the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Sep 2023
Consistent but more intense atmospheric circulation response to Arctic sea ice loss in CMIP6 experiments compared to PAMIP experiments
Abstract. The atmospheric circulation response to Arctic sea ice loss may differ depending on the region of sea ice loss but also on the methodology used to study this impact. Examining the different possible atmospheric circulation responses to sea ice loss is essential, as the Arctic sea ice is not melting uniformly. In this study, we examine the atmospheric response in winter to regional sea ice loss using two different approaches across seven climate models. The sea ice anomaly areas are the pan-Arctic, the Barents-Kara Seas only, and the Sea of Okhotsk only. The first approach involves sensitivity experiments performed within the Polar Amplification Model Intercomparison Project (PAMIP), while the second approach entails a composite analysis in long pre-industrial control simulations from CMIP6. Our results reveal that both approaches lead to consistent atmospheric circulation responses to pan-Arctic sea ice loss, characterized by a negative phase in the North Atlantic Oscillation and a weakening of the stratospheric polar vortex. Similar responses to BK sea ice loss are simulated, albeit with more spread in the PAMIP experiments. The responses to Okhotsk sea ice loss differ and are uncertain in both approaches. Furthermore, larger changes are detected in the composite analysis than in the sensitivity experiments, likely due to a different background state and the presence of confounding factors in the composite analysis. We also find that the atmosphere-ocean coupling does not imply larger circulation changes or a better representation of the eddy momentum feedback in the climate response. These results highlight that sea ice loss in sensitivity experiments yields a weaker atmospheric circulation response compared to the pre-industrial simulations in CMIP6 where the sea ice loss is governed by internal climate variability. A quantification of the role played by factors related to sea ice loss that amplifies the response should be further investigated.
- Preprint
(7371 KB) - Metadata XML
- BibTeX
- EndNote
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1748', Anonymous Referee #1, 04 Oct 2023
GENERAL COMMENTS
This works compares the results of two methods used for understanding the impacts of Arctic and regional sea-ice reduction on the atmospheric circulation, specifically on the NAO and on the stratospheric polar vortex (SPV) strength. The authors find a correspondence between Arctic/BK sea-ice depletion and negative NAO which is consistent across the different approaches, yet showing un-explicable differences in the intensity and geographic location of the surface signal producing negative NAO. The results for Okhotsk-Sea ice depletion are inconsistent across models and methodology.
Although no univocal answer to the primary research question is reached (clearly acknowledged in the manuscript), the results are interesting and pose a new question, i.e. if there exists a correct model setting and methodological approach that should be employed to investigate the atmospheric response to sea-ice reduction. In this direction, a few more tests could be done to verify the CMIP6 compositing method. Moreover, the atmospheric influence on SST and sea-ice anomalies in such simulations should be discussed in more detail, in order to better describe the differences with the setting in which the PAMIP experiments are run.
The general topic of the manuscript matches perfectly with the scope of ESD. The paper is well structured and the arguments are explained clearly and with sufficient completness and put in relation with previous literature - although one analysis appears to be incorrect with respect to the conclusions drawn from it. Under the condition that the specific comments below are addressed satisfyingly by the authors, I would be willing to advise the publication of this manuscript in ESD.
SPECIFIC COMMENTS
1. The CMIP6 composite analysis is not fully convincing to me, as it leaves space for the atmospheric circulation to affect the sea-ice anomalies and SSTs consistently, i.e. for a positive feedback between the three components rather than sea-ice forcing only (and this could be an explanation for the much stronger atmospheric anomalies in the CMIP6 composite compared to the PAMIP sensitivity experiments).
A few analysis would help assessing what is happening in the CMIP6 composite:
- What are the results with no low-pass filtering in the selection of years with low sea-ice extent (described in line 126)? How sensitive are the composites to the month when the sea-ice anomalies are evaluated? What if Arc and BK were constructed based on the November sea-ice anomalies?
- Does a composite selection of years based on a weak SPV or negative NAO in Autumn/early winter recover the composites’ monthly evolution of the sea-ice extension of Figure 1c?
