the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Dec 2023
Storylines of Summer Arctic climate change constrained by Barents-Kara Sea and Arctic tropospheric warming for climate risks assessment
Abstract. While climate models broadly agree on the changes expected to occur over the Arctic with global warming on a pan-Arctic scale (i.e., polar amplification, sea-ice loss, increased precipitation), the magnitude and patterns of those changes at regional and local scales remain uncertain. This limits the usability of climate model projections for risk assessments and their impact on human activities or ecosystems (e.g., fires, permafrost thawing). Whereas any single or ensemble-mean projection may be of limited use to stakeholders, recent studies have shown the value of the storyline approach in providing a comprehensive and tractable set of climate projections that can be used to evaluate changes in environmental or societal risks associated with global warming.
Here, we apply the storyline approach to a large ensemble of CMIP6 models, with the aim of distilling the wide spread in model predictions into four physically plausible outcomes of Arctic summertime climate change. This is made possible by leveraging strong covariability in the climate system, associated with well-known but poorly constrained teleconnections and local processes: specifically, we find that differences in Barents-Kara Sea warming and lower tropospheric warming over polar land regions among CMIP6 models explain most of the inter-model variability in pan-Arctic surface summer climate response to global warming. Based on this novel finding, we compare regional disparities in climate change across the four storylines. Our storyline analysis highlights the fact that, for a given amount of global warming, certain climate risks can be intensified while others may be lessened, relative to a “middle-of-the-road” ensemble mean projection. We find this to be particularly relevant when comparing climate change over terrestrial and marine areas of the Arctic, which can show substantial differences in their sensitivity to global warming. We conclude by discussing potential implications of our findings for modelling climate change impacts on ecosystems and human activities.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(4982 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(4982 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2741', Anonymous Referee #1, 19 Dec 2023
This paper applies the storyline approach developed by Zappa & Shepherd (2017) and Mindlin et al. (2020), which is based on multiple linear regression across CMIP models, to represent the uncertainty in the response of summertime Arctic climate to greenhouse warming in terms of two key drivers: BK SST warming, and pan-Arctic lower tropospheric warming. So far as I am aware this is the first time that the storyline approach has been applied to the Arctic. The key to the storyline approach is an appropriate identification of drivers, for the reasons explained in lines 335-341. The two drivers identified here are interesting is that they are surprisingly complementary. The analysis is competently performed and the results will be of value to the community. I am thus favourably disposed to publication but feel that just a little more effort would make the paper even more valuable. I also have a number of points which require clarification.
Major comments:
Lines 27-28: Here you describe the ArcAmp index as “lower tropospheric warming over polar land regions”, but in the body of the paper it is defined as everything poleward of 55N. Which is it?
Lines 60-61: Sandwiched between a number of incontrovertible statements comes this one about the links between AA and midlatitude weather extremes, using definitive language and with a reference to Cohen et al. (2014). The authors must surely know that this remains a highly contested topic of research. (Morever the statement here is to some extent contradicted by the text on lines 83-90.) I would suggest dropping the sentence (since you are anyway just giving examples here), but alternatively you should use more nuanced language and provide some more recent references to give a more balanced perspective.
Table 1: For 3 of your 4 fields, the fraction of variance explained by your storylines is not particularly impressive. But for spatially noisy fields, FVE at the gridpoint level is something of a misleading statistic. In Mindlin et al. (2020), which looked at precipitation changes, the FVE was also not very impressive, but when aggregated across a region the storylines did span the range of precipitation changes across the models (which is what matters) quite well. See e.g. Table 2 of Mindlin et al. (2020), where with the exception of one region, the range spanned by the storylines is much larger than the median absolute deviation (MAD) between the actual model response and the response predicted for the model by the storylines. I would encourage the authors to do something similar here, because the storylines may have more explanatory power for these 3 fields than the authors realize.
Figure 2: Storylines are easiest to interpret if they are causal, which in ZS17 and M20 was argued by identifying the statistical responses to drivers with relationships found in other studies, e.g. using model intervention. It would useful if the authors could clarify the extent to which Figure 2 aligns with expectations from other studies.
Figure 2: On this point, if I understand correctly, your BKWarm index is associated with increased rather than decreased sea-ice fraction in the Barents Sea. I didn’t see any comment about that. What is your interpretation of this curious relationship, which would seem to undermine the causality of your storylines?
