the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Aug 2025
EC-Earth- and ERA5-driven ensemble hindcasts with the fully coupled ice-sheet–ocean–sea ice–atmosphere–land circum-Antarctic model PARASO
Abstract. The origins of recent and ongoing Antarctic climate trends are topic of debate, partly because trends and variability can originate from both the Antarctic climate system itself as well as from the mid-latitudes. Furthermore, we lack observations for a detailed analysis of these effects. Here, we use the regional ice sheet-ocean-sea ice-atmosphere-land circum-Antarctic model PARASO to produce four hindcasts of the Antarctic climate over the 1985–2014 period. The first is a control simulation, forced by atmospheric and oceanic reanalyses (ERA5 and ORAS5), which realistically reproduces the pre-2017 increase in Antarctic Sea Ice Extent (SIE) and Surface Mass Balance (SMB) of the Antarctic Ice Sheet. In contrast, the other three hindcasts, driven by EC-Earth historical simulations, simulate a declining SIE and increasing SMB over the same period, a behaviour consistent with biases seen in many global climate models, suggesting that biases in these models may be due to misrepresented lower-latitude dynamics or poleward transports. While both ERA5- and EC-Earth-driven simulations reproduce a dipole in sea ice concentration trends—positive in the east and negative in the west—the magnitude differs. The larger negative trend in the West in the EC-Earth-driven simulations feature a stronger intensification and displaced Amundsen Sea Low, enhancing northerly winds, moisture and heat flux between the Ross and Amundsen Sea. In turn, the different trends in SIE between the ERA5 driven and EC-Earth driven hindcasts result in opposing trends for moisture transport towards Antarctica and precipitation. By comparing the agreement between the three EC-Earth driven hindcasts, a small imprint of internal climate variability was found over the Southern Ocean, whereas this imprint over the continent is much stronger. Nonetheless, all EC-Earth driven simulations exhibit a robust positive SMB trend, indicating a link with sea ice decline or with large-scale advection shared across ensemble members.
- Preprint
(8301 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-2889', Anonymous Referee #1, 25 Oct 2025
-
AC2: 'Reply on RC1', Florian Sauerland, 07 Apr 2026
The paper of Florian and co-authors concerns a very important problem – to what degree can we trust to the GCMs when we apply them to simulate the climate round the Antarctica. The clear answer which follows from their study is “no, we cannot”. And this is the main conclusion of the research.
In other words, the authors aim to focus on pit-falls of the global climate models which purpose is to predict the possible future of the Antarctic climate. This is a crusial challenge – to separate imprints of the anthropogenic climate forcing from the climate natural variability. If the modern climate models fail to reproduce Antarctic climate in the nearest past, can we rely on their predictions? The problem is aggravated by the limit amount of the historical observations on the region.
It is also true that the polar regions are significantly different in many aspects from the other regions of the globe. This is the reason for the obvious mismatches between observations (or the reanalysis data) and modeling results.
These mismatches are clearly demonstrated by collating results of the numerical simulations in which the model PARASO is forced with the reanalysis and with the GCM simulations (sea ice extent, 2m air temperature etc.). Since the GCMs demonstrate not only the “wrong” figures but even opposite trends in climatic variables for the historic periods, one become rather sceptical about fasibility of their application to predict the climate of the Antarctica.
The manusript is well structured and clearly written (except some particular details mentioned below). The abstract provide a clear summary of the paper. Authors provide appropriate supplementary materials which illustrate the text. In overall, the study can make a valuable contribution to our understanding and modeling the climate dynamics in the polar regions. In this view the paper clearly addresses very important scientific questions within the scope of ESD.
Reply — Thank you for your detailed comments! We greatly appreciate the time and effort you put into reviewing our manuscript. Please find our replies to your comments below.
Major remarks
Comment 1 — Three numerical experiments cannot by any means called an “ensenble”. Hense the title of the paper must be corrected
Reply — Thanks for your remark. This has indeed also been a topic of discussion within the group of co-authors - as such, we have no issues in removing the word "ensemble" from the title and throughout the manuscript.
Comment 2 — There is no explicit and clear description of the numerical experiments FC00, FC01 and FC02
Reply — The main difference between these is their initialization: The EC-Earth simulations that are used to drive the PARASO experiments differ in their initial conditions in that they use different ensemble members of ORAS5. Also in response to reviewer 2’s remarks, we will now present these EC-Earth simulations in more detail and make sure to include a clearer description in section 2.3.
Comment 3 — What do authors mean under “internal climate variability” which is in the context of the study differs from “interannual variability”
Reply — Thank you for this important remark.
By internal variability we mean processes within the climate system that generate variability on all timescales (instantaneous to multidecadal). This in contrast to external forcing which can be anthropogenic (emission of greenhouse gasses) or natural (eruptions from vulcanos). Interannual variability is the extent to which any signal varies from one year to another, can be measured statistically (e.g. by the variance of the signal) and is independent on how it is generated; however, to a large degree, interannual climate variability is explained by internally-generated climate variability.
The line that the reviewer is referring to in indeed confusing as we say : “we can better differentiate internal variability from interannual variability.”
We therefore will change the associated paragraph to: For models at the resolution used here, often single model realizations of 30 years are analyzed, which is rather short given the slow ocean variability and high interannual variability of the atmosphere. In a single realization, internal climate variability can substantially affect 30-year trends. With our approach where we have three hindcasts driven by EC-Earth, we can get a better insight into the response due to external forcing (greenhouse gas emissions and so on), as the ensemble spread in indicative of the influence of internal variability.
Moreover, we will check the remaining of the manuscript on the correct use of both natural climate variability and interannual variability and will include this clarification.