- Thinking about a possible slow-varying forcing by the atmosphere in the CMIP6 composite, does the decadal variability in the NAO relate to the selection of sea-ice composite years? How does the NAO variability compare between the CMIP6 simulations and the PAMIP experiments?
- How do the fields of SST anomalies look in the coupled PAMIP and in the CMIP6 composite ?
- What happens if you apply the same compositing approach to ERA5? Are the results similar to those obtained from CMIP6 compositing?
2. In Figure 10 you relate the SPV response to the strength of the climatological SPV. What happens if you do the same analysis but comparing tropospheric jet latitude (climatology and response) and strength (climatology and response), considering zonal means of zonal winds over all latitudes or zonal means only over the Atlantic sector.
3. As explained in Methods at the end of Smith et al. 2022, the eddy feedback should be a measure of the efficiency in driving and accelerating the zonal winds. Eddies are disturbances that can only be measured on a synoptic (fast) time scale, hence computing the eddy feedback from monthly data does not estimate the synoptic eddy feedback on the meanflow, but maybe the feedback by slower disturbances. Hence, this analysis is not meaningfull in the sence described by Smith et al. If not otherwise justified, I would recommend removing the analysis from the results and conclusions.
4. In the title the word "response" doesn't reflect the meaning of the composite analysis, where we usually check the state of the atmosphere corresponding to the anomalies in some particular variable, not the actual response to the variable forcing (as you mention in the text other confounding factors are present). Within the text you could also mention more clearly the fact that in a coupled GCM the atmospheric circulation affects the sea ice variability on multiple time-scales.
TECHNICAL CORRECTIONS
Figure 1: Over what months are the sea-ice anomalies in Figures 1,3,5 shown? I would like to see it in the month preceeding the atmospheric anomalies.
Figures 2,4,6,7: The letters (a), (b), (c) are very low and seem to refer to the panels in the lower row. To avoid confusion please place the labels on top of each panel, or else shift closer to the respective panel.
line 2: "to diverse/regional patterns of sea ice loss"
line 9: "albeit showing more spread than"
line 15-16: The last sentence is not clear, please elaborate.
line 20: the trends of the last couple of years could be mentioned.
line 29: "answer" instead of "study".
line 35/36: mention also the reversed influence of circulation on sea-ice anomalies.
line 38: "to complement observational studies".
line 39: "to improve the statistical significance in the.."
line 41: "drivers of atmospheric variability"?
line 55/56: repetition of "decline".
line 58: add reference to Ruggieri, P., Kucharski, F., Buizza, R. and Ambaum, M.H.P. (2017), The transient atmospheric response to a reduction of sea-ice cover in the Barents and Kara Seas. Q.J.R. Meteorol. Soc., 143: 1632-1640. https://doi.org/10.1002/qj.3034
line 68: "each model was run over a .."
line 70: Specify "These experiments, sharing a common methodological approach,.."
line 80: Add ".. or snow cover or atmospheric forcing on the surface state"
line 81/82 and elsewhere: The term "response" is not correct for the compositing, as in specific comment 4.
line 88/93: already specify the numbers of the simulations (1.1, etc)
line 103-105: in experiments 3.1 and 3.2 is the rest of the SIC constrained to the climatology or freely evolving?
line 119: Could you express the percentage in no. models / tot. no. models?
line 120: ".. from the atmospheric state with present-day sea-ice conditions."
line 121: I am not able to distinguish triangles from crosses in the plots. Could you make the difference between the two clearer?
line 128: Could you compute the superposition in the years selected for Arctic, Okhotsk Sea and and Barents Kara composites?
line 172: This is not an underestimation with respect to Smith et al. 2022, but rather a different quantity, since the temporal frequency of the data is different. No point in comparing with Smith et al. 2022. But what does this quantity represent? Can we still call it an eddy feedback, since the (synoptic) eddy temporal variability is smoothed out?
line 179: "four times larger" in amplitude or spatial extension?