Figures 3-6: Your usage of “storyline” is not consistent with the usage in ZS17 and M20. They used the term to describe physically plausible changes, the idea being that the future evolution could follow one of the storylines. You seem to be using it instead to describe deviations from the MMM. Hence, to obtain a plausible future change from one of your storylines, e.g. for climate impact studies, one would need to add the MMM to the storyline. Perhaps you have a reason for doing it the way you did, but it makes the interpretation of Figures 3-6 somewhat confusing. I would suggest following the usage in ZS17 and M20 and having your four storylines be the full changes, i.e. including the MMM.
p.15, final paragraph: You could do just a little more here. One of the benefits of storylines is that each storyline represents correlated aspects of climate change. So rather than describing the way different storylines affect the four different fields you analysed, you could describe the combined changes across those four fields for each storyline. Such a summary, in the spirit of compound events, would exploit the power of storylines.
Minor comments:
Line 63: “Shepard” -> “Shepherd”
Line 333: “ML20” -> “M20”
Line 340 and elsewhere: You refer to “mitigation strategies”, but don’t you actually mean “adaptation strategies”?
Citation: https://doi.org/10.5194/egusphere-2023-2741-RC1 -
AC1: 'Reply on RC1', Xavier Levine, 22 Jan 2024
We’re grateful for the referee’s positive outlook and constructive criticism of our work. We’ve strived to answer the referee’s comments, which we think will improve the clarity of our manuscript. While we’ll update our manuscript to fully address the referee’s comments at a later date, we’re already summarising below some of the changes we intend to make:
L. 27-28: ArcAmp is defined as the lower tropospheric warming over ALL regions poleward of 55N. The definition we originally provided in lines 27-28 was incorrect, and we apologise for this error.
Lines 60-61: We agree that our original statement did not adequately reflect the degree of (un)certainty in the proposed causation mechanisms between AA and midlatitude weather extremes. We will modify line 60-61 to highlight that this is still a topic of active research: “The changes to the Arctic climate system have also been suggested to have caused in increase in the frequency and intensity of certain extreme weather over the Northern Hemisphere mid-latitudes (Cohen et al., 2014), even if the mechanism of action and broader importance of such polar-to-midlatitude teleconnection remains controversial (Vavrus, 2018)”. [Ref: Vavrus S.J., 2018: The influence of Arctic amplification on mid-latitude weather and climate. Current Climate Change Reports, 4, 238-249.]
Table 1: We thank the reviewer for this helpful suggestion. While point-to-point correlation as shown on Table 1 may be the most stringent criterion for evaluating the efficacy of our MLR model, we also agree that it may be overly influenced by spatial noise or slight difference in patterns. We will provide at a later date a table similar to that in Mindlin et al. 2020, which we will post to this comment section.
Figure 2: Indeed, a positive BKWarm index appears to be associated with a greater sea-ice fraction over some areas of the Barents Sea. While we do not have any hypothesis to propose to explain this counter-intuitive result, we note that the significance of this result is low, as very little ice is found over the Barents sea in summer in most models. We will add a statement acknowledging this counter-intuitive aspect of the climate response to the manuscript.
Figures 3-6: We fully agree with the reviewer’s assertion that the total response (MMM + anomalies) is what is most relevant to the end-users. Nevertheless, showing the total response makes it harder to distinguish what differentiate storylines, because storylines’ patterns are strongly influenced by the common MMM component. Since this study is focused on describing and explaining what differentiates storylines (rather than assessing their final response and impacts on the surface climate), we think that showing anomalies with respect to the MMM still remains the clearest way to visualize the differences between storylines. We will add a statement in the result section to make this point clear. For optimal clarity, we will add the total response (MMM + anomalies) in a new appendix section (Appendix B) .
p.15, final paragraph: In the last paragraph of Section 4, we provide a qualitative assessment of our storylines on various risks. While it is primarily based on the 2-m temperature response, which is the target variable best explained by our MLR model, we also mention the mitigating / modulating effect of our other target variable on those climate risks. We are not able to provide more than a brief qualitative assessment of climate risks for each storyline due to the complexity of modeling climate risk from physical changes. However, our storylines will be communicated to and used by climate services experts to model those climate risks within the scope of PolarRES (our funding source project).
Minor comments:
Line 63/333: We will correct these typos.
Line 340 and elsewhere: We also agree with the referee’s suggestion that adaptation may be a better choice of word in this context, although mitigation would also be applicable if narrowly defined (e.g. one could say that building resilient infrastructures in the Arctic may be a form of adaptation to, as much as mitigation against, climate change). For clarity we will change Line 340: “Criterion (i) is critical to the viewpoint of the end-users who need a plausible range of 340 climate change scenarios, for instance to develop adaptation strategies (...)”.