Comment 4 — Terms “decadal variability” and “decadal trends” are not explained
Reply — decadal variability is only used once to describe the trends mentioned above: we cannot be fully sure that these are actual trends, or caused by some variability of ocean conditions which would evolve on very slow (decadal) timescales that would eventually even out again. As it is used only once in the manuscript, we will reformulate the sentence avoiding the term. The trends are called decadal because they are calculated over a multi-decadal period. We can see the confusion and therefore we now replaced all decadal trends with "30-year trends".
Comment 5 — The conclusion that the authors’ approach can help to “differentiate internal variability from interannual variability” is too optimistic (see comments below)
Reply — Indeed the use of the wording "internal variability" and "interannual variability" will be improved (see Comment 3)
Previous studies hypothesized that the misrepresentation of Antarctic sea ice (and other climate variables describing the climate of the Southern Ocean) might be due to issues in the representation of polar climate processes within GCMs. In our study we show that the modeled climate in the coastal oceans of Antarctica is similar for all three GCM-driven PARASO runs (and different from the ERA5-driven one). Thus we show that the cause for polar misrepresentations in GCMs lies predominantly in misrepresentations of climatic influences from lower latitudes.
While the similar trends between all three EC-Earth-driven hindcasts could also be due to coincidence, the similarity between these three runs also in their interannual variability further leads us to believe that the predominant reason for the difference between ERA5-driven and EC-Earth-driven runs is indeed in the influence of lower latitudes. The influence of internal variability within the PARASO domain thus seems much less relevant in comparison. However, what we have succeeded in is indeed not a differentiation between the two definitions of variability. Rather, we have ruled them both out as potential causes for the misrepresentation of trends. Accordingly, we will adapt our conclusions to represent this more accurately.
Other comments (Replies are in italic)
Line 15: “internal climate variability”. What is it? According to the WMO definition climate is measured by 30-yr period
Thank you for this remark. We follow here the definition of the IPCC where internal variability is opposed to external forcings and can happen on all time scales (virtually instantaneous to centuries). We will remove the word ’climate’ to be more concise.
Ref: https://archive.ipcc.ch/publications_and_data/ar4/wg1/en/ch9s9-1.htmlLine 63: daily-weekly time step to couple an ice sheet to other components of a system. Such small steps seem to be excessively small. A comment here would be pertinent.
This seems to be a misunderstanding, the time steps mentioned here are the model time steps of the different submodels. We will rephrase this to be more precise.
Line 67: three experiments can hardly be called an “ensemble”. This is just only three numerical experiments with sligtly different perturbed boundary conditions of the same model. Of course, no statistics can be ruled out of the results, which is the essence of the ensemble approach in climate modeling.
In line also with your major comment 1, we will adapt the naming here and throughout the manuscript.
Lines 112-113: I see a contradiction here. In case surface changes of the Antarctic Ice Sheet (AIS) are negligible during a 30-yr numerical experiment (and it is in the real world except may be the processes on the boundary ice sheet-ice shelf), why the model is called “fully coupled” in the context of the AIS? I suppose that the actual geometry of the AIS is used, isn’t it?
Indeed, surface changes to the antarctic ice sheet are not fed back into the atmospheric model, and in that sense, our model is not fully coupled. As described in Pelletier et al., 2022, however, the ice sheet model is used to prescribe ice-shelf cavities to the ocean model and is thus coupled for the aspects that could play a role on shorter timescales. . This is part of the offline coupling mentioned in line 108 and will be be made clearer in the description.
Ref.: Pelletier, C., Fichefet, T., Goosse, H., Haubner, K., Helsen, S., Huot, P.-V., Kittel, C., Klein, F., Le clec'h, S., van Lipzig, N. P. M., Marchi, S., Massonnet, F., Mathiot, P., Moravveji, E., Moreno-Chamarro, E., Ortega, P., Pattyn, F., Souverijns, N., Van Achter, G., Vanden Broucke, S., Vanhulle, A., Verfaillie, D., and Zipf, L.: PARASO, a circum-Antarctic fully coupled ice-sheet–ocean–sea-ice–atmosphere–land model involving f.ETISh1.7, NEMO3.6, LIM3.6, COSMO5.0 and CLM4.5, Geosci. Model Dev., 15, 553–594, https://doi.org/10.5194/gmd-15-553-2022, 2022.
Line 181: 20-yr or 20-kyr? 20 years seems a rather short period of relaxation, no?
This is indeed a mistake - it is actually 20 kyr. Thank you for noticing.
Lines 210 and 211: Correct sea name is “BellingShausen”
We will correct this here and throughout the manuscript.
Figure 1: More illustrative would be to demonstrate also absolute and relative differences in ice extent. I think the conclusions would be less optimistic: good coincidence only in November-December period.
(See our reply to your next comment.)
Lines 213-221: Discussion on the sea-ice extent simulations. The same note as to Fig. 1. I think, a short discussion will be plausible on how the mismatches in sea ice extent affect the overall simulation.
Thank you for this important remark: The relative impact of the underestimation is much larger in austral summer than in winter. We will make sure to discuss this in the manuscript.
Lines 242-245: “Since all EC-Earth driven experiments share quasi-identical initial states, all of those simulations are affected by similar forcing signals (due to greenhouse gases etc.) and model drift, while the timing of the internal variability may vary. Differences in long-term trends should therefore be the result of internal climate variability, generated outside and/or inside the domain.” Not necessarily. These trends can aslo indicate the sensitivity of the model to the initial conditions or to different setup of FC00, FC01 and FC02. Isn’t it?
We acknowledge that the formulation can be improved. Since the setup of FC00, FC01, and FC02 is identical except for slight perturbations in the initial conditions of the driving EC-Earth simulations as well as PARASO, any differences between the EC-Earth driven PARASO runs should be due to internal variability (inside or outside the domain) only. The sensitivity of the model to the small variations in initial conditions can be regarded as internal variability within the climatology of the model. We will improve this part in the manuscript.