line 193: I see only 2 models showing significant NAO in both PAMIP and CMIP6 composites (not considering the model mean)
line 198: zonal wind weakening peaks in the stratosphere only in CMIP6, not in the PAMIP case, right? You could mention that the most significant decrease seems to be limited to the tropospheric levels.
line 271: estimate superposition (no. common years / tot. no. selected years for each composite?)
line 282-283: the sentence is unclear and too general (e.g. what does inherent differences mean), please re-formulate.
par. 308-316 and par. 401-413 and conclusions: Apart from coupling and the other factors considered here, the two methods differ in allowing natural variability over long time scales (CMIP6), or running from differing initial conditions over only 14 month (PAMIP). This is not cancelled out simply by filtering out the interdecadal changes in sea-ice fraction. Moreover, the SSTs in low sea-ice CMIP6 years evolve consistently and together with the sea-ice field under ocean-atmosphere feedback mechanisms, which is not the case in nudged sea-ice experiments, where SSTs may experience atmospheric influence producing anomalies that diverge from the nudged sea-ice ones. These, and other possible relevant differences, should be mentioned when stating that the divergent results are not related to coupling.
line 321-324 and conclusions: mention that the parameters were computed differently, with daily rather than monthly fields!
line 336: repetition of previous sentence?
line 348: and with daily fields!
Citation: https://doi.org/10.5194/egusphere-2023-1748-RC1 -
RC2: 'Comment on egusphere-2023-1748', Anonymous Referee #2, 25 Oct 2023
The manuscript provides a comparison of two methods that attempt to estimate the atmospheric response to sea-ice loss. The first method employs ‘intervention’ experiments, where sea ice perturbations are specified in model simulations. The second approach relies on composites during low sea ice years simulated in free-running model experiments The key result is there is some similarity in the response estimates from the two methods, but also considerable differences – most notably, the estimated responses from the composite analysis (hereafter referred to as CMIP6) are generally of greater magnitude than from the perturbation experiments (hereafter referred to as PAMIP). The manuscript discusses several potential causes of this difference but is unable to make any firm conclusions.
I think the aim of the paper is sensible: it is a valuable endeavour to compare the two methods and the results have important implications. I don’t perceive any major errors in the analyses, but I think their interpretation needs additional thought. I recommend a major revision to improve the clarity of the paper and its messages. I list of number of points below, roughly in order of where they appear in the manuscript, rather than their severity.
- Title and abstract (and throughout). A more precise use of wording is needed. I think it is more accurate to say that the two approaches give different estimates of the response to sea-ice loss, not that the responses are different.
- Introduction: Whilst true that the composite approach can capture other factors associated with sea ice loss, it may also capture factors that co-occur with sea-ice loss but are not physically related to sea-ice loss. One benefit of the composite approach that isn’t explicitly mentioned is that it can be applied to observations/reanalyses to provide an observational estimate of the response.
- The comparison between CMIP6 and PAMIP is hindered by the very different patterns but especially magnitudes of sea-ice loss. The projected loss of Arctic sea ice prescribed in the PAMIP experiments is much larger than the interannual variability of sea ice in the PI control experiments. I wonder if a closer match could be obtained by comparing the pre-industrial and present-day simulations from PAMIP, rather than the present-day and future experiments. An alternative would be to use a pattern scaling approach and normalize the estimated responses per unit loss of sea ice. This would have the effect of reducing the magnitude of the response estimated in PAMIP relative to CMIP6. I find it a bit misleading to say the “atmospheric response is consistent across both approaches” (line 205). If the difference in the magnitudes of sea-ice loss is accounted for, the response estimates would look very different in magnitude, if not in pattern.
- A major difference I note between PAMIP and CMIP6 from Figure 2 is that the CMIP6 estimate lacks the strong low pressure anomalies in the regions of winter ice loss. It would be useful to see maps of the winter sea ice anomalies to know whether this difference comes from differing patterns of sea ice loss or whether the methods are detecting different ‘responses’ to a similar forcing.