Citation: https://doi.org/10.5194/egusphere-2023-2741-AC1
-
AC1: 'Reply on RC1', Xavier Levine, 22 Jan 2024
-
RC2: 'Comment on egusphere-2023-2741', Anonymous Referee #2, 20 Jan 2024
This paper uses a storyline approach to analyze CMIP6 projections of summertime Arctic climate change, which offers unique information for evaluating societal/environmental risk that goes beyond evaluating the multi-model mean. The authors effectively show that uncertainty in two parameters, Barents-Kara sea SST (BkWarm) and Arctic 850hPa temperature (ArcAmp), modulate the spatial pattern of near-surface temperature trends, particularly the relative warming over terrestrial and marine areas. The four storylines produced by these predictors have important implications for the risk of permafrost thawing, wildfire risk, and Arctic Ocean navigability.
As a reader who is familiar with Arctic amplification processes, but new to the storyline approach, I was impressed by the utility this methodology offers, and think these results will be useful to the polar climate science community. That said, I have two major concerns with the current manuscript. First, I found that there was insufficient consideration given to physical mechanisms connecting the predictors to the storylines. Second, it is not clear that the Arctic lower-tropospheric warming predictor (ArcAmp), is quantified in the optimal way, or how different definitions of this term might affect the robustness of the results. I hope my comments below will help the authors address these issues. I recommend major revisions.
Major Comments:
- L182-183: The definition of ArcAmp. How do we know that 850 hPa is the appropriate level to evaluate lower-tropospheric Arctic warming? Attention must be given to the distinct seasonality and vertical structure of Arctic-amplified warming, as well as the different warming mechanisms operating at different pressure levels (e.g., Graversen et al. 2008, Screen and Simmonds 2012, Feldl et al. 2020, Kaufman and Feldl 2022). Finally, it should be noted that the changing Arctic atmospheric energy budget is distinct over land and ocean (Deser et al. 2010). Beyond justifying their choice with prior research on these topics, the authors may want to test the robustness of their storyline results to different definitions of ArcAmp, using different pressure levels and horizontal domains for the spatial average.
- L185-193: The authors assert that their choice of predictor variables is justified due to “(i) their ability to explain a large fraction of the inter-model variability in climate change projections, and to (ii) their connection to a wide array of climatic phenomena in the Arctic and in midlatitude regions, especially near-surface warming.” While (i) is given sufficient evaluation in the subsequent analysis, there is far less work to justify (ii). I recognize that the importance of the Barents-Kara Sea and lower-tropospheric warming for surface climate has been demonstrated in previous studies, as the authors note in the introduction (L83-90). But the mechanisms underpinning these connections are not used to evaluate the storyline results. Why does lower-tropospheric warming influence terrestrial and marine areas differently (Fig. 2b)? Why does the Barents-Kara Sea have outsized influence relative to other areas of the Arctic Ocean?
- L336-343: Similar to my above comment, I was left wanting more discussion of the connection of the storyline predictor terms to physical processes (Criterion ii). The discussion section as written seems to focus almost exclusively to the implications of the different scenarios for climate risks (Criterion i).
Minor Comments:
- L60-61: This is still a heavily debated mechanism, so uncertainty in this consequence of AA should be made clear (e.g. Cohen et al. 2020)
- L120-122: Does having varying numbers of ensemble members included in each model average create any issues regarding inconsistent signal-to-noise ratios across them? Would using a single ensemble member for each model lead to a more consistent treatment?
- L155-158, L194-199: I am curious if the authors could elaborate on the condition of orthoganality amongst the predictors. They make a convincing case to me that the weak correlation between BK-Warm and ArcAmp (r^2 = .15) in this study is acceptable, but it would be useful to future users of this method to have an idea of what an unacceptably high correlation would be, and why.
- 3-6: Significance stippling is only shown for the multi-model mean. Why not show it for the storylines as well?
References:
Deser, Clara, et al. "The seasonal atmospheric response to projected Arctic sea ice loss in the late twenty-first century." Journal of Climate 23.2 (2010): 333-351.
Graversen, Rune G., et al. "Vertical structure of recent Arctic warming." Nature 451.7174 (2008): 53-56.
Screen, James A., Clara Deser, and Ian Simmonds. "Local and remote controls on observed Arctic warming." Geophysical Research Letters 39.10 (2012).
Feldl, Nicole, et al. "Sea ice and atmospheric circulation shape the high-latitude lapse rate feedback." NPJ climate and atmospheric science 3.1 (2020): 41.