Section 4.1 (also line 309): What is “pluri-decadal” in the context of this paper? Ten days or ten years? In the second case statistics has no sence. In the text, trends over 30 years (30 septembers and 30 marches) are described (i.e. in Table 1). Then the term “decadal” is really confusing.
Thank you for this comment. By ’pluri-decadal’ we mean trends over 30 years. We will reformulate this accordingly throughout the manuscript.
Line 267: “ensemble averaged”: does it mean an avearge over three values?
This is the trend calculated from the averaged values of the SIE, so three values per year. We will clarify this by removing ’ensemble’.
Line 267-268: The follwing is not clear: “The EC-Earth hindcasts themselves are not showing a negative SIE trend (Mann-Kendall test for the ensemble averaged September SIE: p = 0.972”. In the figure 4 one can clearly see a negative trend in EC-Earth hindcasts – from 17.5 mln. sq. km to 15 mln. sq. km over 30 years (meaning over 30 septermbers). Shown in Table 1 treds are statistically significant …
Thank you for this comment. The trends shown in Table 1 and Figure 4 (in blues) are trends seen in the PARASO simulations. The trends mentioned in Lines 267-268 are trends seen in the EC-Earth global simulations that were eventually used to drive PARASO. We will add the trends in the EC-Earth global simulations to Fig. 4 (as well as the seasonal cycle in these simulations in Fig. 1).
Line 288: “bufferzone”: may be “buffer zone”?
This is indeed better and we will change this in the manuscript.
Figures 7 and 8: Shown are anomalies, not absolute values, aren’t they? Then the colour scales and figure captions must be changed to avoid confusion.
Thank you, we will modify the figures and captions accordingly.
Figure 8: The color scale is indicated as 700 mb but in the figure caption is written 850 mb. What is true?
We noticed that the label in Fig. 8 is indeed incorrect, we show the 850hPa level as indicated in the caption.
Table 3: It is not clear over which decade averaging is carried out. The indicated period is 1985-2014. Are the trends estimated over annually averaged values? Isn’t the period of ten years (ten values) is long enough to figure out a realistic trend? By what means the trends were estimated?
Thank you for this remark. The given trends are the trends calculated for the whole 30-year simulation period, here and in all other tables containing trends. The term "decadal" referred here to the unit, as it is the rate of change in the given variables, based on a linear trend, per decade (i.e. K per decade, as in Table 2). We will remove the term to make this clearer and just term them trends, as the unit is already given in the table headers.
Figure 9: Evaporation and precipitation TRENDS
We will modify this in the colourbar labels.
Line 357: “After all, precipitation that does not have a local source has to have an origin somewhere else” – This is not entirely correct if the internal part of the Antarctca is considered. Here the precipitation has another (local) mechanism and is formed at the upper boundary of the inversion layer. Anyway, the related part of the text must be clarified in case anything else was supposed.
We acknowledge that this sentence does not add any value and will remove it altogether.
Table 4. The same note as for the Table 3. What decade is taken? Or the average among 3 decades is indicated? What do the authors mean under “decadal trend”?
We will remove the word ’decadal’, see our reply to your previous comment on Table 3.
Line 425: “the slow ocean variability”: the ice sheet is much less variable compared to the ocean, may be except the gronding line migration.
Thank you for this remark, we will mention variability in the ice sheet as well.
Lines 427-428: “With our approach where we have three hindcasts driven by EC-Earth, we can better differentiate internal variability from interannual variability”. The arguments in support of this statement seem not to be very strong. What was clearly demonstarted is that the GCM output driven model PARASO failed to reproduce some crusial regional climate features compared to the case when PARASO was forced by the reanalysis. By itself, this would be a very important conclusion. It is also not clear what “internal” variability means.
Our formulation here was indeed not very clear and we will improve this in the revised manuscript. If the trends between the three EC-Earth driven runs would have been very different, this would have pointed to strong influence of internal variability (primarily at timescales from several years to decades, e.g. ocean conditioning) as the three EC-Earth global runs only differ in their ocean initial state. Different evolution from year to year (interannual variability), but with similar longer-term trends (30 years) point towards a dominant influence of the external climate forcing, such as greenhouse gases or in our case also the processes external to the domain (e.g. in the tropics and subtropics).
Citation: https://doi.org/10.5194/egusphere-2025-2889-AC2
-
AC2: 'Reply on RC1', Florian Sauerland, 07 Apr 2026
-
RC2: 'Comment on egusphere-2025-2889', Christoph Kittel, 03 Mar 2026
Sauerland et al. present reconstructions of recent Antarctic climate using an almost fully coupled modelling framework. This is an interesting and valuable study, particularly given the well-known difficulty of running a fully coupled model without significant drift. From my understanding, the main message of the manuscript is that the atmospheric driving largely explains the simulated sea-ice trends, including the contrast with the opposite trends typically produced by GCMs. In this context, the improved physics of PARASO does not appear to substantially improve the representation of sea-ice trends compared to the driving GCM itself. It would therefore be helpful to show the direct results of the GCM (P21L460) in order to better assess the added value of the coupled framework.
Most of the simulated sea-ice trends and associated changes are attributed to atmospheric processes. However, the manuscript does not discuss potential changes at the oceanic boundaries. Overall, the manuscript is comprehensive, but it would benefit from a simplification of the text to better highlight the main storyline. This is particularly noticeable in Section 4.1 on sea ice. For example, parts of the caption of Figure 3 repeat information that is already obvious (e.g., stating that ice shelves are land). The manuscript would also gain clarity from a more structured development of the results and clearer links between the different analyses so that the central narrative emerges more distinctly.
I understand the rationale behind the debiasing of EC-Earth; however, the correction is applied at levels very close to the surface. This raises the question of whether the correction is dynamically consistent across the full atmospheric column. For instance, moisture advection typically occurs at levels well above the surface (e.g., around 850 hPa). If the correction is not vertically consistent, could it alter the dynamical forcing from EC-Earth or affect the structure of atmospheric circulation patterns? In that respect, could ERA5 be considered more reliable than EC-Earth even in the stratosphere, given that data assimilation is still performed at those altitudes and may constrain large-scale atmospheric properties more consistently?
