the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Jan 2024
Evaluating an accelerated forcing approach for improving computational efficiency in coupled ice sheet-ocean modelling
Abstract. Coupled ice sheet-ocean models are increasingly being developed and applied to important questions pertaining to processes at the Greenland and Antarctic Ice Sheet margins, which play a pivotal role in ice sheet stability and sea level rise projections. One of the challenges of such coupled modelling activities is the timescale discrepancy between ice and ocean dynamics. This discrepancy, combined with the high computational cost of ocean models due to their finer temporal resolution, limits the time frame that can be modeled. In this study, we introduce an “accelerated forcing” approach to address the timescale discrepancy and thus improve computational efficiency in a framework designed to couple evolving ice geometry to ice shelf cavity circulation. This approach is based on the assumption that the ocean adjusts faster to imposed changes than the ice sheet, with the ocean viewed as being in a slowly varying quasi-steady state over timescales of ice geometry change. By assuming that the ocean-induced ice draft change rate during one coupling interval can be reflected by a quasi-steady state change rate during a shortened coupling interval (equal to the regular coupling interval divided by a constant factor), we can reduce the ocean model simulation duration. We first demonstrate that the mean cavity residence time, derived from stand-alone ocean simulations, can guide the selection of suitable scenarios for this approach. We then evaluate the accelerated forcing approach by comparing basal melting response under the accelerated forcing with that under the regular forcing based on idealized coupled ice sheetocean experiments. Our results suggest that: the accelerated approach can yield comparable melting responses to those under the regular forcing when the model is subjected to steady far-field ocean conditions or time-varying conditions with timescales much shorter than the cavity residence time. However, it is not suitable when the timescale of the accelerated ocean conditions is not significantly different from the cavity residence time. When used carefully, the accelerated approach can be a useful tool in coupled ice sheet-ocean modelling.
- Preprint
(14058 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on gmd-2023-244', Nicolas Jourdain, 12 Mar 2024
The comment was uploaded in the form of a supplement: https://gmd.copernicus.org/preprints/gmd-2023-244/gmd-2023-244-RC1-supplement.pdf
- AC2: 'Reply on RC1', Qin Zhou, 05 Jul 2024
-
RC2: 'Comment on gmd-2023-244', Moritz Kreuzer, 01 Apr 2024
Summary
The study evaluates the applicability of an accelerated forcing methodology in the scope of high resolution ice-ocean coupling. The authors motivate their investigation by the typical time scale discrepancy between ice and ocean dynamics and the corresponding disparity in simulation time. First, the idea and methodology of an accelerated coupling approach is presented and the models used in the study are described.
Then, standalone setups of two ocean models are used to derive ice-shelf melt rates for warm, cold and mean far-field ocean forcing conditions, as well as oscillating profiles between the warm and cold case at different frequencies. The authors use the mean cavity response time (MCRT; for the mean forcing case) as a characteristic variable to evaluate the derived melt rates. They find that averaged cavity melt rates for oscillating far-field forcings that have significantly higher frequencies than the MCRT (periods <=10% of MCRT) are mostly in the range of 90-110% of melt rates from time averaged forcing. However, forcing frequencies that are in the same order of the MCRT or substantially lower do either under- or overestimate the mean melt rates. Based on these ocean standalone simulations the authors anticipate that the accelerating forcing approach would only work in the first case, which they subsequently test in coupled ice-ocean simulations.
The test scenarios for coupled ice-ocean simulations are structured in three categories: 1. constant cold-to-warm forcing as well as two periodic forcings with fast (2.) and slow (3.) varying time scales. For all categories different acceleration factors are tested (between 1.5 and 10) and evaluated to the baseline scenario (no acceleration) in terms of cavity averaged melt rates and total ocean volume changes (time series) as well as spatially fields of melt rates and ice draft changes. The authors find that acceleration works well for spun-up simulations in the constant forcing as well as the fast-varying ocean forcing case, but not in the slow-varying case.
Finally, the authors conclude, that their presented approach of accelerated forcing in high-resolution ocean-ice coupled models is also applicable in real-world applications for ocean to ice forcing that varies over century-long timescales.General comments
The paper addresses a relevant and scientifically interesting topic which fits very well in the scope of GMD. It presents a novel investigation of testing the impact of asynchronous coupling in idealized setups and introduces a useful metric (MCRT) to assess the applicability of the approach. The study has a sound methodology and follows a clear experimental design with valid, clear and justified assumptions. The manuscript is well written, with fluent and precise language. The title is appropriate and reflects the contents of the paper well. The abstract provides a concise and complete summary of the presented work.