Cohen, Judah, et al. "Divergent consensuses on Arctic amplification influence on midlatitude severe winter weather." Nature Climate Change 10.1 (2020): 20-29.
Kaufman, Zachary S., and Nicole Feldl. "Causes of the Arctic’s Lower-Tropospheric Warming Structure." Journal of Climate 35.6 (2022): 1983-2002.
Citation: https://doi.org/10.5194/egusphere-2023-2741-RC2 -
AC2: 'Reply on RC2', Xavier Levine, 22 May 2024
We thank the referee for the thoughtful and constructive review of our work, which has helped improve the quality and readability of our manuscript. While we’ll update our manuscript to fully address the referee’s comments at a later date, we’re already summarizing below some of the changes we intend to make. The referee will find the cited figures and tables in the pdf supplement attached to our response.
Major comments:
L182-183: We agree with the reviewer that Arctic atmospheric warming shows strong seasonal and vertical contrast, and this is already noticeable in the climate change response observed in the past decades (Graversen et al., 2008). As shown in past studies (e.g., Fig. 1 in Graversen et al., 2008), warming rates generally peak in the mid-troposphere (~600 hPa) in summer, but in the boundary layer (<800 hPa) in winter. Likewise, we find that tropospheric warming is nearly uniform across levels in summer (Fig. 1a), unlike winter when it peaks in the boundary layer (Fig. 1b). In summer, we expect the marine boundary layer to be sensitive primarily to changes in ocean currents and state, while the terrestrial boundary layer to respond primarily to atmospheric processes. This is due to the fact that the troposphere is often found to be stably stratified over the Arctic ocean in summer, consistently with the Arctic ocean acting as a heat sink for the boundary layer (Tjernström and Graversen, 2009). The effect of this stable stratification on the climate response is made readily apparent when comparing separately over ocean and land regions how surface temperature changes co-vary with the Arctic temperature profile changes. Specifically, we find that the mid-tropospheric warming (between 850 and 500 hPa) covaries much more strongly with land surface warming than ocean surface warming (Fig. 2a,b), as would be expected from a thermal decoupling of the marine boundary layer from the free troposphere.
Overall, we select the predictors of our Arctic storylines such as to best explain the warming of the marine and terrestrial boundary layers in the Arctic. Our profile analysis suggests that free tropospheric warming (above 850 hPa) may be well suited for explaining changes in the terrestrial warming, due to its strong correlation with land warming (Fig. 2b). In our study, we select the Arctic-mean 850-hPa temperature change as our predictor for the Arctic free tropospheric warming, due to its strong performance for explaining 2-m temperature changes in the Arctic in a multivariate linear regression (MLR) framework in conjunction with our other predictor, the Barents-Kara sea warming. Compared with other vertical levels, the 850 hPa level maximizes the variance explained by the MLR (Table 1, top row) while being only weakly correlated with the Barents-Kara sea warming (Table 1, bottom row). The relevance of ArcAmp to land temperatures is confirmed by our pattern response analysis (Fig. 3a). We select the Barents-Kara sea warming as a predictor for the marine boundary layer. We picked the Barents-Kara sea as our reference region specifically because of its role as a climate gateway between the North Atlantic and the Arctic ocean (e.g. Smedsrud et al., 2013). Our pattern response analysis confirms the strong covariability of the Barents-Kara seas warming with both the Central Arctic and North Atlantic sea warming (Fig. 3b). However, our analysis does not seek to establish a causation between North Atlantic, Barents-Kara or Central Arctic warming, but rather to leverage the strong co-variability across all three regions with global warming. Hence, Central Arctic or North Atlantic warming could have been chosen as predictors instead of the Barents-Kara sea warming, with little difference in the overall fitness of the model (see Table 2).
=> We will expand section 2.4 ("Choice of predictor indices”) to better explain those points, as follows: “In this study, we select Arctic atmospheric amplification and Barents-Kara Sea warming as our predictors (...). Both ‘ArcAmp’ and ‘BKWarm’ are defined over the extended summer season (May to October). We choose those two predictors owing to their ability to explain a large fraction of the inter-model variability in climate change projections in the Arctic, specifically the warming of the boundary layer over marine and terrestrial regions. Indeed, comparing the vertical profile of temperature against surface temperature in the Arctic regions shows a strong covariability over land but weak covariability over marine areas (see Fig 2a,b), consistent with the thermal decoupling of the marine boundary layer from the free troposphere in summer (e.g., Tjernström and Graversen, 2009). Over ocean regions, the warming of the marine boundary layer is found to warm coherently across the Central Arctic, Barents-Kara, and North Atlantic regions, in agreement with a coherent increase in sea surface temperature across those regions. Due to its role as a climate gateway between the North Atlantic and the Arctic ocean (e.g. Smedsrud et al., 2013), we select the Barents-Kara sea as our reference region for defining our ocean warming predictor in the Arctic. Conversely, we select the 850 hPa Arctic mean temperature warming as our second predictor due to its high degree of covariability with the warming of the terrestrial boundary layer and low degree of covariability with the marine boundary layer warming (see Table 1). ” We will be adding an appendix B in the manuscript featuring our vertical profile analysis (Fig. 2, Tables 1 & 2).