Finally, although the authors provide a detailed evaluation of their configurations, the manuscript does not sufficiently discuss the implications of the identified biases for the simulations themselves. I would suggest including a comparison with in situ station data for atmospheric variables particularly wind rather than relying primarily on comparisons with ERA5 over the ice sheet, especially for the reanalysis-driven simulations. While this is not the most critical issue, it would strengthen the robustness of the evaluation.
See also the remaining minor comments.
P2 L29 , however. Correct the . location.
P3 L66 : The authors could try to rewrite the sentence to avoid the repition of « boundary and initial conditions ».
P3L80 : The authors can remove « in the boundary conditions »P4 L112: Is this only related to your spectific set up or to a real absence of ice surface change over 30yr ? Is the model fully coupled then ? For me it’s not the case as there is no coupling of the ice sheet towards the atmosphere. (also for P20L430: I’d better use nearly fully-coupled model)
P5L114: You can probably remove the two forcings occurences here. I’d also suggest to remove « regarding the forcing » P5 L121. I think there are enough (too much) “forcing” occurences in that paragraph so it’s clear you are talking about that.
P5L124: Does this change of forcing frequency have an impact on the ocean model results ?P6L172 : Consider to replace « snow mantle » by snowpack or snowpack properties (and then list wich properties you are initializing)
P6L189 : in instead of In.
P10 L239-251: I’d suggest the author to move these paragraphes at the end of the method section as they also explain where you did the evaluation, and especially the Figure 3 should be placed before the results of the evaluation and the becoming Fig 1.
P12 L282 – P13 L284: Isn't that a bit too simplistic to justify the decrease in sea ice in this region?
Figure 6 could be placed in Appendix as it is not essential for the main story.
P13 L286: I’d suggest to cut the sentence in two:
To examine the possible impact of different warming rates applied at the boundaries of the regional model (from ERA5 or EC-Earth), we compare the yearly average changes in air temperature and specific humidity. These averages are calculated along the boundary of the evaluation domain (i.e., the outermost grid cells of the evaluation domain and the innermost grid cells of the buffer zone) for the ERA5- and EC-Earth-driven hindcasts (Fig. 6). I understand you want to give important information (what is the boundary of the evaluation domain), but this often makes sentences quite confusing, with many elements in a single long sentence.P13 L296 : initially instead of Initially.
P13 L301 : ERA5-driven to be consistent.
P14 L305 : I suggest to seperate this part in two: one for ERA5, the other one for EC-Earth simulations, adding some connectors such as In contrast, but also avoiding the repetition of « a cooling trend ». This observation applies to many passages in the paper where the authors could simplify their sentences by avoiding repetitions of this kind and linking the same elements to shorten the sentences.
In the ERA5-driven hindcast, the cooling at the lateral boundaries mainly results in cooling over the Southern Ocean. The model simulates a cooling trend of −0.11 K per decade at 2 m and −0.15 K per decade at 700 hPa (Table 2). In general, the 2 m and 700 hPa temperature trends are consistent in both sign and significance, although their magnitudes differ slightly. This cooling is consistent with the simulated increase in sea ice extent (SIE) in the ERA5-driven hindcast, in agreement with previous findings (e.g., Comiso et al., 2017). No significant temperature trend is found over the Antarctic continent.
In contrast, the EC-Earth-driven hindcasts simulate warming over the Southern Ocean, with a mean 2 m air temperature trend of +0.30 K per decade (Table 2). They also produce positive 2 m temperature trends over Antarctica, although with larger inter-member variability (ranging from +0.14 to +0.27 K per decade) and lower robustness of the trend estimates (e.g., the trend is not statistically significant for member FC02). The higher variability over the continent compared to the ocean highlights the strong contribution of internal variability to multi-decadal temperature trends over Antarctica, consistent with the weakly significant trends reported in observations (Turner et al., 2016, 2020).
Figure 8: Yellow and green stripes are really difficult to see. Consider to add them on Figure 3 and removing on Fig8.
Table 3: How is surface mass balance computed ? From what I understand it is P-E, but this is a simplified definition and not the real surface mass balance.
P16 334-338 : surprising is not really suited for scientific publication. Furthermore, this is another example of long sentences with too much less important information. I suggest to rewrite these sentences.Because saturation water vapour pressure decreases with temperature, colder conditions are generally expected to reduce evaporation and limit moisture transport from the warmer mid-latitudes toward the continent, ultimately leading to lower precipitation. Nevertheless, the ERA5-driven hindcast indicates an increase in precipitation over the Southern Ocean of 196.9 Gt yr⁻¹ per decade (Table 3), despite the simulated cooling (Table 2). Over the continent, however, both temperature and precipitation trends remain insignificant.
The authors show a reduction in humidity at 700 Hpa, but this is not the only level that brings humidity. I think it would be better to show the integrated humidity transport across the entire air column and according to wind direction in Figure 6 (so MT boundary). This would help to better illustrate this “surprising” result, i.e., an increase in precipitation while decreasing in temperature.
P17L361: Remove surprising (also P19L389). To what extent could this apparent contrast be related to your non-uniform EC Earth bias on the air column?Figure 10 : Caption (a) ERA5 and (b) EC-Earth- driven hindcasts to be consistent with the remaining part of the manuscript but also reducing the lenght of the caption. In the figure, B should not be hindcats but EC-Earth if you set A to ERA5. For the rest, I'm not sure I understand the caption in relation to the graph (decrease where I see an increase according to the figures in the figure). Moisture transport (SMB) is also inadequate.