- Related to (3), although the BKS sea-ice loss looks similar in October between PAMIP and CMIP6, PAMIP has larger change through winter. Generally, whilst it appears feasible to capture a monthly anomaly of similar magnitude in CMIP6 than PAMIP, the CMIP6 anomalies doesn’t persist for many months, whereas in PAMIP (representing a long-term trend rather than internal variability) has sea-ice loss in all months. Again, if the response estimates were scaled by the magnitudes of winter sea-ice loss, the apparent discrepancies between PAMIP and CMIP6 would be even greater.
- To my eye, there is little consistency in either the spatial pattern or magnitude of the response estimates to BKS sea-ice loss. I think the interpretation that “Similar responses to BK sea ice loss are simulated” (line 9) is questionable. I would say they are inconsistent. It might be interesting to calculate the pattern correlation between the maps, in figure 4 and elsewhere to quantity the level of agreement (or not).
- For Okhotsk sea ice loss, there is a reasonable match in December between PAMIP and CMIP6, but PAMIP has larger losses in Jan-March.
- On the effect of coupling – the papers cited as motivation for this exploration of the effect of coupling which suggest a stronger response to sea-ice loss in coupled models refer to the equilibrium response and the importance of slow ocean adjustments. At least based on these papers, coupling would not be expected to amplify the fast response to interannual variations in sea ice. So, I think this section is a bit misguided.
- On the eddy feedback – the methods imply that the eddy feedback parameter was calculated from monthly mean data. Is that correct? If so, this is not the same as Smith et al, which used higher frequency (daily or sub-daily) input data. Eddies are short-lived phenomena. It is unclear (and unjustified) that the calculating the eddy feedback parameter from monthly data is suitable. This is a major caveat and at worst, invalidates any conclusions from this part of the analysis.
- On the QBO – the authors have misrepresented the work by Labe et al. The issue is not whether sea-ice loss changes the QBO, but whether the QBO (through its effect of the background state) modulates the response to sea-ice loss. Therefore, the pertinent thing to check is the QBO mean state in PAMIP and CMIP6. The weaker SPV mean state could be related to more easterly QBO in CMIP6 relative to PAMIP. Labe et al suggests a stronger SPV response to sea-ice loss during QBO-E than QBO-W.
- I think there is a possible fourth explanation for the differences between CMIP6 and PAMIP that is not well addressed, although it is related to coupling and confounding factors. Composites will reveal relationships in two directions – both the atmospheric response to sea ice and sea ice response to atmospheric (circulation) changes. For example, BKS sea-ice loss is associated with the negative NAO. This might partly reflect a change in the NAO in response to BKS sea-ice loss but could also reflect a reduction in BKS sea ice driven by the negative NAO. The latter would not be captured by the PAMIP simulations, whether uncoupled or coupled (the coupled PAMIP simulations aim to capture the effect of coupling in the response to sea-ice loss, but they do not capture atmospheric forcing of sea ice, because the sea ice is constrained by design). The several month lag in the composite analysis may not fully circumvent this issue, as discussed in Blackport and Screen 2021. Smith et al 2017 also raised this issue.
- A fifth explanation, as mentioned above, is the sea ice loss is not consistent between approaches, with differing patterns, magnitudes and seasonality.
- Conclusions: The authors are unable to make firm conclusions about the causes of the difference between methods, which is unfortunate but understandable. A key conclusion I take from this paper, but that isn’t mentioned, is that comparisons between models and observations need to be interpreted with extreme caution. Often observed correlations or composites are compared to model perturbation experiments, and their differences have been interpreted as weaknesses in models (e.g, Mori et al., 2019; Cohen et al 2020). However, this submission clearly shows significant differences when applying two methods to the same models. Thus, any apparent differences between observations and models could be attributed to different methods rather than model biases. It might be worth discussing your results in the context of the arguments presented by Blackport and Screen 2021.