L185-193: We’ve addressed some of those points in our previous comment about the choice of predictors. To briefly summarize, we find a strong covariability between free tropospheric warming and land temperatures, while the marine boundary layer is found to warm coherently across the Arctic ocean and North Atlantic. We expect surface temperature over terrestrial areas to be more sensitive to atmospheric processes than ocean regions, due to the strong thermal inertia of the ocean and the stable thermal stratification of the troposphere in summer over the Arctic ocean. The processes tying temperature anomalies in the free troposphere to that of the surface over land likely involve multiple atmospheric feedbacks, such as cloud and clear sky radiative feedbacks and boundary layer mixing changes; those feedbacks are beyond the scope of this study, however, although we would like be analyze them in a future study. Regarding our selection of the Barents-Kara sea warming (BKWarm), we motivate it primarily based on its known importance as a mediator of climate change between the North Atlantic and Central Arctic. As mentioned in our previous comments, our study does not seek to outline a mechanism connecting those 3 regions, as it would require an in-depth analysis of changes in ocean current, upper-ocean mixing, and surface fluxes, which are beyond the scope of this study.
=> We will expand section 2.4 ("Choice of predictor indices”) to explicitly state the limitations of our study. “In this study, we select Arctic atmospheric amplification and Barents-Kara Sea warming as our predictors (...). Conversely, we select the 850 hPa Arctic mean temperature warming (...) marine boundary layer warming (see Table 1). The processes tying temperature anomalies in the free troposphere to that of the surface over land likely involve multiple atmospheric feedbacks, such as radiative or boundary layer mixing changes, which is beyond the scope of this study. Likewise, while our study leverages the connection between the North Atlantic, Barents-Kara and Central Arctic warming to produce a predictor for the marine boundary layer warming, it does not seek to outline a mechanism connecting those three regions as it would require an in-depth analysis of changes in ocean current, upper-ocean mixing, and surface fluxes, which is beyond the scope of this study.”
L336-343: Our study does not address the mechanism behind the links of our predictors with regional and global changes in the atmospheric circulation, which is beyond the scope of this study. Yet, our predictors are connected to changing surface climate on a global scale, tying our predictors to possible global circulation changes with global warming. For instance, the response of the 2-m temperature to changes in ArcAmp and BKWarm show distinctively different spatial patterns (Fig. 4). ArcAmp is associated with a bipolar climate change signal, with warming over the Northern Hemisphere land regions and over the Pacific ocean, but a cooling over the Southern ocean (Fig. 4a). This is likely linked to changes in sensible and latent poleward heat fluxes (Kauffman and Feldl, 2022), reaching far across hemispheres. BKWarm is associated with a strong warming of the northern hemisphere high latitude marine regions and slight cooling of the deep tropics (Fig. 4b). We intend to explore more in depth those teleconnections in a future study.
Minor comments:
L60-61: We agree with the reviewer that our original statement did not adequately reflect the degree of (un)certainty in the proposed causation mechanisms between AA and mid-latitudes weather extremes. We will modify line 60-61 to highlight that this is still a topic of active research: “The changes to the Arctic climate system have also been suggested to have caused an increase in the frequency and intensity of certain extreme weather over the Northern Hemisphere mid-latitudes (Cohen et al., 2014), even if the mechanism of action and broader importance of such polar-to-midlatitude teleconnection remains controversial (Vavrus, 2018)”.