P19L397-398: rewrite the sentence and remove this is thanks (twice). P19 L397: Is it really better than EC-Earth? Although I agree with the authors, I don’t think they demonstrate this improvment in your paper. There are also stand alone NEMO simulation that can correctly reproduce the sea ice trend (Goose et al., 2024) so the authors cannot use this to justify the potential improvment in PARASO. Furthermore, as you mention, when forced by EC-Earth,P ARASO cannot reproduce the correct trend.
P20L405 : I am wandering what is the effect from ocean boundaries condition compared to the atmospheric. The authors never mention changes in the ocean boundaries.
P20L415 : The formulation suggesting that the chain of processes “reaches the SMB” is conceptually unclear. SMB is a diagnostic quantity resulting from multiple processes, rather than an endpoint that can be “reached.” Please rephrase for clarity.
Finally, I would like to apologize to the authors and editor for my delay in submitting this review.Citation: https://doi.org/10.5194/egusphere-2025-2889-RC2 -
AC1: 'Reply on RC2', Florian Sauerland, 07 Apr 2026
Major remarks
Comment 1 — Sauerland et al. present reconstructions of recent Antarctic climate using an almost fully coupled modelling framework. This is an interesting and valuable study, particularly given the well-known difficulty of running a fully coupled model without significant drift. From my understanding, the main message of the manuscript is that the atmospheric driving largely explains the simulated sea-ice trends, including the contrast with the opposite trends typically produced by GCMs. In this context, the improved physics of PARASO does not appear to substantially improve the representation of sea-ice trends compared to the driving GCM itself. It would therefore be helpful to show the direct results of the GCM (P21L460) in order to better assess the added value of the coupled framework.
Reply — First we want to thank you for your detailed comments which will help to significantly improve our manuscript. We agree with the assessment that the GCM runs should have been presented in more detail. As such, we have included the equivalents of figures 1, 2 and 4 for the driving EC-Earth runs in the supplement and discuss them in the result section now as well.
Comment 2 — Most of the simulated sea-ice trends and associated changes are attributed to atmospheric processes. However, the manuscript does not discuss potential changes at the oceanic boundaries. Overall, the manuscript is comprehensive, but it would benefit from a simplification of the text to better highlight the main storyline. This is particularly noticeable in Section 4.1 on sea ice. For example, parts of the caption of Figure 3 repeat information that is already obvious (e.g., stating that ice shelves are land). The manuscript would also gain clarity from a more structured development of the results and clearer links between the different analyses so that the central narrative emerges more distinctly.
Reply — Thank you for this comment. We will rework Sec. 4 (especially 4.1) and to improve clarity following your suggestions. We will also add the oceanic boundaries as a potential source of differences.
Comment 3 — I understand the rationale behind the debiasing of EC-Earth; however, the correction is applied at levels very close to the surface. This raises the question of whether the correction is dynamically consistent across the full atmospheric column. For instance, moisture advection typically occurs at levels well above the surface (e.g., around 850 hPa). If the correction is not vertically consistent, could it alter the dynamical forcing from EC-Earth or affect the structure of atmospheric circulation patterns? In that respect, could ERA5 be considered more reliable than EC-Earth even in the stratosphere, given that data assimilation is still performed at those altitudes and may constrain large-scale atmospheric properties more consistently?
Reply — Indeed, bias correction is always introducing the risk of creating imbalances in the dynamical and hydrostatic balances, for both vertically consistent and variable correction rates. The INT2LM preprocessor in COSMO-CLM aims to resolve these imbalances. The issues you mention above with bias correcting driving data for regional models are similar to issues that arise in the frequently used framework of the pseudo global warming approach. Here, re-analyses that drive the regional model are perturbed using the climate change signal from the GCM. The framework including the risk of imbalances is clearly described in Brogli et al. (2023). A discussion including this reference will be added to the manuscript.
As for the reliability of ERA5 and EC-Earth in the stratosphere, we discussed this for a while before creating the simulations and came to the conclusion that the stratosphere is not especially relevant for moisture transport. As there is little evidence to the reliability of both ERA-5 and EC-Earth in the stratosphere around Antarctica, we implemented the current version. We discuss this now in the discussion section.
Ref.: Brogli, R., Heim, C., Mensch, J., Sørland, S. L., and Schär, C.: The pseudo-global-warming (PGW) approach (2023): methodology, software package PGW4ERA5 v1.1, validation, and sensitivity analyses, Geosci. Model Dev., 16, 907–926, https://doi.org/10.5194/gmd-16-907-2023, 2023.
Comment 4 — Finally, although the authors provide a detailed evaluation of their configurations, the manuscript does not sufficiently discuss the implications of the identified biases for the simulations themselves. I would suggest including a comparison with in situ station data for atmospheric variables particularly wind rather than relying primarily on comparisons with ERA5 over the ice sheet, especially for the reanalysis-driven simulations. While this is not the most critical issue, it would strengthen the robustness of the evaluation.
Reply — Thank you for this suggestion. We agree that the lack of a comparison with in situ data is a weak point of the paper, adding such a detailed validation would therefore indeed be beneficial. However, an in-depth analysis in this regard would essentially mean adding another section to an already comprehensive manuscript. As it is not a critical issue, we therefore decided to focus on improving the rest of the paper, with the added note that ERA5 itself has been validated using in-situ data in other papers (e.g., Gossart et al., 2019), and put this under future work instead. Furthermore, we don’t expect PARASO to perform better than ERA5: as we do not use any form of data assimilation within the model, it would be already a fantastic result if the ERA5 results are resembled in the model.
Ref.: Gossart, A., S. Helsen, J. T. M. Lenaerts, S. V. Broucke, N. P. M. van Lipzig, and N. Souverijns, 2019: An Evaluation of Surface Climatology in State-of-the-Art Reanalyses over the Antarctic Ice Sheet. J. Climate, 32, 6899–6915, https://doi.org/10.1175/JCLI-D-19-0030.1.
