme. Also the authors can take a previously documented extreme event and show how the high resolution of the PolarRES models reveals a greater detail about the temperature extremes over complex terrain than previously understood.
We feel we are justified in focusing primarily on the warm extremes but agree that some discussion of cold or snowy extremes is justified to demonstrate the all-round practical uses of the database. The June 2006 event does stand out, particularly as an exceptional warm spike in austral winter. Snowfall anomalies for June 2006 will be tricky but we can add images of precipitation anomalies, possibly in supplementary material. We will address the last point about resolution by bringing in the ERA5 reanalysis.
6. Discussion and Conclusion: I feel that the benefits of having an extreme event database are not fully discussed or realized in these sections. Here it would be interesting if the authors can state how their database can provide greater insight into already documented extremes or uncover previously undocumented extreme events. Finally, as the PolarRES dataset is a product that should benefit the Antarctic scientific community, the authors can do a better job “selling” their product in the Conclusion by mentioning examples of other disciplines that would be interested in this dataset like ecologists studying greening on the Antarctic Peninsula (Roland et al. 2024) or biologists studying microbe response to extreme weather (Barrett et al. 2024).
We will have a more thorough examination of the literature documenting extreme events plus highlighting extremes that have not been covered. We also agree that the Conclusion is currently too brief and will benefit from discussing other disciplines that could benefit from the dataset. Thank you for the citations which are very useful.
Minor comments
Line 21: Having two “which”s in the opening sentence is clunky. Would recommend phrasing the end of the sentence
Yes, good point: we will rephrase the last part of the sentence.
Line 41: Please clarify here that the increased vulnerability to calving results from ocean waves adding stress to the ice shelf front when sea ice is absent.
Yes we will address this point.
Line 40-42: Recommend splitting this sentence into two; first focusing on the fohn wind that brings warm air over the ice shelf then discussing the sea ice disintegration and exposure to ocean swells when the strong wind reaches the ice shelf front.
We will bear this point in mind. We may do this but it will depend on how our introduction runs after we rewrite the introduction to make it more coherent and give a clearer message as per major point 1.
Line 42: Massom et al. (2018) only discusses the ocean swell stress on ice shelves. You need citations for the fohn wind. For example, (Turton et al. 2020; Wille et al. 2022; Elvidge et al. 2020).
Indeed, thanks for the helpful citations.
Line 44: Yes extreme precipitation events add to the surface mass balance and offset loses from melting, but does precipitation actually reduce surface meltwater ponding? van Wessem et al. (2023) only discusses melt pond formation and nothing about precipitation. You can look at (Maclennan et al. 2022; Turner et al. 2019; Wille et al. 2021; González-Herrero et al. 2023; Adusumilli et al. 2021) for papers about extreme precipitation.
Good point, we will check out these papers and adapt our comments according to their findings.
Line 45-50: The authors should also mention that the mass balance has been slightly negative to neutral over the past five years or so due to an increase in extreme snowfall events. See (Kolbe et al. 2026; Wang et al. 2023, 2025)
Yes good point, this is an important omission from this section of the introduction.
Line 52: Add an oxford comma
This will be pending rewriting of the Introduction as per major point 1, but if the sentence remains as is we will add a comma.
Line 58: Specify that this is basal melt.
We will specify that it is basal melt.
Line 59-60: Please add a citation for atmospheric rivers (Zhu and Newell 1998).
Yes, this one slipped through; thanks for the reference provided.
Line: 213-216: Please rephrase this sentence. Using the word “for” three times feels awkward.
We intend to address this by replacing the third "for" in the last part of the sentence.
Figure 3: Please clarify the months of the seasons in the caption.
Yes good point, the figure "as is" could especially confuse readers who are used to boreal rather than austral seasons. We will add DJF, MAM etc. to the caption.
Line 307-310: Check grammar. It feels like a run-on sentence.
We intend to split the sentence with a "However," or along those lines.
Line 337: “over Amery”. Check grammar.
We will change this to "the Amery ice shelf".
Table 3 and 4: Recommend including observational values where available. It would help the reader understand the model biases.