L120-122: This is a fair point made by the reviewer. Ideally, we would have liked to use multiple members for all models, and the same number of members for all models. Initially, we were concerned that having only one member for all members would make it challenging to disentangle natural variability from climate change in our storyline analysis. Fortunately, we show that individual members of the same model are more alike than the ensemble-mean of different models, as shown by our 2 predictors (Fig. 1b in the manuscript). This provides strong evidence for the climate change response to dominate over natural variability, and suggests that using only one member would provide similar storylines. We will add a point to this effect in section 2.1 (“Model data”): “Our analysis uses a set of 42 climate models from CMIP6 (...). As most models only have a few members, setting a maximum of 15 members seems a reasonable trade-off for reducing internal variability while including as many models as possible. We find little difference in using only a single member or an ensemble-mean of members, as the climate projections are dominated by the effect of the climate forcing with only a small contribution from natural variability (see Fig. 1b)”.
L155-158: Orthogonality is not a necessity for constructing storylines since one can build a meaningful MLR model with covarying predictors, yet orthogonality remains desirable for interpreting changes. However, we do not offer any specific guidance on what may constitute a ‘good enough’ orthogonality (i.e., what may be the maximum value of a correlation coefficient), as this may be dependent on the context and situation. It is also possible to use a modified linear regression analysis when predictors are more strongly correlated, which preserves some of the essential properties of interpretability for the storylines (see ‘sequential regression’ in Mindlin et al., 2020).
Fig 3-6: We will add significance stippling on the storyline plots (Figs. 3, 4, 5, 6), which will consist of all regions in which at least one of the response coefficients shows significance (as defined and shown in the normalized response coefficients, Fig. 2 ).
References:
Graversen, R.G., Mauritsen, T., Tjernström, M., Källén, E. and Svensson, G., 2008. Vertical structure of recent Arctic warming. Nature, 451, 53-56.
Tjernström, M. and Graversen, R.G., 2009. The vertical structure of the lower Arctic troposphere analysed from observations and the ERA‐40 reanalysis. Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography, 135, 431-443.
Smedsrud, L.H., Esau, I., Ingvaldsen, R.B., Eldevik, T., Haugan, P.M., Li, C., Lien, V.S., Olsen, A., Omar, A.M., Otterå, O.H. and Risebrobakken, B., 2013. The role of the Barents Sea in the Arctic climate system. Reviews of Geophysics, 51, 415-449.
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2741', Anonymous Referee #1, 19 Dec 2023
This paper applies the storyline approach developed by Zappa & Shepherd (2017) and Mindlin et al. (2020), which is based on multiple linear regression across CMIP models, to represent the uncertainty in the response of summertime Arctic climate to greenhouse warming in terms of two key drivers: BK SST warming, and pan-Arctic lower tropospheric warming. So far as I am aware this is the first time that the storyline approach has been applied to the Arctic. The key to the storyline approach is an appropriate identification of drivers, for the reasons explained in lines 335-341. The two drivers identified here are interesting is that they are surprisingly complementary. The analysis is competently performed and the results will be of value to the community. I am thus favourably disposed to publication but feel that just a little more effort would make the paper even more valuable. I also have a number of points which require clarification.
Major comments:
Lines 27-28: Here you describe the ArcAmp index as “lower tropospheric warming over polar land regions”, but in the body of the paper it is defined as everything poleward of 55N. Which is it?
Lines 60-61: Sandwiched between a number of incontrovertible statements comes this one about the links between AA and midlatitude weather extremes, using definitive language and with a reference to Cohen et al. (2014). The authors must surely know that this remains a highly contested topic of research. (Morever the statement here is to some extent contradicted by the text on lines 83-90.) I would suggest dropping the sentence (since you are anyway just giving examples here), but alternatively you should use more nuanced language and provide some more recent references to give a more balanced perspective.
Table 1: For 3 of your 4 fields, the fraction of variance explained by your storylines is not particularly impressive. But for spatially noisy fields, FVE at the gridpoint level is something of a misleading statistic. In Mindlin et al. (2020), which looked at precipitation changes, the FVE was also not very impressive, but when aggregated across a region the storylines did span the range of precipitation changes across the models (which is what matters) quite well. See e.g. Table 2 of Mindlin et al. (2020), where with the exception of one region, the range spanned by the storylines is much larger than the median absolute deviation (MAD) between the actual model response and the response predicted for the model by the storylines. I would encourage the authors to do something similar here, because the storylines may have more explanatory power for these 3 fields than the authors realize.
Figure 2: Storylines are easiest to interpret if they are causal, which in ZS17 and M20 was argued by identifying the statistical responses to drivers with relationships found in other studies, e.g. using model intervention. It would useful if the authors could clarify the extent to which Figure 2 aligns with expectations from other studies.
Figure 2: On this point, if I understand correctly, your BKWarm index is associated with increased rather than decreased sea-ice fraction in the Barents Sea. I didn’t see any comment about that. What is your interpretation of this curious relationship, which would seem to undermine the causality of your storylines?