Minor comments (Replies are in italic)
P2 L29 , however. Correct the . location.
We will move "however" to the beginning of the sentence.
P3 L66 : The authors could try to rewrite the sentence to avoid the repition of "boundary and initial conditions".
We will reformulate this and the following sentence to: "For this purpose, we created a total of four hindcasts with PARASO with different boundary and initial conditions: For the first experiment, we take these from ERA5 (atmosphere) and ORAS5 (ocean). For the other three experiments, we use slightly perturbed conditions from three different EC-Earth GCM hindcasts, which themselves start from the same but slightly perturbed initial conditions."
P3L80 : The authors can remove "in the boundary conditions"
We will do this.
P4 L112: Is this only related to your spectific set up or to a real absence of ice surface change over 30yr ? Is the model fully coupled then ? For me it’s not the case as there is no coupling of the ice sheet towards the atmosphere. (also for P20L430: I’d better use nearly fully-coupled model)
Indeed the ice sheet is not coupled back to the atmosphere, due to the changes in the ice sheet not being relevant enough to have an impact on the atmosphere. However, PARASO can take care of that feedback, so it is a fully coupled model. The feedback of the elevation was only ignored here since it is not relevant for this study. As PARASO is also called a fully coupled model in Pelletier et al., 2022, we continue calling it so here.
Ref.: Pelletier, C., Fichefet, T., Goosse, H., Haubner, K., Helsen, S., Huot, P.-V., Kittel, C., Klein, F., Le clec'h, S., van Lipzig, N. P. M., Marchi, S., Massonnet, F., Mathiot, P., Moravveji, E., Moreno-Chamarro, E., Ortega, P., Pattyn, F., Souverijns, N., Van Achter, G., Vanden Broucke, S., Vanhulle, A., Verfaillie, D., and Zipf, L.: PARASO, a circum-Antarctic fully coupled ice-sheet–ocean–sea-ice–atmosphere–land model involving f.ETISh1.7, NEMO3.6, LIM3.6, COSMO5.0 and CLM4.5, Geosci. Model Dev., 15, 553–594, https://doi.org/10.5194/gmd-15-553-2022, 2022.
P5L114: You can probably remove the two forcings occurences here. I’d also suggest to remove "regarding the forcing" P5 L121. I think there are enough (too much) “forcing” occurences in that paragraph so it’s clear you are talking about that.
Thank you for this remark, we will rewrite the paragraph and reduce the number of "forcings".
P5L124: Does this change of forcing frequency have an impact on the ocean model results ?
Although we did not test that separately, given the slower evolution of ocean properties compared to atmosphere and other climate components this is unlikely to cause any major disturbances on ocean properties. In addition, the forcing frequency only plays a role outside the coupled domain (where NEMO is forced by ERA5/EC-Earth). The air-sea coupling time step within the coupled domain is the same whether the model is forced by EC-Earth or ERA5 at its boundaries
P6L172 : Consider to replace "snow mantle" by snowpack or snowpack properties (and then list wich properties you are initializing)
A good suggestion that we have implemented accordingly!
P6L189 : in instead of In.
We will change this.
P10 L239-251: I’d suggest the author to move these paragraphes at the end of the method section as they also explain where you did the evaluation, and especially the Figure 3 should be placed before the results of the evaluation and the becoming Fig 1.
Thank you for this suggestion, we will move these paragraphs. We will also move Fig. 3 to Section 2.1.
P12 L282 – P13 L284: Isn’t that a bit too simplistic to justify the decrease in sea ice in this region?
This sentence was indeed oversimplified, we will reformulate it to a weaker statement.
Figure 6 could be placed in Appendix as it is not essential for the main story.
Figure 6 will be reworked to include moisture transport trends and the entire vertical columns — this should make it more relevant. However, in order to reduce the total number of figures and to improve comparability we will combine Figures 5, 7, and 8 to a single figure.
P13 L286: I’d suggest to cut the sentence in two:
To examine the possible impact of different warming rates applied at the boundaries of the regional model (from ERA5 or EC-Earth), we compare the yearly average changes in air temperature and specific humidity. These averages are calculated along the boundary of the evaluation domain (i.e., the outermost grid cells of the evaluation domain and the innermost grid cells of the buffer zone) for the ERA5- and EC-Earth-driven hindcasts (Fig. 6). I understand you want to give important information (what is the boundary of the evaluation domain), but this often makes sentences quite confusing, with many elements in a single long sentence.
Also in response to your major comment 2 we will improve Section 4 and break up long sentences where possible.
P13 L296 : initially instead of Initially.
Thank you
P13 L301 : ERA5-driven to be consistent.
Thank you.
P14 L305 : I suggest to seperate this part in two: one for ERA5, the other one for EC-Earth simulations, adding some connectors such as In contrast, but also avoiding the repetition of « a cooling trend ». This observation applies to many passages in the paper where the authors could simplify their sentences by avoiding repetitions of this kind and linking the same elements to shorten the sentences.
In the ERA5-driven hindcast, the cooling at the lateral boundaries mainly results in cooling over the Southern Ocean. The model simulates a cooling trend of −0.11 K per decade at 2 m and −0.15 K per decade at 700 hPa (Table 2). In general, the 2 m and 700 hPa temperature trends are consistent in both sign and significance, although their magnitudes differ slightly. This cooling is consistent with the simulated increase in sea ice extent (SIE) in the ERA5-driven hindcast, in agreement with previous findings (e.g., Comiso et al., 2017). No significant temperature trend is found over the Antarctic continent.