- It is open to interpretation whether the composite approach is suitable for quantifying the response to sea ice, given all the confounding factors discussed. In my opinion, the striking differences between PAMIP and CMIP6 reveal that the composite approach is not suitable. I’m interested to know if the authors agree, or not? If not, what evidence is there that the composite approach provides a better estimate of the response to sea-ice loss than the perturbation experiment approach? The author may prefer not to and if so, I would respect that wish, but I think the discussion and conclusions would benefit from some reflection of relative strengths and weaknesses of the two approaches. Some of the potential weaknesses of the PAMIP approach – lack of coupling, weak eddy feedbacks – appear to have been ruled out, or at least, deemed not to be primary causes of the differences between PAMIP and CMIP6. The explanations remaining are dependence on the mean state and confounding factors and the latter, seems to point to major weaknesses of the composite approach.
Citation: https://doi.org/10.5194/egusphere-2023-1748-RC2
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1748', Anonymous Referee #1, 04 Oct 2023
GENERAL COMMENTS
This works compares the results of two methods used for understanding the impacts of Arctic and regional sea-ice reduction on the atmospheric circulation, specifically on the NAO and on the stratospheric polar vortex (SPV) strength. The authors find a correspondence between Arctic/BK sea-ice depletion and negative NAO which is consistent across the different approaches, yet showing un-explicable differences in the intensity and geographic location of the surface signal producing negative NAO. The results for Okhotsk-Sea ice depletion are inconsistent across models and methodology.
Although no univocal answer to the primary research question is reached (clearly acknowledged in the manuscript), the results are interesting and pose a new question, i.e. if there exists a correct model setting and methodological approach that should be employed to investigate the atmospheric response to sea-ice reduction. In this direction, a few more tests could be done to verify the CMIP6 compositing method. Moreover, the atmospheric influence on SST and sea-ice anomalies in such simulations should be discussed in more detail, in order to better describe the differences with the setting in which the PAMIP experiments are run.
The general topic of the manuscript matches perfectly with the scope of ESD. The paper is well structured and the arguments are explained clearly and with sufficient completness and put in relation with previous literature - although one analysis appears to be incorrect with respect to the conclusions drawn from it. Under the condition that the specific comments below are addressed satisfyingly by the authors, I would be willing to advise the publication of this manuscript in ESD.
SPECIFIC COMMENTS
1. The CMIP6 composite analysis is not fully convincing to me, as it leaves space for the atmospheric circulation to affect the sea-ice anomalies and SSTs consistently, i.e. for a positive feedback between the three components rather than sea-ice forcing only (and this could be an explanation for the much stronger atmospheric anomalies in the CMIP6 composite compared to the PAMIP sensitivity experiments).
A few analysis would help assessing what is happening in the CMIP6 composite:
- What are the results with no low-pass filtering in the selection of years with low sea-ice extent (described in line 126)? How sensitive are the composites to the month when the sea-ice anomalies are evaluated? What if Arc and BK were constructed based on the November sea-ice anomalies?
- Does a composite selection of years based on a weak SPV or negative NAO in Autumn/early winter recover the composites’ monthly evolution of the sea-ice extension of Figure 1c?
- Thinking about a possible slow-varying forcing by the atmosphere in the CMIP6 composite, does the decadal variability in the NAO relate to the selection of sea-ice composite years? How does the NAO variability compare between the CMIP6 simulations and the PAMIP experiments?
- How do the fields of SST anomalies look in the coupled PAMIP and in the CMIP6 composite ?
- What happens if you apply the same compositing approach to ERA5? Are the results similar to those obtained from CMIP6 compositing?
2. In Figure 10 you relate the SPV response to the strength of the climatological SPV. What happens if you do the same analysis but comparing tropospheric jet latitude (climatology and response) and strength (climatology and response), considering zonal means of zonal winds over all latitudes or zonal means only over the Atlantic sector.
3. As explained in Methods at the end of Smith et al. 2022, the eddy feedback should be a measure of the efficiency in driving and accelerating the zonal winds. Eddies are disturbances that can only be measured on a synoptic (fast) time scale, hence computing the eddy feedback from monthly data does not estimate the synoptic eddy feedback on the meanflow, but maybe the feedback by slower disturbances. Hence, this analysis is not meaningfull in the sence described by Smith et al. If not otherwise justified, I would recommend removing the analysis from the results and conclusions.