One potential issue with this is that the distribution of extremes at AWS sites may not be representative of the ice shelf as a whole, because of localised effects such as fohn winds. We will compare observational values with the RCM values and if this proves not to be a substantial issue we will include them where available. If not, we will explain the decision not to include them at the next stage of the revisions.
Figure 4: It would be helpful if there was a panel for the station observations as well. Then the simulated temperature spikes in the 2006 can be validated with observations.
We agree with including station observations in the panel, especially as there are long running station observations at Larsen C.
Figure 5: Is the annual count just the sum of the four seasons? Or is it the number of days that exceed the annually defined 95th quantile? Please ensure it is the former and specify in the caption.
Yes, it is just the sum of the four seasons: we will specify this in the caption.
Line 372-373: Please clarify if this statement refers to all ice shelves? Also calculating the trend for annual extremes and listing it in the upper right hand corners would be interesting.
We will change this to "the Amery ice shelf".
Line 416: Double periods.
We will remove the second stray period just before "Correlations".
Line 451: The storm that triggered the final collapse of the Larsen A was identified as an atmospheric river which is worth mentioning here (Scambos et al. 2000; Wille et al. 2022)
We will note this and examine whether this atmospheric river corresponds to the storm that shows up in the extremes database on 24/25 January 1995.
Citation: https://doi.org/10.5194/egusphere-2026-2270-AC2 -
AC3: 'Reply on RC2', Ian Simpson, 19 Aug 2026
Line 504-508: This event was extensively documented in Bozkurt et al. (2018) which discussed an atmospheric river and resulting foehn wind over the Antarctic Peninsula.
Thanks for this, we were not aware of that analysis and will add in a reference to it. There is also a grammatical error (two spaces after "up to 8°C") that we will address.
Citation: https://doi.org/10.5194/egusphere-2026-2270-AC3
-
RC3: 'Comment on egusphere-2026-2270', Anonymous Referee #3, 17 Jul 2026
This study constructs a meteorological extreme event database covering eight major Antarctic ice shelves spanning 1995-2023 using four regional climate models, focusing on four core variables: surface air temperature, precipitation, surface pressure and wind speed. The authors validate model outputs against in-situ station observations and diagnose the driving mechanisms of extreme events via key climate modes including the SAM, ENSO and regional sea ice variability. Multiple well-documented ice shelf collapse and extreme heatwave cases are selected for targeted case analysis, rendering the topic highly relevant to the cutting-edge research on Antarctic ice shelf climate change.
Overall, the manuscript presents an interesting effort to build a database of Antarctic meteorological extremes over key ice shelves using a multi-model ensemble of Regional Climate Models (RCMs). While the objective is highly relevant to understanding ice shelf stability, there are fundamental methodological flaws regarding the statistical definitions and data gap-filling. These issues must be addressed to ensure the scientific validity and readability of the study. The manuscript reads more like a database documentation with preliminary descriptive statistics, while rigorous assessments of methodological robustness, inter-model consistency, and validation of detected extreme events remain insufficient. I recommend Major Revisions based on the following critical points.
Major Comments
- The manuscript defines extreme events on its own (e.g., specific percentile thresholds within a 1-day or 3-day window) (L207-L212). However, most existing studies on extreme events have adopted definitions established by the Expert Team on Climate Change Detection and Indices (ETCCDI). Please explain why this study did not adopt or adjust the existing extreme index definitions, particularly regarding the selection of the base period for percentiles.
- The methodology described for estimating missing model data during the periods 1995–2000 and 2020–2023 is inappropriate for an analysis of climate extremes (L213-L224). The authors state that they developed linear regressions ($y = ax + b$) to estimate the four-model mean using the available subset of RCMs to retain consistency with the 2001–2019 daily percentiles. However, linear regression is fundamentally designed to capture the central tendency and inherently minimizes variance. Extreme weather events, such as the 95th or 98th percentiles, exist at the absolute tails of the distribution and are characterized by high variance and non-linearity.
By applying linear regression to gap-fill daily extreme values, the methodology artificially suppresses this critical variance. Consequently, true extreme peaks will be mathematically flattened, inevitably leading to false negatives where severe events fail to breach the established percentile thresholds. Furthermore, extrapolating extreme values using a simple linear fit can produce physically unrealistic results.