In contrast, the EC-Earth-driven hindcasts simulate warming over the Southern Ocean, with a mean 2 m air temperature trend of +0.30 K per decade (Table 2). They also produce positive 2 m temperature trends over Antarctica, although with larger inter-member variability (ranging from +0.14 to +0.27 K per decade) and lower robustness of the trend estimates (e.g., the trend is not statistically significant for member FC02). The higher variability over the continent compared to the ocean highlights the strong contribution of internal variability to multi-decadal temperature trends over Antarctica, consistent with the weakly significant trends reported in observations (Turner et al., 2016, 2020).
Thank you, we will follow your suggestion.
Figure 8: Yellow and green stripes are really difficult to see. Consider to add them on Figure 3 and removing on Fig 8.
This is actually a great suggestion as it also makes Figure 8 more clear and comparable to the other figures of its kind!
Table 3: How is surface mass balance computed ? From what I understand it is P-E, but this is a simplified definition and not the real surface mass balance.
The given numbers are indeed P-E, this is not clearly communicated and we updated the caption now. This is indeed a simplification, but the only other component calculated in PARASO is the melt, which is so small over the ice sheet that including this would not change the trends. Redistribution of blowing snow is not included in PARASO, which is a limitation that will be mentioned in the discussion.
P16 334-338 : surprising is not really suited for scientific publication. Furthermore, this is another example of long sentences with too much less important information. I suggest to rewrite these sentences.
We will remove the words surprising from the manuscript.
Because saturation water vapour pressure decreases with temperature, colder conditions are generally expected to reduce evaporation and limit moisture transport from the warmer mid-latitudes toward the continent, ultimately leading to lower precipitation. Nevertheless, the ERA5-driven hindcast indicates an increase in precipitation over the Southern Ocean of 196.9 Gt yr-1 per decade (Table 3), despite the simulated cooling (Table 2). Over the continent, however, both temperature and precipitation trends remain insignificant.
Thank you for this comment, this is indeed correct. The reduction in specific humidity due to the cooling must therefore be more than compensated for by the changing atmospheric dynamics. We will add this aspect to the manuscript.
The authors show a reduction in humidity at 700 Hpa, but this is not the only level that brings humidity. I think it would be better to show the integrated humidity transport across the entire air column and according to wind direction in Figure 6 (so MT boundary). This would help to better illustrate this “surprising” result, i.e., an increase in precipitation while decreasing in temperature.
We think this is a good way of making Figure 6 more relevant and to make our findings more robust — Figure 6 will be changed and the discussion of it adapted accordingly.
P17L361: Remove surprising (also P19L389). To what extent could this apparent contrast be related to your non-uniform EC Earth bias on the air column?
Thank you, we will check the paper for other surprising occurrences of the word "surprising". The EC-Earth debiasing is unlikely to be an issue here as we so far only compared the 700 hPa level, right at the boundary, which following Figure A1 is not severely affected by biases. Also, as mentioned in our reply yo your related comment, the INT2LM preprocessor aims to ensure some consistency here as well. The extension of our analysis to the entire air column (see reply to your previous comment) and visualisation of the EC-Earth output should fully clear this up, but applying a climatological bias correction should not be causing any trends in itself.
Figure 10 : Caption (a) ERA5 and (b) EC-Earth- driven hindcasts to be consistent with the remaining part of the manuscript but also reducing the lenght of the caption. In the figure, B should not be hindcats but EC-Earth if you set A to ERA5. For the rest, I’m not sure I understand the caption in relation to the graph (decrease where I see an increase according to the figures in the figure). Moisture transport (SMB) is also inadequate.
Thank you, we will modify the figure and caption following your suggestions. We will use EC-Earth-driven and ERA5-driven consistently throughout the paper and in the figures.
P19L397-398: rewrite the sentence and remove this is thanks (twice). P19 L397: Is it really better than EC-Earth? Although I agree with the authors, I don’t think they demonstrate this improvment in your paper. There are also stand alone NEMO simulation that can correctly reproduce the sea ice trend (Goose et al., 2024) so the authors cannot use this to justify the potential improvment in PARASO. Furthermore, as you mention, when forced by EC-Earth, PARASO cannot reproduce the correct trend.
Thank you for this observation. Also in response to some of your previous comments, we add the equivalent of Figs. 1, 2, and 4 for EC-Earth data (i.e. the input data). Especially the comparison of PARASO - ERA5 (Fig. 2) vs. EC-Earth - ERA5 (new figure) should sufficiently show the added value of PARASO compared to EC-Earth alone. We furthermore modify the discussion to take some focus away from this aspect which is an important, but not the whole outcome of our study.
P20L405 : I am wandering what is the effect from ocean boundaries condition compared to the atmospheric. The authors never mention changes in the ocean boundaries.
This is a valid point, and we will include a brief separate discussion of the influence of ocean boundary conditions in the manuscript. We believe that they play a large role in causing the difference between EC-Earth and ERA5-driven data, especially as they are very similar for the EC-Earth driven runs (see Figure A4). Generally though, the conclusions we drew are valid for the influence of both oceanic and atmospheric boundary conditions.
P20L415 : The formulation suggesting that the chain of processes “reaches the SMB” is conceptually unclear. SMB is a diagnostic quantity resulting from multiple processes, rather than an endpoint that can be “reached.” Please rephrase for clarity.
Thank you, we will rephrase this accordingly.
Finally, I would like to apologize to the authors and editor for my delay in submitting this review. Thank you for your comments!
Additional remark: During the review process, we noticed a wrong attribution of the observational dataset used to validate modelled sea ice extent (OSI). For consistency, we now take the observations from ERA5 (as in ERA5 sea ice is input data based on the OSI-SAF dataset). We will adapt the manuscript accordingly. The results and interpretation remains the same.
