the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
The evolution of future Antarctic surface melt using PISM-dEBM-simple
Maria Zeitz
Uta Krebs-Kanzow
Ricarda Winkelmann
Abstract. It is virtually certain that Antarctica's contribution to sea-level rise will increase with future warming, although competing mass balance processes hamper accurate quantification of the exact magnitudes. Today, ocean-induced melting underneath the floating ice shelves dominates mass losses, but melting at the surface will gain importance as global warming continues. Meltwater at the ice surface has crucial implications for the ice sheet's stability, as it increases the risk of hydrofracturing and ice-shelf collapse that could cause enhanced glacier outflow into the ocean. Simultaneously, positive feedbacks between the atmosphere and the ice elevation and albedo can accelerate mass losses and increase the ice sheet's sensitivity to warming. However, due to long response times it may take hundreds to thousands of years until the ice sheet fully adjusts to the environmental changes. Therefore, ice sheet model simulations must be computationally fast and capture the relevant feedbacks, including the ones at the ice–atmosphere interface.
Here we use the novel surface melt module dEBM-simple, coupled to the Parallel Ice Sheet Model (PISM), to estimate the impact of 21st-century atmospheric warming on Antarctic surface melt and long-term ice dynamics. As an enhancement compared to the widely adopted positive degree-day (PDD) scheme, dEBM-simple includes an implicit diurnal cycle and computes melt not only from the temperature, but also from the influence of solar radiation and changes in ice albedo, thus accounting for the melt–albedo feedback. We calibrate PISM-dEBM-simple to reproduce historical and present-day Antarctic surface melt rates given by the regional climate model RACMO2.3p2 and use the calibrated model to assess the range of possible future surface melt trajectories under SSP5-8.5 warming projections, extended beyond 2100 under fixed climatological conditions. Our findings reveal a substantial speed-up in ice flow associated with large-scale elevation reductions in sensitive ice-sheet regions, underscoring the critical role of self-reinforcing ice-sheet–atmosphere feedbacks on future mass losses and sea-level contribution from the Antarctic Ice Sheet on centennial to millennial timescales.
- Preprint
(12581 KB) -
Supplement
(23365 KB) - BibTeX
- EndNote
Julius Garbe et al.
Status: final response (author comments only)
-
RC1: 'Comment on tc-2022-249', Anonymous Referee #1, 07 Feb 2023
Summary
This paper presents a new surface melt scheme in the Parallel Ice Sheet Model (PISM) coined dEBM-simple. The scheme is forced by air temperature and includes parameterizations for solar radiation, atmospheric transmissivity, and albedo evolution. The authors first tune their model using historical melt simulated by the regional climate model RACMO2.3p2 forced by CESM2. Ensuing comparison between RACMO2-CESM2 and dEBM-simple reveals a reasonably strong agreement in the simulation of mean annual and monthly surface melt rates. Based on this, the authors then force their model to 2100 using output from a SSP5-8.5 forcing of RACMO2-CESM2, and then to 2300 by repeating end-of-century forcing. Results show increasing melt, runoff, and snowfall, ice acceleration, and the importance of ice-albedo and ice-elevation feedbacks.
This is a very well written manuscript supported by clear, nicely illustrated figures. The introduction to the paper provides a great overview of the science and motivation for the work. Development of the dEBM-simple model is an important effort to bridge the gap between simple PDD melt schemes and more complex and computationally expensive energy balance approaches implemented in regional climate models. The model and methods are well described and the uncertainty quantification by varying parameter values is welcome. I do have several concerns that I would like the authors to address, particularly as it relates to the evaluation of the dEBM-simple results and determination of meltwater runoff.
I thank the authors for their time in considering my evaluation.
Major comments
The authors claim the validity of their new dEBM-simple approach by comparing it with historical (1950-2015) melt rates simulated by RACMO2. The major issue I see with this is that model parameters were specifically tuned to match RACMO2 over this same period, and thus this comparison does not represent an independent check on dEBM-simple’s validity.