- The manuscript documents a prominent warm summer bias in HCLIM; at the Amery Ice Shelf in East Antarctica, the 98th percentile temperature simulated by HCLIM exceeds outputs from other models by more than 8 °C. The multi-model ensemble mean is simply calculated via unweighted arithmetic averaging without any weighting scheme to offset systematic model biases, which raises critical concerns regarding the reliability of derived extreme event statistics.
- Daily climatological percentiles are computed based on the 2001-2019 reference period using the four-model ensemble mean. The reference period of only 19 years is relatively short, which may introduce substantial biases to percentile thresholds and subsequent extreme identification. In addition, the full study period (1995-2023) spans merely 29 years, an insufficient time length for robust long-term trend detection of extreme events. The discussion of trend results must be more cautious and well-qualified.
- The abstract is overly general and fails to highlight core quantitative findings. A full rewrite of the abstract is required to incorporate key specific results from the analysis.
- At least two to three well-documented historical Antarctic extreme events should be comprehensively examined to demonstrate that the proposed database can reliably capture observed extreme meteorological episodes.
- The results section overemphasizes descriptive documentation of the database rather than highlighting novel scientific insights. The authors shall restructure this section to prioritize interpretative scientific discoveries.
Minor Comments
- The introduction should more explicitly articulate the novelty and research gaps addressed by this work.
- Subplots in Figure 2 adopt inconsistent axis limits, which hinders cross-panel comparison. The zero-reference lines (x=0, y=0) should be uniformly marked across all subplots for consistency.
- Tick labels on the y-axis of Figure 3 are illegible. The vertical axis interval should be widened, and font sizes enlarged for readability.
- Line traces in Figure 4 are indistinct, and axis labels use overly small fonts; graphical revisions are necessary to improve clarity.
- The line charts used to represent the annual and seasonal counts of extreme events (Figures 5 and 6) are not an effective visualization choice. Figures 5, 6, 8, 10 and 11 contain identical legends across subplots. A single shared legend per figure is sufficient, and all axis text fonts should be enlarged.
- Latitude and longitude annotations in Figures 7 and 9 are too small to distinguish and require resizing.
- Statistical p-values should be italicized (e.g., p < 0.01) following standard journal formatting conventions.
- The current correlation tables only present tabular values without visual spatial patterns of correlation coefficients. Supplementary spatial maps illustrating the geographical distribution of these correlations are recommended.
- Manuscript formatting errors require correction: Line 113 contains a typo “Southern Annual Mode”, which should be revised to “Southern Annular Mode”.
- A whitespace should be inserted between numerical values and their corresponding units throughout the text.
- Standardize unit formatting uniformly across the full text, including mm day⁻¹ and m s⁻¹. (or mm/day and m/s)
- Reference formatting is inconsistent; all citations should be unified following the journal’s official style guide.
Citation: https://doi.org/10.5194/egusphere-2026-2270-RC3 -
AC4: 'Reply on RC3', Ian Simpson, 19 Aug 2026
Reviewer 3
This study constructs a meteorological extreme event database covering eight major Antarctic ice shelves spanning 1995-2023 using four regional climate models, focusing on four core variables: surface air temperature, precipitation, surface pressure and wind speed. The authors validate model outputs against in-situ station observations and diagnose the driving mechanisms of extreme events via key climate modes including the SAM, ENSO and regional sea ice variability. Multiple well-documented ice shelf collapse and extreme heatwave cases are selected for targeted case analysis, rendering the topic highly relevant to the cutting-edge research on Antarctic ice shelf climate change.
Overall, the manuscript presents an interesting effort to build a database of Antarctic meteorological extremes over key ice shelves using a multi-model ensemble of Regional Climate Models (RCMs). While the objective is highly relevant to understanding ice shelf stability, there are fundamental methodological flaws regarding the statistical definitions and data gap-filling. These issues must be addressed to ensure the scientific validity and readability of the study. The manuscript reads more like a database documentation with preliminary descriptive statistics, while rigorous assessments of methodological robustness, inter-model consistency, and validation of detected extreme events remain insufficient. I recommend Major Revisions based on the following critical points.