4. In the title the word "response" doesn't reflect the meaning of the composite analysis, where we usually check the state of the atmosphere corresponding to the anomalies in some particular variable, not the actual response to the variable forcing (as you mention in the text other confounding factors are present). Within the text you could also mention more clearly the fact that in a coupled GCM the atmospheric circulation affects the sea ice variability on multiple time-scales.
TECHNICAL CORRECTIONS
Figure 1: Over what months are the sea-ice anomalies in Figures 1,3,5 shown? I would like to see it in the month preceeding the atmospheric anomalies.
Figures 2,4,6,7: The letters (a), (b), (c) are very low and seem to refer to the panels in the lower row. To avoid confusion please place the labels on top of each panel, or else shift closer to the respective panel.
line 2: "to diverse/regional patterns of sea ice loss"
line 9: "albeit showing more spread than"
line 15-16: The last sentence is not clear, please elaborate.
line 20: the trends of the last couple of years could be mentioned.
line 29: "answer" instead of "study".
line 35/36: mention also the reversed influence of circulation on sea-ice anomalies.
line 38: "to complement observational studies".
line 39: "to improve the statistical significance in the.."
line 41: "drivers of atmospheric variability"?
line 55/56: repetition of "decline".
line 58: add reference to Ruggieri, P., Kucharski, F., Buizza, R. and Ambaum, M.H.P. (2017), The transient atmospheric response to a reduction of sea-ice cover in the Barents and Kara Seas. Q.J.R. Meteorol. Soc., 143: 1632-1640. https://doi.org/10.1002/qj.3034
line 68: "each model was run over a .."
line 70: Specify "These experiments, sharing a common methodological approach,.."
line 80: Add ".. or snow cover or atmospheric forcing on the surface state"
line 81/82 and elsewhere: The term "response" is not correct for the compositing, as in specific comment 4.
line 88/93: already specify the numbers of the simulations (1.1, etc)
line 103-105: in experiments 3.1 and 3.2 is the rest of the SIC constrained to the climatology or freely evolving?
line 119: Could you express the percentage in no. models / tot. no. models?
line 120: ".. from the atmospheric state with present-day sea-ice conditions."
line 121: I am not able to distinguish triangles from crosses in the plots. Could you make the difference between the two clearer?
line 128: Could you compute the superposition in the years selected for Arctic, Okhotsk Sea and and Barents Kara composites?
line 172: This is not an underestimation with respect to Smith et al. 2022, but rather a different quantity, since the temporal frequency of the data is different. No point in comparing with Smith et al. 2022. But what does this quantity represent? Can we still call it an eddy feedback, since the (synoptic) eddy temporal variability is smoothed out?
line 179: "four times larger" in amplitude or spatial extension?
line 193: I see only 2 models showing significant NAO in both PAMIP and CMIP6 composites (not considering the model mean)
line 198: zonal wind weakening peaks in the stratosphere only in CMIP6, not in the PAMIP case, right? You could mention that the most significant decrease seems to be limited to the tropospheric levels.
line 271: estimate superposition (no. common years / tot. no. selected years for each composite?)
line 282-283: the sentence is unclear and too general (e.g. what does inherent differences mean), please re-formulate.
par. 308-316 and par. 401-413 and conclusions: Apart from coupling and the other factors considered here, the two methods differ in allowing natural variability over long time scales (CMIP6), or running from differing initial conditions over only 14 month (PAMIP). This is not cancelled out simply by filtering out the interdecadal changes in sea-ice fraction. Moreover, the SSTs in low sea-ice CMIP6 years evolve consistently and together with the sea-ice field under ocean-atmosphere feedback mechanisms, which is not the case in nudged sea-ice experiments, where SSTs may experience atmospheric influence producing anomalies that diverge from the nudged sea-ice ones. These, and other possible relevant differences, should be mentioned when stating that the divergent results are not related to coupling.
line 321-324 and conclusions: mention that the parameters were computed differently, with daily rather than monthly fields!
line 336: repetition of previous sentence?
line 348: and with daily fields!