I am also not fully convinced that dEBM-simple is doing a better job than the PDD scheme. For example, all analyses for the historical period are quite similar, and the PDD scheme better matches RACMO2’s melt magnitude at the end of the century (Table 2). In addition, the maps of “present-day” (via the CESM2 historical scenario) melt (Figure 2a, Figure S7a) and their difference with RACMO2 (Figure 2b, Figure S7b) seem to suggest that the PDD scheme may be doing a better job at capturing some of the spatial characteristics of melt (e.g., across Ross, Ronne-Filchner, the Ross/Amundsen/Bellingshausen coasts) compared to RACMO2-CESM2 that is shown in the Figure S4 inset. The RACMO2 data in the Figure S4 inset also seem to better agree with the spatial distribution of melt determined via satellite observations (cf. the cited Trusel et al 2013 paper), which suggests that dEBM-simple is underestimating melt in these low-melt regions. Note that this is counter to what the authors describe in the text of a general overestimation of melt in low-melt areas. The maps of dEBM-simple and PDD vs RACMO2 over the AP also seem to show lower biases in melt over the high-melt AP using the PDD scheme. The statistics listed (slopes and R values) are all quite similar as well, so I am somewhat concerned about whether dEBM-simple is doing a better job.
The coastline/ice shelf extent displayed in Figure 2 and other maps is not an accurate reflection of the current extent of grounded and floating ice in Antarctica. For example, most ice shelves in West Antarctica appear to be missing, as is George VI on the AP. I presume this is because the PISM model does not simulate ice shelves here? Could the authors please comment on this in the text? Also, while ice shelves of Queen Maud Land appear to be present, they do not all seem to be marked as ice shelves. For example, the melt pattern of what looks to be Roi Baudouin ice shelf exists, but it’s not shown as an ice shelf. Why is this?My last, and perhaps most important, major concern is with how runoff is calculated. The authors estimate 50% of surface melt becomes runoff. In comparison, the RACMO2.3p2 melt/runoff ratio over the contemporary period is ~6%, and in 2090-2100 is ~23%. Over the contemporary period, the dEBM-simple results therefore unrealistically suggest 9x the amount of runoff compared to RACMO2. At the end of the century, dEBM-simple produces ~160% of the runoff of RACMO2, yet only 70% of the amount of RACMO2’s melt. Estimating runoff to be fixed at 50% of melt is not physical, and this has very important consequences for other conclusions of the presented manuscript including constraining the future SMB, total mass budget, surface elevation changes, and ice dynamics.
Minor comments
L17: Please clarify here if the speed up in ice flow from elevation reductions are related to SMB decreases
L58: Regarding supraglacial lakes “play a major role in the ice sheet mass balance in East Antarctica” – My understanding is that the cited Stokes et al 2019 and Arthur et al 2022 papers assess the presence and variability of supraglacial lakes, but not their role in the ice sheet mass balance. Given that the cited papers do not discuss runoff of water from the lakes to the ocean (to my knowledge), the lakes are important in an energy balance/surface hydrology/ice shelf stability perspective, but not currently important in terms of the ice sheet mass budget.
L97: Description of PDD schemes here not entirely correct. They do not assume melt is “proportional to the number of days” above zero, but rather the cumulative sum of air temperatures above zero. I believe you more correctly describe PDD schemes later in the manuscript.
L245: Does the atmospheric transmissivity calculation include the effects of cloud climatologies in any way? This seems like it would be a relatively easy way to crudely account for cloudiness and perhaps reduce discrepancies with RACMO2 in cloudy areas like on the western AP, where dEBM-simple produces more melt than RACMO2.
L263: It’s stated here that “RACMO data show no clear dependence between melt and albedo values under historic and present-day climate conditions”, but is this true? For example, the cited Jakobs et al 2021 paper uses RACMO2 to show the melt-albedo feedback is important. Other work, like that of Lenaerts et al 2016 (Nature Climate Change; doi:10.1038/NCLIMATE3180 ) also show the importance of the melt-albedo feedback.
L370: Change to “historical”
Figure 1c/d: The units here (and associated manuscript text) regarding monthly melt rates are confusing to me. First, should these be per month, not year (as these are monthly melt rates)? Second, are the values actually mm w.e. averaged over some area, not Gt? I fail to see how monthly melt rates could be several hundred Gt, yet yearly rates shown in panel a are <160 Gt/yr.
L450: When discussing how dEBM-simple tends to underestimate melt in high-intensity regions and overestimate melt in low-intensity regions (notwithstanding my above ‘major’ comments to this regard), it would be helpful to include as supplementary figures maps of the difference between dEBM- and PDD-derived melt and RACMO2 across all of Antarctica, both for the present-day and future. This would allow for a better understanding of where (and potentially why) discrepancies exist between the methods.