Major Comments
1. The manuscript defines extreme events on its own (e.g., specific percentile thresholds within a 1-day or 3-day window) (L207-L212). However, most existing studies on extreme events have adopted definitions established by the Expert Team on Climate Change Detection and Indices (ETCCDI). Please explain why this study did not adopt or adjust the existing extreme index definitions, particularly regarding the selection of the base period for percentiles.
We did not choose the ETCCDI indices because we feel that most of them are less suitable for Antarctica than for mid-latitude regions such as Europe or North America, and will not adequately identify short-lived but critical extreme events over the Antarctic ice shelves. We could not use 1961-1990 or similar as a base period, mainly because we have a limited span of years for which RCM runs were available, only one of which has runs extending earlier than 1995 (and then only to 1981 for RACMO2).
The methodology described for estimating missing model data during the periods 1995–2000 and 2020–2023 is inappropriate for an analysis of climate extremes (L213-L224). The authors state that they developed linear regressions ($y = ax + b$) to estimate the four-model mean using the available subset of RCMs to retain consistency with the 2001–2019 daily percentiles. However, linear regression is fundamentally designed to capture the central tendency and inherently minimizes variance. Extreme weather events, such as the 95th or 98th percentiles, exist at the absolute tails of the distribution and are characterized by high variance and non-linearity.
By applying linear regression to gap-fill daily extreme values, the methodology artificially suppresses this critical variance. Consequently, true extreme peaks will be mathematically flattened, inevitably leading to false negatives where severe events fail to breach the established percentile thresholds. Furthermore, extrapolating extreme values using a simple linear fit can produce physically unrealistic results.
We acknowledge these deficiencies in using linear regression, but more complex forms of regression and other forms of infilling will introduce problems of their own, and so changing this facet of the methodology is likely to be a considerable additional undertaking for little extra reward. We will test the reliability of the regressions by selecting a year within the common period and test the potential errors by removing individual models, applying the regression and comparing with the actual four model mean. We hope that this will be sufficient to address these concerns about the regressions.
1. The manuscript documents a prominent warm summer bias in HCLIM; at the Amery Ice Shelf in East Antarctica, the 98th percentile temperature simulated by HCLIM exceeds outputs from other models by more than 8 °C. The multi-model ensemble mean is simply calculated via unweighted arithmetic averaging without any weighting scheme to offset systematic model biases, which raises critical concerns regarding the reliability of derived extreme event statistics.
Again we acknowledge this problem but we feel that adding weighting to the scheme will add a lot of extra complexity for limited reward, e.g. the HCLIM warm bias is different for different ice shelves and different times of the year, so it will not be straightforward to provide a weighted multi-model mean. We will produce another version of the database that does not include HCLIM as well as one that does, so that the version without HCLIM can be used in future projects that for example assess periods above and below freezing, for which the HCLIM warm bias will be a problem.
2. Daily climatological percentiles are computed based on the 2001-2019 reference period using the four-model ensemble mean. The reference period of only 19 years is relatively short, which may introduce substantial biases to percentile thresholds and subsequent extreme identification. In addition, the full study period (1995-2023) spans merely 29 years, an insufficient time length for robust long-term trend detection of extreme events. The discussion of trend results must be more cautious and well-qualified.
We will address this by emphasising the short study period when discussing Tables 5 to 8, which look at the relationships between extreme events and the SAM, ENSO and sea ice extent, for which the reliability of the trends will indeed be limited by the small sample size, and potentially influenced by a few individual events. For the trends in mean temperature and extreme events, however, we have already highlighted that significance is limited by the small sample size and that more data are needed.
3. The abstract is overly general and fails to highlight core quantitative findings. A full rewrite of the abstract is required to incorporate key specific results from the analysis.
We think it may not require a full re-write but we agree that the abstract "as is" certainly lacks key specific results and will need expanding and some re-writing to include these results.
4. At least two to three well-documented historical Antarctic extreme events should be comprehensively examined to demonstrate that the proposed database can reliably capture observed extreme meteorological episodes.
To address this point we will discuss two to three extreme events more comprehensively (provisionally including the 2006 and 2019/20 events), and, addressing points from other reviewers, add the ERA5 reanalysis to the study for three of the ice shelves (Larsen, George VI and Amery) so as to highlight the benefits of the greater resolution of the RCMs.