I have no major concerns, but a few remarks. More specific comments and suggestions for improvement are given in the attached pdf.
The introduction motivates the following work well. However, more background on previous work about asynchronous ice-ocean coupling would be great to give the reader more context to the study like: What studies have used asynchronous coupling so far? Is there already some literature that compares synchronous vs asynchronous coupling?
I have a few general remarks for plots (more specific ones are given as annotations to the pdf):
- for all spatial plots of ISOMIP+ domain (barotropic stream function, melt rates, ice draft changes): it would be helpful to either mark the grounding line as a line or to shade the grounded areas (e.g. light gray) to be able to distinguish regions with values close to zero and grounded areas.
- Do 2d plots show values out of the colorbar range (e.g. Fig 3c)? If so, please indicate this by adding out-of-range extensions to the colorbars and give maximum values in text/caption.
- If contour spacing can't be easily inferred from colorbar, please provide this information in the caption.
- Please make sure that all colorbars have meaningful ticks. Fig 3b & 10c have no tick for lower bound.
The simulation times (model & represented time) are all given in months. Personally I would find it more intuitive to speak of years, and where the precise number of months is important (e.g. start of spun-up simulations, etc), this information could be given in brackets in months. This remark also applies to time axis of the plots.
I disagree with some statements that are made in the discussion and conclusion section:
The study shows that the accelerated forcing approach is not suitable when the time scale of periodical forcing is in the order of the MCRT. When the forcing period is significantly longer, mean melt rates are higher than in the mean-state (Fig. 6). I am wondering how Fig. 6 would look like if it would be extended to the right, with longer forcing periods. Would it stay constant at comparable levels like the longest tested time scales, or will it converge asymptotically to a higher value? How long would that tail be, and how big the differences? The authors argue in the discussion that for a 30 year forcing period the cavity is assumed to be in equilibrium with the forcing at all times (can this be proven somehow?). But already the 20 year period deviates significantly, which is already a factor 5 higher than the MCRT of 4 years. So then the question arises, what is a minimum factor that is required to still yield realistic results? I understand that it is challenging to test much greater forcing periods due to long computation times and that this might not be feasible. However, I feel that there was little evidence given that 30 years is in equilibrium with the forcing and therefore would be same as 300 years, whereas in the same time it is shown that 20 years is already too short.
That also relates to my second point in the discussion/conclusion: about the application for real-world scenarios. In global warming projections/scenarios the slowly-varying background forcing would not be periodically, but rather steadily increasing at comparable slow rates. When using this acceleration method for coupled simulations, the warming rate would be increased in the accelerated ocean compared to the unaccelerated case. Also in this case it is of great importance to know what would be still acceptable rates of changing forcing without impacting the results too much. Again, here it would help if testing of more than 30 years periodic forcing is possible, as a maximum change rate can be inferred from periodic forcing. However, as this seems not feasible for the given setup/resources, linear increasing forcing from a cold to a warm state at different rates could be an option? Especially for the ice-ocean coupled setup, this would be interesting.
Also important for the applicability of real world scenarios are the time scales of natural variability of ocean to ice forcing, e.g. at decadal timescales (Jenkins et al., 2018). I interpret the current result of the study that the MCRT for Filchner-Ronne/Ross (4-8 years) would conflict with oscillating forcing on decadal time scales. This could possibly impose major challenges for the applicability of the accelerated approach on real world scenarios. I am not certain whether this is a deal-breaker for later applications, but I would like to see much more discussion about this. Stating that the ocean forcing in real world scenarios varies mostly over century-long time scales, seems a bit too simple in my view. Concerning this matter, also a discussion of mixed time scale forcing seems to be interesting, like seasonal forcing overlayed by decadal oscillations with a steady increase in the background signal.
The authors provide a publicly accessible archive (similar as in Zhao et al. 2022, https://doi.org/10.5194/gmd-15-5421-2022) with detailed information about where to obtain the source code of ROMS, ElmerIce and the FISOC coupler (URLs + git commits) including configuration and restart files. As I've never worked with the described models, it is beyond my expertise to judge whether the given information and files are sufficient to reproduce the realized experiments. As far as I can tell the archive does not include information and restart files for the FVCOM model. I request the authors to check this, and update if necessary. Furthermore it would be helpful to include concrete information about the model versions and where to obtain the source code already in the manuscript, e.g. in the code availability section.Also, please make sure, DOIs are provided for all references.