Figure 2: There’s an apparent circular/wavy pattern appearing between 1000 and 1500 mm w.e./yr. Could you comment on what is producing this? This also appears in Figure S7.
Figure 2c, S7c: Please define what “n” is.
Figure 3: The colors for melt and runoff are hard to distinguish. Please use a different color for runoff. Also, in the caption, it states that positive values of surface melt and runoff denote mass losses. Presumably, this should only say that runoff is mass loss, correct? Lastly, the albedo map in panel c is difficult to assess as it is not compared with present-day albedo. I would suggest rather than plotting the absolute value of albedo, the difference with present-day albedo could be plotted.
L474+L492: This is actually the “western” margin considering south is up.
L657: Again, I suggest the authors create maps of difference between their results and RACMO2 because the regional/spatial perspective (i.e., “high-intensity melt regions” and “low-intensity melt regions”) cannot be assessed in the scatter plots (i.e., Figures 1d, 2c, etc.).
L693: Following on from my final “major” comment – I do not think the uncertainty produced by assuming runoff to be 50% of melt is properly explored. What would happen if runoff was fixed at 10%? 25%? Alternatively, what would the future ice dynamical evolution be if PISM was forced using (presumably more reliable) runoff rates prescribed directly from RACMO2?
Code and data availability: I would encourage the authors to consider uploading their code (particularly that to make the figures) to GitHub or Zenodo, rather than making it available “upon reasonable request”. The figures are all very nicely designed, and the community would benefit by being able to easily look at the underlying code!
Citation: https://doi.org/10.5194/tc-2022-249-RC1 -
RC2: 'Comment on tc-2022-249', Ella Gilbert, 09 Feb 2023
Review of “The evolution of future Antarctic surface melt using PISM-dEBM-simple”, Garbe et al. submitted to The Cryosphere Jan 2023
Summary
The manuscript explores present and future Antarctic surface melting using a new surface melt module, dEBM-simple, coupled to the ice sheet model, PISM. The authors evaluate their configuration’s robustness with respect to surface melting as calculated by a positive degree day (PDD) model and the RACMO regional climate model. They show good agreement between their surface melt results and those produced by the more sophisticated model, RACMO. They emphasise the relative computational efficiency of dEBM-simple-PISM in comparison to running a complex model like RACMO and its superiority over PDD-based melt estimates.
The manuscript is very well presented, with a compelling argument and clear figures. It is a welcome contribution to the field that showcases an important tool. I have some general comments and suggestions that I feel would improve the manuscript, which are detailed below. My main concerns relate to the method of tuning present day / historical melt parameters to the same model that is used for validation (especially without a thorough discussion of RACMO’s own errors and limitations) and the need for more justification of what we can learn from the simulations out to 3000. However, in general I think it would be highly suited to publication in the journal, subject to the authors making adjustments in light of my and other reviewers’ comments.
Thanks to the authors for an interesting paper. EG
General comments / suggestions
Further discussion of the limitations of the PDD method could be included (e.g. in the lit review) to set up the significance of the work and usefulness of dEBM-simple
Some (brief) quantitative comparison between the simple/full versions of dEBM results would be informative
The temperature-melt index of Orr et al. (2022) could also be an interesting comparison for your work, to put your results into further context. https://doi.org/10.1175/JCLI-D-22-0386.1
How well do RACMO / dEBM-simple capture melt associated with orographic features around the edges of the ice sheet e.g. foehn winds / adiabatically warmed katabatic outflow? I’m thinking especially of the Antarctic Peninsula – it seems from Fig 2 that there is limited melt adjacent to the mountains over Larsen C for example. At 27 km it is doubtful that RACMO will capture these sort of dynamics – even the 8 km PISM grid might be too coarse. What does this mean for melt estimates?
The tuning of melt parameters to RACMO foreshadows the results. Although RACMO is undeniably a good model for estimating melt, it still contains errors and there is limited discussion of this in the paper. It would be better if dEBM-simple could be independently validated, for example against observational/satellite datasets, and then compared with RACMO. Otherwise the comparison in sect 5.1 against RACMO is somewhat meaningless because if you tune your melt parameters to match RACMO output, it’s unsurprising that the results are similar.
Clearly it is difficult to project beyond 2100 without any kind of post-2100 emissions scenarios. I recognise that you have attempted to address the lack of such input data here, but how useful is the fixed 2100 simulation for telling us about the deep future? It tells us more about the feedbacks and impacts of the 21st century high-emissions scenario than anything beyond 2100. I would like to see a little more justification and discussion of what we can learn from this particular experiment in Sect 5.4.