5. The results section overemphasizes descriptive documentation of the database rather than highlighting novel scientific insights. The authors shall restructure this section to prioritize interpretative scientific discoveries.
We will change the emphasis of this section to place greater emphasis on the scientific implications of our statistical analysis of the database.
Minor Comments
1. The introduction should more explicitly articulate the novelty and research gaps addressed by this work.
As suggested by other reviewers, we will restructure the introduction so that concepts follow on from each other more coherently and the novelty and research gaps are more evident to the reader.
2. Subplots in Figure 2 adopt inconsistent axis limits, which hinders cross-panel comparison. The zero-reference lines (x=0, y=0) should be uniformly marked across all subplots for consistency.
I don't see the axis limits problem - they all look the same on each of the plots. The non-uniform marking of the x and y-axis = 0 lines is a consequence of how Microsoft Word removes detail when resizing the images in the document, we will resize them in other image editing software so that they display correctly in Word.
3. Tick labels on the y-axis of Figure 3 are illegible. The vertical axis interval should be widened, and font sizes enlarged for readability.
Agreed; we will substitute these with intervals of 4 or 5 rather than 2, with larger font sizes.
4. Line traces in Figure 4 are indistinct, and axis labels use overly small fonts; graphical revisions are necessary to improve clarity.
Again this is likely a Word auto-removing detail when resizing issue so we will need to resize the figures in other image editing software before putting them into Word. We agree that the axis labels, particularly the x-axis labels, are too small and will be enlarged in the next revision.
5. The line charts used to represent the annual and seasonal counts of extreme events (Figures 5 and 6) are not an effective visualization choice. Figures 5, 6, 8, 10 and 11 contain identical legends across subplots. A single shared legend per figure is sufficient, and all axis text fonts should be enlarged.
Yes we will enlarge the text axis fonts and have one legend covering each of the plots as is standard.
6. Latitude and longitude annotations in Figures 7 and 9 are too small to distinguish and require resizing.
Yes we will enlarge them, they do appear very small.
7. Statistical p-values should be italicized (e.g., p < 0.01) following standard journal formatting conventions.
Yes we will do this.
8. The current correlation tables only present tabular values without visual spatial patterns of correlation coefficients. Supplementary spatial maps illustrating the geographical distribution of these correlations are recommended.
We feel that this is unnecessary and will add more figures that are incidental rather than central. A map of the geographical locations of the ice shelves is already provided in Figure 1.
9. Manuscript formatting errors require correction: Line 113 contains a typo “Southern Annual Mode”, which should be revised to “Southern Annular Mode”.
Yes, well spotted, we will revise this.
10. A whitespace should be inserted between numerical values and their corresponding units throughout the text.
Yes, we see this is the standard formatting in Cryosphere papers, and so we will amend this.
11. Standardize unit formatting uniformly across the full text, including mm day⁻¹ and m s⁻¹. (or mm/day and m/s)
We will check for inconsistencies and amend them.
12. Reference formatting is inconsistent; all citations should be unified following the journal’s official style guide.
We will again check for inconsistencies and amend them.
Citation: https://doi.org/10.5194/egusphere-2026-2270-AC4
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 187 | 59 | 10 | 256 | 10 | 11 |
- HTML: 187
- PDF: 59
- XML: 10
- Total: 256
- BibTeX: 10
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In this study, the reproducibility of basic meteorological parameters simulated by four regional climate models is evaluated through comparison with AWS observations over three ice shelves. In addition, using the outputs from these four models, the authors discuss trends in temperature and precipitation over eight ice shelves, as well as their relationships with large-scale atmospheric circulation modes such as ENSO and SAM. However, I have several concerns about this study, and I believe that substantial revisions are required. Therefore, I recommend that the manuscript be reconsidered for publication after major revision. My main reasons are as follows.
- Lines 156–159: One of the strengths of this study is the use of high-resolution regional climate models. However, it has also been reported that ERA5 can reproduce these variables to some extent (Yamada et al., 2026). Therefore, I encourage the authors to include ERA5 in the analysis and demonstrate how much added value these regional models provide relative to the reanalysis.
Reference: Yamada, K., et al. (2026). Interannual variations of precipitation events at Dome Fuji station, Antarctica. Journal of Geophysical Research: Atmospheres, 131, e2025JD045296.