Figures 3-6: Your usage of “storyline” is not consistent with the usage in ZS17 and M20. They used the term to describe physically plausible changes, the idea being that the future evolution could follow one of the storylines. You seem to be using it instead to describe deviations from the MMM. Hence, to obtain a plausible future change from one of your storylines, e.g. for climate impact studies, one would need to add the MMM to the storyline. Perhaps you have a reason for doing it the way you did, but it makes the interpretation of Figures 3-6 somewhat confusing. I would suggest following the usage in ZS17 and M20 and having your four storylines be the full changes, i.e. including the MMM.
p.15, final paragraph: You could do just a little more here. One of the benefits of storylines is that each storyline represents correlated aspects of climate change. So rather than describing the way different storylines affect the four different fields you analysed, you could describe the combined changes across those four fields for each storyline. Such a summary, in the spirit of compound events, would exploit the power of storylines.
Minor comments:
Line 63: “Shepard” -> “Shepherd”
Line 333: “ML20” -> “M20”
Line 340 and elsewhere: You refer to “mitigation strategies”, but don’t you actually mean “adaptation strategies”?
Citation: https://doi.org/10.5194/egusphere-2023-2741-RC1 -
AC1: 'Reply on RC1', Xavier Levine, 22 Jan 2024
We’re grateful for the referee’s positive outlook and constructive criticism of our work. We’ve strived to answer the referee’s comments, which we think will improve the clarity of our manuscript. While we’ll update our manuscript to fully address the referee’s comments at a later date, we’re already summarising below some of the changes we intend to make:
L. 27-28: ArcAmp is defined as the lower tropospheric warming over ALL regions poleward of 55N. The definition we originally provided in lines 27-28 was incorrect, and we apologise for this error.
Lines 60-61: We agree that our original statement did not adequately reflect the degree of (un)certainty in the proposed causation mechanisms between AA and midlatitude weather extremes. We will modify line 60-61 to highlight that this is still a topic of active research: “The changes to the Arctic climate system have also been suggested to have caused in increase in the frequency and intensity of certain extreme weather over the Northern Hemisphere mid-latitudes (Cohen et al., 2014), even if the mechanism of action and broader importance of such polar-to-midlatitude teleconnection remains controversial (Vavrus, 2018)”. [Ref: Vavrus S.J., 2018: The influence of Arctic amplification on mid-latitude weather and climate. Current Climate Change Reports, 4, 238-249.]
Table 1: We thank the reviewer for this helpful suggestion. While point-to-point correlation as shown on Table 1 may be the most stringent criterion for evaluating the efficacy of our MLR model, we also agree that it may be overly influenced by spatial noise or slight difference in patterns. We will provide at a later date a table similar to that in Mindlin et al. 2020, which we will post to this comment section.
Figure 2: Indeed, a positive BKWarm index appears to be associated with a greater sea-ice fraction over some areas of the Barents Sea. While we do not have any hypothesis to propose to explain this counter-intuitive result, we note that the significance of this result is low, as very little ice is found over the Barents sea in summer in most models. We will add a statement acknowledging this counter-intuitive aspect of the climate response to the manuscript.
Figures 3-6: We fully agree with the reviewer’s assertion that the total response (MMM + anomalies) is what is most relevant to the end-users. Nevertheless, showing the total response makes it harder to distinguish what differentiate storylines, because storylines’ patterns are strongly influenced by the common MMM component. Since this study is focused on describing and explaining what differentiates storylines (rather than assessing their final response and impacts on the surface climate), we think that showing anomalies with respect to the MMM still remains the clearest way to visualize the differences between storylines. We will add a statement in the result section to make this point clear. For optimal clarity, we will add the total response (MMM + anomalies) in a new appendix section (Appendix B) .
p.15, final paragraph: In the last paragraph of Section 4, we provide a qualitative assessment of our storylines on various risks. While it is primarily based on the 2-m temperature response, which is the target variable best explained by our MLR model, we also mention the mitigating / modulating effect of our other target variable on those climate risks. We are not able to provide more than a brief qualitative assessment of climate risks for each storyline due to the complexity of modeling climate risk from physical changes. However, our storylines will be communicated to and used by climate services experts to model those climate risks within the scope of PolarRES (our funding source project).
Minor comments:
Line 63/333: We will correct these typos.