I have also noted this below, but I would benefit from better explanation of how the isolation-dependent and temperature-dependent components of melt are separated.
Specific comments
170-172 Does the thinning/ loss of ice shelves feed back on the speed of glaciers and therefore ice discharge? Apologies if I’ve missed this elsewhere.
230 Is RACMO melt corrected/validated before optimising to it? Optimising or tuning to an RCM (even a good one) still introduces error. It is also not an entirely independent comparison for the results.
232-234 So, this means half of the meltwater generated runs off the surface? This is surely a significant overestimate for the present climate, when very little runoff occurs except over ice shelves? As far as I undertand, 50% runoff may be a valid assumption for ice shelves by the end of the century (c.f. Gilbert & Kittel, 2021 – Fig 3) but it still strikes me as high for the grounded ice sheet.
292-295 Choice of averaging period for climate data (RACMO) – why this one? Does it affect the boundary conditions? And why does it differ from the ocean forcing? (presumably because of data availability?)
296 Are the climatic BCs just repeatedly applied? E.g. the same climatology is applied every year for 22000 years?
310 - 314 Here you justify your use of RACMO2.3. I think this would be strengthened by acknowledging the biases in RACMO too (if I recall correctly from the Mottram et al paper, RACMO still under-estimated the SMB slightly, although less than some of the other ensemble members)? You could also note that RACMO has one of the more sophisticated surface schemes, which has feedbacks on the quality of its atmospheric outputs.
323 “The precipitation field is independent of the evolving ice-sheet geometry” – meaning that there are no changes in orography-precipitation interactions as the ice sheet evolves? Could be worth spelling out the implications of this statement.
Fig 1 caption. Add that RMSE values are shown for each model compared to RACMO in panel a)
419-422 but the melt peak is captured well in Fig 1c. , with virtually zero difference between RACMO/the 2 models in Jan. This is encouraging given that this is when melt is most intense
423 Missing processes such as?
425 did you do any sensitivity tests to explore the impact of piecewise vs default interpolation?
Sect 5.2 / Tab 2 This is perhaps a point for the discussion, but it would be interesting to see how your results compare to previous estimates of future SMB change, e.g. Kittel et al 2021 (https://tc.copernicus.org/articles/15/1215/2021/), Lenaerts et al. 2016 (https://link.springer.com/article/10.1007/s00382-015-2907-4), Donat-Magnin et al 2021 (https://tc.copernicus.org/articles/15/571/2021/)
500-501 As per my previous point re: resolution/the hydrostatic nature of RACMO, could this under-estimate actually be even greater if RACMO itself is under-estimating high-intensity melt hotspots? Although I acknowledge that you state that this has limited impact on overall totals…
509+ Worth re-stating here that the melt equation describes average melt **when temperatures are above the melt threshold ** I was originally confused by this as melting can of course only occur when there is a surplus of energy available to do melting and (sub-)surface temperatures are at the melting point. But re-reading section 2.2.2 I see the equation considers only melting when the temperature condition is met. A little more explanation of what the separate components of ‘temperature-driven’ and ‘SW-driven’ melt really mean would be welcome.
What causes SW-driven melt to increase? Presumably this is related to the albedo feedback darkening the surface, reducing SWup and resulting in a measurable difference in SWnet?
Fig 5 I can’t really see the purple shading in the main panel - can you make it darker? Also the grey text in the smaller panel is too light to read unless I zoom in really far and squint!
Para beginning 620 – not sure this summary is needed. You can probably just launch straight into your discussion points.
643 How does the temperature-dependent split of rain/snow compare against rain/snow inputs from RACMO? Did you look at that?
700 Evaporation may also become more important in future (especially under strong warming scenarios like SSP5-8.5)
Technical corrections
411 remove comma after “both”
531 “as high as few degrees” à “as high as a few degrees”
622 “to serve as full-fledged” à “to serve as a fully-fledged”
706-710 Very long sentence! Suggest splitting into two.
Citation: https://doi.org/10.5194/tc-2022-249-RC2
Julius Garbe et al.
Julius Garbe et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 374 | 150 | 13 | 537 | 43 | 11 | 6 |
- HTML: 374
- PDF: 150
- XML: 13
- Total: 537
- Supplement: 43
- BibTeX: 11
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1