Citation: https://doi.org/10.5194/egusphere-2023-1748-RC1 -
RC2: 'Comment on egusphere-2023-1748', Anonymous Referee #2, 25 Oct 2023
The manuscript provides a comparison of two methods that attempt to estimate the atmospheric response to sea-ice loss. The first method employs ‘intervention’ experiments, where sea ice perturbations are specified in model simulations. The second approach relies on composites during low sea ice years simulated in free-running model experiments The key result is there is some similarity in the response estimates from the two methods, but also considerable differences – most notably, the estimated responses from the composite analysis (hereafter referred to as CMIP6) are generally of greater magnitude than from the perturbation experiments (hereafter referred to as PAMIP). The manuscript discusses several potential causes of this difference but is unable to make any firm conclusions.
I think the aim of the paper is sensible: it is a valuable endeavour to compare the two methods and the results have important implications. I don’t perceive any major errors in the analyses, but I think their interpretation needs additional thought. I recommend a major revision to improve the clarity of the paper and its messages. I list of number of points below, roughly in order of where they appear in the manuscript, rather than their severity.
- Title and abstract (and throughout). A more precise use of wording is needed. I think it is more accurate to say that the two approaches give different estimates of the response to sea-ice loss, not that the responses are different.
- Introduction: Whilst true that the composite approach can capture other factors associated with sea ice loss, it may also capture factors that co-occur with sea-ice loss but are not physically related to sea-ice loss. One benefit of the composite approach that isn’t explicitly mentioned is that it can be applied to observations/reanalyses to provide an observational estimate of the response.
- The comparison between CMIP6 and PAMIP is hindered by the very different patterns but especially magnitudes of sea-ice loss. The projected loss of Arctic sea ice prescribed in the PAMIP experiments is much larger than the interannual variability of sea ice in the PI control experiments. I wonder if a closer match could be obtained by comparing the pre-industrial and present-day simulations from PAMIP, rather than the present-day and future experiments. An alternative would be to use a pattern scaling approach and normalize the estimated responses per unit loss of sea ice. This would have the effect of reducing the magnitude of the response estimated in PAMIP relative to CMIP6. I find it a bit misleading to say the “atmospheric response is consistent across both approaches” (line 205). If the difference in the magnitudes of sea-ice loss is accounted for, the response estimates would look very different in magnitude, if not in pattern.
- A major difference I note between PAMIP and CMIP6 from Figure 2 is that the CMIP6 estimate lacks the strong low pressure anomalies in the regions of winter ice loss. It would be useful to see maps of the winter sea ice anomalies to know whether this difference comes from differing patterns of sea ice loss or whether the methods are detecting different ‘responses’ to a similar forcing.
- Related to (3), although the BKS sea-ice loss looks similar in October between PAMIP and CMIP6, PAMIP has larger change through winter. Generally, whilst it appears feasible to capture a monthly anomaly of similar magnitude in CMIP6 than PAMIP, the CMIP6 anomalies doesn’t persist for many months, whereas in PAMIP (representing a long-term trend rather than internal variability) has sea-ice loss in all months. Again, if the response estimates were scaled by the magnitudes of winter sea-ice loss, the apparent discrepancies between PAMIP and CMIP6 would be even greater.
- To my eye, there is little consistency in either the spatial pattern or magnitude of the response estimates to BKS sea-ice loss. I think the interpretation that “Similar responses to BK sea ice loss are simulated” (line 9) is questionable. I would say they are inconsistent. It might be interesting to calculate the pattern correlation between the maps, in figure 4 and elsewhere to quantity the level of agreement (or not).
- For Okhotsk sea ice loss, there is a reasonable match in December between PAMIP and CMIP6, but PAMIP has larger losses in Jan-March.