Furthermore, I share the concern by Nicolas Jourdain (RC1) about how to deal with calving fluxes in more realistic/non-local applications.
References:
Jenkins, A., Shoosmith, D., Dutrieux, P., Jacobs, S., Kim, T. W., Lee, S. H., Ha, H. K., and Stammerjohn, S.: West Antarctic Ice Sheet retreat in the Amundsen Sea driven by decadal oceanic variability, Nature Geosci, 11, 733–738, https://doi.org/10.1038/s41561-018-0207-4, 2018.- AC3: 'Reply on RC2', Qin Zhou, 05 Jul 2024
-
RC3: 'Comment on gmd-2023-244', Anonymous Referee #3, 10 Apr 2024
The Authors seek to provide a solution to a current problem limiting the use of coupled ice-ocean models, namely the increased computational expense of the ocean part of the model when compared to the ice side. I find this a very worthwhile and relevant topic suitable for the journal. The justification, methodology and results are well presented. I feel that at present the authors are slightly over selling the potential use of their method without some further additions and clarifications within the discussion section.
1) In the current model framework, ice calving and the resultant freshwater input to the ocean is ignored. Similarly for any models that use real freshwater fluxes on the ocean time step. Do the authors envisage any potential problems with their accelerated forcing scheme if such processes were to be included?
2) Likewise, the calving front is currently fixed in time. if it were to move in time would the approach still hold? My inclination would be that 1) and 2) are not deal breakers, but I would appreciate some discussion about them.
3) In regards to ice dynamics melting near or at the grounding line is of crucial importance to represent accurately. As such it is a potential concern that the differences when using the acceleration factor are located in such areas. It would be good to see further discussion on this topic included.
4) At present domain wide volume changes are shown from the ocean side only. It would be good to see some plots of ice Volume Above Floatation to get an idea of the relative impacts of the acceleration method upon sea level predictions. It would also be good to see some measure of the rate of grounding line retreat over time. Perhaps along a central profile, or a measure of domain grounded area?
If the above points are addressed, as well as those of the other reviewers (with who I find myself in agreement) I am in agreement with), I would be happy to recommend publication.
Minor comments /typos:
L 79 "circulation to flush"
L 85 I think a brief mention of the model domain to be used should be included here to help orientate the reader, as it is a little ambiguous what is meant by the MISOMIP1 framework.
L 175 Is there a reason that different oscillation periods are being used for each set up?
L 216 'Notably, although..'
L 401 'with the exception of the relative...changes exceding 10%.....'Figure 2 Caption - 'curry colored' could be confusing for non native english speakers.
Citation: https://doi.org/10.5194/gmd-2023-244-RC3 - AC1: 'Reply on RC3', Qin Zhou, 04 Jul 2024
- AC4: 'Comment on gmd-2023-244', Qin Zhou, 05 Jul 2024
- AC5: 'Comment on gmd-2023-244', Qin Zhou, 21 Sep 2024
Status: closed
-
RC1: 'Comment on gmd-2023-244', Nicolas Jourdain, 12 Mar 2024
The comment was uploaded in the form of a supplement: https://gmd.copernicus.org/preprints/gmd-2023-244/gmd-2023-244-RC1-supplement.pdf
- AC2: 'Reply on RC1', Qin Zhou, 05 Jul 2024
-
RC2: 'Comment on gmd-2023-244', Moritz Kreuzer, 01 Apr 2024
Summary
The study evaluates the applicability of an accelerated forcing methodology in the scope of high resolution ice-ocean coupling. The authors motivate their investigation by the typical time scale discrepancy between ice and ocean dynamics and the corresponding disparity in simulation time. First, the idea and methodology of an accelerated coupling approach is presented and the models used in the study are described.
Then, standalone setups of two ocean models are used to derive ice-shelf melt rates for warm, cold and mean far-field ocean forcing conditions, as well as oscillating profiles between the warm and cold case at different frequencies. The authors use the mean cavity response time (MCRT; for the mean forcing case) as a characteristic variable to evaluate the derived melt rates. They find that averaged cavity melt rates for oscillating far-field forcings that have significantly higher frequencies than the MCRT (periods <=10% of MCRT) are mostly in the range of 90-110% of melt rates from time averaged forcing. However, forcing frequencies that are in the same order of the MCRT or substantially lower do either under- or overestimate the mean melt rates. Based on these ocean standalone simulations the authors anticipate that the accelerating forcing approach would only work in the first case, which they subsequently test in coupled ice-ocean simulations.