- Lines 207–212: The definition of extreme events includes both one-day exceedances of the 2nd, 5th, 95th, and 98th percentiles and three consecutive days exceeding the 10th or 90th percentiles. However, the physical rationale for using these specific thresholds is not fully explained. The authors should clarify why these thresholds were selected and whether the main conclusions are sensitive to this choice.
- Lines 213–224: In this study, values for periods with missing model data are estimated using linear regression based on the available models. I assume that the regression coefficients were derived from the relationships among the models during the overlapping period, 2000–2020. If this understanding is correct, the relationships derived from 2000–2020 are applied to estimate values in the 2020s. However, because several extreme events occurred during the 2020s, these relationships may not necessarily hold for that period. At a minimum, the authors should compare the estimated values for the non-overlapping periods with available observations to evaluate the accuracy of this approach. In addition, the manuscript does not provide sufficient details on how the regression coefficients were calculated. Please explain the procedure more clearly.
- Lines 298–306: The authors appear to use the difference between the observed daily temperature and the observed 95th percentile temperature as the x-axis. However, it is unclear why this metric was chosen. I suggest that the authors explain the rationale for using this quantity in the Methods or Results section. If the objective is to evaluate model performance across different temperature conditions, would it not be more straightforward to use the observed daily temperature itself as the x-axis?
- Lines 307–311: Although the authors note that a similar warm bias has been reported in previous studies, the underlying cause of this warm bias is not discussed. It would be helpful for readers if the authors briefly addressed the possible reasons for this bias, or referred to previous studies that have investigated its origin.
- Lines 380–383: Similar findings have also been discussed in the following study. I recommend citing this reference.
References: Sato, K., and I. Simmonds, 2021: Antarctic skin temperature warming related to enhanced downward longwave radiation associated with increased atmospheric advection of moisture and temperature. Environ. Res. Lett., 16, 064059, https://doi.org/10.1088/1748-9326/ac0211.
- Lines 410–426, Tables 5–8: The relationships between ENSO, SAM, sea-ice extent, and the meteorological variables appear to vary considerably among seasons. Therefore, I suggest presenting seasonal correlation maps between ENSO, SAM, and sea-ice extent and each meteorological variable (temperature, precipitation, surface pressure, and wind speed) for four seasons. Such figures would provide a more comprehensive view of the seasonal dependence of these relationships than the current tables alone.
- Table 2: I suggest adding the model surface elevations at each observation site. Since the model elevations differ from the actual station elevations and elevation corrections were applied, including these values would help readers better assess the model performance.
- Table 2: It would also be useful to include the results for the model mean, as this is one of the primary products used throughout the study.
- Table 2: I also recommend including the corresponding ERA5 results. This would provide a useful benchmark and allow readers to evaluate the added value of the regional climate models over the driving reanalysis.
- Table 2: In addition to the correlation coefficients and RMSE, I suggest reporting the mean bias for each variable. While the temperature bias can be inferred from Figure 2, the biases in surface pressure and wind speed cannot be readily assessed from the current presentation. Including the mean bias would provide a more complete evaluation of model performance.
- Figure 2: The caption states that the Larsen Ice Shelf data cover the period 1985–2016. Does this refer to the observation period? What is the actual overlapping period used for the comparison between the observations and the model simulations? In addition, were the same time periods used for all four models at the Amery G3 and Fossil Bluff sites? For example, in the case of Fossil Bluff, RACMO2 extends to 2023, whereas the other models are only available until 2020. It is not clear whether all four models were compared over the same common period. If different periods were used, the comparison would not be directly comparable, and the analysis should instead be restricted to the common period.
- Figure 2: The caption refers to a "vertical line." Which line does this refer to? Is it the line at 0°C on the x-axis? If so, does every point to the right of this line represent a temperature exceeding the 95th percentile? Please clarify this in the figure or caption.
- Figure 4: I suggest adding the corresponding observational results to this figure. Including observations would allow readers to directly evaluate the realism of the model simulations.
- Figures 5 and 6: It would be helpful to provide the numerical values of the seasonal trends for each ice-shelf region to figures. This would allow readers to quantitatively compare the magnitude of the trends among regions and seasons.