Citation: https://doi.org/10.5194/egusphere-2025-2889-AC1
-
AC1: 'Reply on RC2', Florian Sauerland, 07 Apr 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,418 | 153 | 42 | 1,613 | 53 | 57 |
- HTML: 1,418
- PDF: 153
- XML: 42
- Total: 1,613
- BibTeX: 53
- EndNote: 57
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
A review of the paper
EC-Earth- and ERA5-driven ensemble hindcasts with the fully coupled ice-sheet–ocean–sea ice–atmosphere–land circum-Antarctic model PARASO
By Florian Sauerland and co-authors
The paper of Florian and co-authors concerns a very important problem – to what degree can we trust to the GCMs when we apply them to simulate the climate round the Antarctica. The clear answer which follows from their study is “no, we cannot”. And this is the main conclusion of the research.
In other words, the authors aim to focus on pit-falls of the global climate models which purpose is to predict the possible future of the Antarctic climate. This is a crusial challenge – to separate imprints of the anthropogenic climate forcing from the climate natural variability. If the modern climate models fail to reproduce Antarctic climate in the nearest past, can we rely on their predictions? The problem is aggravated by the limit amount of the historical observations on the region.
It is also true that the polar regions are significantly different in many aspects from the other regions of the globe. This is the reason for the obvious mismatches between observations (or the reanalysis data) and modeling results.
These mismatches are clearly demonstrated by collating results of the numerical simulations in which the model PARASO is forced with the reanalysis and with the GCM simulations (sea ice extent, 2m air temperature etc.). Since the GCMs demonstrate not only the “wrong” figures but even opposite trends in climatic variables for the historic periods, one become rather sceptical about fasibility of their application to predict the climate of the Antarctica.
The manusript is well structured and clearly written (except some particular details mentioned below). The abstract provide a clear summary of the paper. Authors provide appropriate supplementary materials which illustrate the text. In overall, the study can make a valuable contribution to our understanding and modeling the climate dynamics in the polar regions. In this view the paper clearly addresses very important scientific questions within the scope of ESD.
There are several major remarks:
Other cpmments (some of them intersect with the above ones)
Line 15: “internal climate variability”. What is it? According to the WMO definition climate is measured by 30-yr period
Line 63: daily-weekly time step to couple an ice sheet to other components of a system. Such small steps seem to be excessively small. A comment here would be pertinent.
Line 67: three experiments can hardly be called an “ensemble”. This is just only three numerical experiments with sligtly different perturbed boundary conditions of the same model. Of course, no statistics can be ruled out of the results, which is the essence of the ensemble approach in climate modeling.
Lines 112-113: I see a contradiction here. In case surface changes of the Antarctic Ice Sheet (AIS) are negligible during a 30-yr numerical experiment (and it is in the real world except may be the processes on the boundary ice sheet-ice shelf), why the model is called “fully coupled” in the context of the AIS? I suppose that the actual geometry of the AIS is used, isn’t it?
Line 181: 20-yr or 20-kyr? 20 years seems a rather short period of relaxation, no?
Lines 210 and 211: Correct sea name is “BellingShausen”
Figure 1: More illustrative would be to demonstrate also absolute and relative differences in ice extent. I think the conclusions would be less optimistic: good coincidence only in November-December period.
Lines 213-221: Discussion on the sea-ice extent simulations. The same note as to Fig. 1. I think, a short discussion will be plausible on how the mismatches in sea ice extent affect the overall simulation.
Lines 242-245: “Since all EC-Earth driven experiments share quasi-identical initial states, all of those simulations are affected by similar forcing signals (due to greenhouse gases etc.) and model drift, while the timing of the internal variability may vary. Differences in long-term trends should therefore be the result of internal climate variability, generated outside and/or inside the domain.” Not necessarily. These trends can aslo indicate the sensitivity of the model to the initial conditions or to different setup of FC00, FC01 and FC02. Isn’t it?
Section 4.1 (also line 309): What is “pluri-decadal” in the context of this paper? Ten days or ten years? In the second case statistics has no sence. In the text, trends over 30 years (30 septembers and 30 marches) are described (i.e. in Table 1). Then the term “decadal” is really confusing.
Line 267: “ensemble averaged”: does it mean an avearge over three values?
Line 267-268: The follwing is not clear: “The EC-Earth hindcasts themselves are not showing a negative SIE trend (Mann-Kendall test for the ensemble averaged September SIE: p = 0.972”. In the figure 4 one can clearly see a negative trend in EC-Earth hindcasts – from ~17.5 mln. sq. km to ~15 mln. sq. km over 30 years (meaning over 30 septermbers). Shown in Table 1 treds are statistically significant …
Line 288: “bufferzone”: may be “buffer zone”?
Figures 7 and 8: Shown are anomalies, not absolute values, aren’t they? Then the colour scales and figure captions must be changed to avoid confusion.
Figure 8: The color scale is indicated as 700 mb but in the figure caption is written 850 mb. What is true?
Table 3: It is not clear over which decade averaging is carried out. The indicated period is 1985-2014. Are the trends estimated over annually averaged values? Isn’t the period of ten years (ten values) is long enough to figure out a realistic trend? By what means the trends were estimated?
Figure 9: Evaporation and precipitation TRENDS
Line 357: “After all, precipitation that does not have a local source has to have an origin somewhere else” – This is not entirely correct if the internal part of the Antarctca is considered. Here the precipitation has another (local) mechanism and is formed at the upper boundary of the inversion layer. Anyway, the related part of the text must be clarified in case anything else was supposed.
Table 4. The same note as for the Table 3. What decade is taken? Or the average among 3 decades is indicated? What do the authors mean under “decadal trend”?
Line 425: “the slow ocean variability”: the ice sheet is much less variable compared to the ocean, may be except the gronding line migration.
Lines 427-428: “With our approach where we have three hindcasts driven by EC-Earth, we can better differentiate internal variability from interannual variability”. The arguments in support of this statement seem not to be very strong. What was clearly demonstarted is that the GCM output driven model PARASO failed to reproduce some crusial regional climate features compared to the case when PARASO was forced by the reanalysis. By itself, this would be a very important conclusion. It is also not clear what “internal” variability means.