The test scenarios for coupled ice-ocean simulations are structured in three categories: 1. constant cold-to-warm forcing as well as two periodic forcings with fast (2.) and slow (3.) varying time scales. For all categories different acceleration factors are tested (between 1.5 and 10) and evaluated to the baseline scenario (no acceleration) in terms of cavity averaged melt rates and total ocean volume changes (time series) as well as spatially fields of melt rates and ice draft changes. The authors find that acceleration works well for spun-up simulations in the constant forcing as well as the fast-varying ocean forcing case, but not in the slow-varying case.
Finally, the authors conclude, that their presented approach of accelerated forcing in high-resolution ocean-ice coupled models is also applicable in real-world applications for ocean to ice forcing that varies over century-long timescales.General comments
The paper addresses a relevant and scientifically interesting topic which fits very well in the scope of GMD. It presents a novel investigation of testing the impact of asynchronous coupling in idealized setups and introduces a useful metric (MCRT) to assess the applicability of the approach. The study has a sound methodology and follows a clear experimental design with valid, clear and justified assumptions. The manuscript is well written, with fluent and precise language. The title is appropriate and reflects the contents of the paper well. The abstract provides a concise and complete summary of the presented work.
I have no major concerns, but a few remarks. More specific comments and suggestions for improvement are given in the attached pdf.
The introduction motivates the following work well. However, more background on previous work about asynchronous ice-ocean coupling would be great to give the reader more context to the study like: What studies have used asynchronous coupling so far? Is there already some literature that compares synchronous vs asynchronous coupling?
I have a few general remarks for plots (more specific ones are given as annotations to the pdf):
- for all spatial plots of ISOMIP+ domain (barotropic stream function, melt rates, ice draft changes): it would be helpful to either mark the grounding line as a line or to shade the grounded areas (e.g. light gray) to be able to distinguish regions with values close to zero and grounded areas.
- Do 2d plots show values out of the colorbar range (e.g. Fig 3c)? If so, please indicate this by adding out-of-range extensions to the colorbars and give maximum values in text/caption.
- If contour spacing can't be easily inferred from colorbar, please provide this information in the caption.
- Please make sure that all colorbars have meaningful ticks. Fig 3b & 10c have no tick for lower bound.
The simulation times (model & represented time) are all given in months. Personally I would find it more intuitive to speak of years, and where the precise number of months is important (e.g. start of spun-up simulations, etc), this information could be given in brackets in months. This remark also applies to time axis of the plots.
I disagree with some statements that are made in the discussion and conclusion section:
The study shows that the accelerated forcing approach is not suitable when the time scale of periodical forcing is in the order of the MCRT. When the forcing period is significantly longer, mean melt rates are higher than in the mean-state (Fig. 6). I am wondering how Fig. 6 would look like if it would be extended to the right, with longer forcing periods. Would it stay constant at comparable levels like the longest tested time scales, or will it converge asymptotically to a higher value? How long would that tail be, and how big the differences? The authors argue in the discussion that for a 30 year forcing period the cavity is assumed to be in equilibrium with the forcing at all times (can this be proven somehow?). But already the 20 year period deviates significantly, which is already a factor 5 higher than the MCRT of 4 years. So then the question arises, what is a minimum factor that is required to still yield realistic results? I understand that it is challenging to test much greater forcing periods due to long computation times and that this might not be feasible. However, I feel that there was little evidence given that 30 years is in equilibrium with the forcing and therefore would be same as 300 years, whereas in the same time it is shown that 20 years is already too short.
That also relates to my second point in the discussion/conclusion: about the application for real-world scenarios. In global warming projections/scenarios the slowly-varying background forcing would not be periodically, but rather steadily increasing at comparable slow rates. When using this acceleration method for coupled simulations, the warming rate would be increased in the accelerated ocean compared to the unaccelerated case. Also in this case it is of great importance to know what would be still acceptable rates of changing forcing without impacting the results too much. Again, here it would help if testing of more than 30 years periodic forcing is possible, as a maximum change rate can be inferred from periodic forcing. However, as this seems not feasible for the given setup/resources, linear increasing forcing from a cold to a warm state at different rates could be an option? Especially for the ice-ocean coupled setup, this would be interesting.