- On the effect of coupling – the papers cited as motivation for this exploration of the effect of coupling which suggest a stronger response to sea-ice loss in coupled models refer to the equilibrium response and the importance of slow ocean adjustments. At least based on these papers, coupling would not be expected to amplify the fast response to interannual variations in sea ice. So, I think this section is a bit misguided.
- On the eddy feedback – the methods imply that the eddy feedback parameter was calculated from monthly mean data. Is that correct? If so, this is not the same as Smith et al, which used higher frequency (daily or sub-daily) input data. Eddies are short-lived phenomena. It is unclear (and unjustified) that the calculating the eddy feedback parameter from monthly data is suitable. This is a major caveat and at worst, invalidates any conclusions from this part of the analysis.
- On the QBO – the authors have misrepresented the work by Labe et al. The issue is not whether sea-ice loss changes the QBO, but whether the QBO (through its effect of the background state) modulates the response to sea-ice loss. Therefore, the pertinent thing to check is the QBO mean state in PAMIP and CMIP6. The weaker SPV mean state could be related to more easterly QBO in CMIP6 relative to PAMIP. Labe et al suggests a stronger SPV response to sea-ice loss during QBO-E than QBO-W.
- I think there is a possible fourth explanation for the differences between CMIP6 and PAMIP that is not well addressed, although it is related to coupling and confounding factors. Composites will reveal relationships in two directions – both the atmospheric response to sea ice and sea ice response to atmospheric (circulation) changes. For example, BKS sea-ice loss is associated with the negative NAO. This might partly reflect a change in the NAO in response to BKS sea-ice loss but could also reflect a reduction in BKS sea ice driven by the negative NAO. The latter would not be captured by the PAMIP simulations, whether uncoupled or coupled (the coupled PAMIP simulations aim to capture the effect of coupling in the response to sea-ice loss, but they do not capture atmospheric forcing of sea ice, because the sea ice is constrained by design). The several month lag in the composite analysis may not fully circumvent this issue, as discussed in Blackport and Screen 2021. Smith et al 2017 also raised this issue.
- A fifth explanation, as mentioned above, is the sea ice loss is not consistent between approaches, with differing patterns, magnitudes and seasonality.
- Conclusions: The authors are unable to make firm conclusions about the causes of the difference between methods, which is unfortunate but understandable. A key conclusion I take from this paper, but that isn’t mentioned, is that comparisons between models and observations need to be interpreted with extreme caution. Often observed correlations or composites are compared to model perturbation experiments, and their differences have been interpreted as weaknesses in models (e.g, Mori et al., 2019; Cohen et al 2020). However, this submission clearly shows significant differences when applying two methods to the same models. Thus, any apparent differences between observations and models could be attributed to different methods rather than model biases. It might be worth discussing your results in the context of the arguments presented by Blackport and Screen 2021.
- It is open to interpretation whether the composite approach is suitable for quantifying the response to sea ice, given all the confounding factors discussed. In my opinion, the striking differences between PAMIP and CMIP6 reveal that the composite approach is not suitable. I’m interested to know if the authors agree, or not? If not, what evidence is there that the composite approach provides a better estimate of the response to sea-ice loss than the perturbation experiment approach? The author may prefer not to and if so, I would respect that wish, but I think the discussion and conclusions would benefit from some reflection of relative strengths and weaknesses of the two approaches. Some of the potential weaknesses of the PAMIP approach – lack of coupling, weak eddy feedbacks – appear to have been ruled out, or at least, deemed not to be primary causes of the differences between PAMIP and CMIP6. The explanations remaining are dependence on the mean state and confounding factors and the latter, seems to point to major weaknesses of the composite approach.
Citation: https://doi.org/10.5194/egusphere-2023-1748-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 479 | 167 | 36 | 682 | 36 | 41 |
- HTML: 479
- PDF: 167
- XML: 36
- Total: 682
- BibTeX: 36
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 180 | 27 |
| Germany | 2 | 130 | 19 |
| China | 3 | 46 | 6 |
| France | 4 | 41 | 6 |
| United Kingdom | 5 | 37 | 5 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 180