Also important for the applicability of real world scenarios are the time scales of natural variability of ocean to ice forcing, e.g. at decadal timescales (Jenkins et al., 2018). I interpret the current result of the study that the MCRT for Filchner-Ronne/Ross (4-8 years) would conflict with oscillating forcing on decadal time scales. This could possibly impose major challenges for the applicability of the accelerated approach on real world scenarios. I am not certain whether this is a deal-breaker for later applications, but I would like to see much more discussion about this. Stating that the ocean forcing in real world scenarios varies mostly over century-long time scales, seems a bit too simple in my view. Concerning this matter, also a discussion of mixed time scale forcing seems to be interesting, like seasonal forcing overlayed by decadal oscillations with a steady increase in the background signal.
The authors provide a publicly accessible archive (similar as in Zhao et al. 2022, https://doi.org/10.5194/gmd-15-5421-2022) with detailed information about where to obtain the source code of ROMS, ElmerIce and the FISOC coupler (URLs + git commits) including configuration and restart files. As I've never worked with the described models, it is beyond my expertise to judge whether the given information and files are sufficient to reproduce the realized experiments. As far as I can tell the archive does not include information and restart files for the FVCOM model. I request the authors to check this, and update if necessary. Furthermore it would be helpful to include concrete information about the model versions and where to obtain the source code already in the manuscript, e.g. in the code availability section.Also, please make sure, DOIs are provided for all references.
Furthermore, I share the concern by Nicolas Jourdain (RC1) about how to deal with calving fluxes in more realistic/non-local applications.
References:
Jenkins, A., Shoosmith, D., Dutrieux, P., Jacobs, S., Kim, T. W., Lee, S. H., Ha, H. K., and Stammerjohn, S.: West Antarctic Ice Sheet retreat in the Amundsen Sea driven by decadal oceanic variability, Nature Geosci, 11, 733–738, https://doi.org/10.1038/s41561-018-0207-4, 2018.- AC3: 'Reply on RC2', Qin Zhou, 05 Jul 2024
-
RC3: 'Comment on gmd-2023-244', Anonymous Referee #3, 10 Apr 2024
The Authors seek to provide a solution to a current problem limiting the use of coupled ice-ocean models, namely the increased computational expense of the ocean part of the model when compared to the ice side. I find this a very worthwhile and relevant topic suitable for the journal. The justification, methodology and results are well presented. I feel that at present the authors are slightly over selling the potential use of their method without some further additions and clarifications within the discussion section.
1) In the current model framework, ice calving and the resultant freshwater input to the ocean is ignored. Similarly for any models that use real freshwater fluxes on the ocean time step. Do the authors envisage any potential problems with their accelerated forcing scheme if such processes were to be included?
2) Likewise, the calving front is currently fixed in time. if it were to move in time would the approach still hold? My inclination would be that 1) and 2) are not deal breakers, but I would appreciate some discussion about them.
3) In regards to ice dynamics melting near or at the grounding line is of crucial importance to represent accurately. As such it is a potential concern that the differences when using the acceleration factor are located in such areas. It would be good to see further discussion on this topic included.
4) At present domain wide volume changes are shown from the ocean side only. It would be good to see some plots of ice Volume Above Floatation to get an idea of the relative impacts of the acceleration method upon sea level predictions. It would also be good to see some measure of the rate of grounding line retreat over time. Perhaps along a central profile, or a measure of domain grounded area?
If the above points are addressed, as well as those of the other reviewers (with who I find myself in agreement) I am in agreement with), I would be happy to recommend publication.
Minor comments /typos:
L 79 "circulation to flush"
L 85 I think a brief mention of the model domain to be used should be included here to help orientate the reader, as it is a little ambiguous what is meant by the MISOMIP1 framework.
L 175 Is there a reason that different oscillation periods are being used for each set up?
L 216 'Notably, although..'
L 401 'with the exception of the relative...changes exceding 10%.....'Figure 2 Caption - 'curry colored' could be confusing for non native english speakers.
Citation: https://doi.org/10.5194/gmd-2023-244-RC3 - AC1: 'Reply on RC3', Qin Zhou, 04 Jul 2024
- AC4: 'Comment on gmd-2023-244', Qin Zhou, 05 Jul 2024
- AC5: 'Comment on gmd-2023-244', Qin Zhou, 21 Sep 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 502 | 128 | 47 | 677 | 45 | 42 |
- HTML: 502
- PDF: 128
- XML: 47
- Total: 677
- BibTeX: 45
- EndNote: 42
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1