Line 340 and elsewhere: We also agree with the referee’s suggestion that adaptation may be a better choice of word in this context, although mitigation would also be applicable if narrowly defined (e.g. one could say that building resilient infrastructures in the Arctic may be a form of adaptation to, as much as mitigation against, climate change). For clarity we will change Line 340: “Criterion (i) is critical to the viewpoint of the end-users who need a plausible range of 340 climate change scenarios, for instance to develop adaptation strategies (...)”.
Citation: https://doi.org/10.5194/egusphere-2023-2741-AC1
-
AC1: 'Reply on RC1', Xavier Levine, 22 Jan 2024
-
RC2: 'Comment on egusphere-2023-2741', Anonymous Referee #2, 20 Jan 2024
This paper uses a storyline approach to analyze CMIP6 projections of summertime Arctic climate change, which offers unique information for evaluating societal/environmental risk that goes beyond evaluating the multi-model mean. The authors effectively show that uncertainty in two parameters, Barents-Kara sea SST (BkWarm) and Arctic 850hPa temperature (ArcAmp), modulate the spatial pattern of near-surface temperature trends, particularly the relative warming over terrestrial and marine areas. The four storylines produced by these predictors have important implications for the risk of permafrost thawing, wildfire risk, and Arctic Ocean navigability.
As a reader who is familiar with Arctic amplification processes, but new to the storyline approach, I was impressed by the utility this methodology offers, and think these results will be useful to the polar climate science community. That said, I have two major concerns with the current manuscript. First, I found that there was insufficient consideration given to physical mechanisms connecting the predictors to the storylines. Second, it is not clear that the Arctic lower-tropospheric warming predictor (ArcAmp), is quantified in the optimal way, or how different definitions of this term might affect the robustness of the results. I hope my comments below will help the authors address these issues. I recommend major revisions.
Major Comments:
- L182-183: The definition of ArcAmp. How do we know that 850 hPa is the appropriate level to evaluate lower-tropospheric Arctic warming? Attention must be given to the distinct seasonality and vertical structure of Arctic-amplified warming, as well as the different warming mechanisms operating at different pressure levels (e.g., Graversen et al. 2008, Screen and Simmonds 2012, Feldl et al. 2020, Kaufman and Feldl 2022). Finally, it should be noted that the changing Arctic atmospheric energy budget is distinct over land and ocean (Deser et al. 2010). Beyond justifying their choice with prior research on these topics, the authors may want to test the robustness of their storyline results to different definitions of ArcAmp, using different pressure levels and horizontal domains for the spatial average.
- L185-193: The authors assert that their choice of predictor variables is justified due to “(i) their ability to explain a large fraction of the inter-model variability in climate change projections, and to (ii) their connection to a wide array of climatic phenomena in the Arctic and in midlatitude regions, especially near-surface warming.” While (i) is given sufficient evaluation in the subsequent analysis, there is far less work to justify (ii). I recognize that the importance of the Barents-Kara Sea and lower-tropospheric warming for surface climate has been demonstrated in previous studies, as the authors note in the introduction (L83-90). But the mechanisms underpinning these connections are not used to evaluate the storyline results. Why does lower-tropospheric warming influence terrestrial and marine areas differently (Fig. 2b)? Why does the Barents-Kara Sea have outsized influence relative to other areas of the Arctic Ocean?
- L336-343: Similar to my above comment, I was left wanting more discussion of the connection of the storyline predictor terms to physical processes (Criterion ii). The discussion section as written seems to focus almost exclusively to the implications of the different scenarios for climate risks (Criterion i).
Minor Comments:
- L60-61: This is still a heavily debated mechanism, so uncertainty in this consequence of AA should be made clear (e.g. Cohen et al. 2020)
- L120-122: Does having varying numbers of ensemble members included in each model average create any issues regarding inconsistent signal-to-noise ratios across them? Would using a single ensemble member for each model lead to a more consistent treatment?
- L155-158, L194-199: I am curious if the authors could elaborate on the condition of orthoganality amongst the predictors. They make a convincing case to me that the weak correlation between BK-Warm and ArcAmp (r^2 = .15) in this study is acceptable, but it would be useful to future users of this method to have an idea of what an unacceptably high correlation would be, and why.
- 3-6: Significance stippling is only shown for the multi-model mean. Why not show it for the storylines as well?
References:
Deser, Clara, et al. "The seasonal atmospheric response to projected Arctic sea ice loss in the late twenty-first century." Journal of Climate 23.2 (2010): 333-351.
Graversen, Rune G., et al. "Vertical structure of recent Arctic warming." Nature 451.7174 (2008): 53-56.
Screen, James A., Clara Deser, and Ian Simmonds. "Local and remote controls on observed Arctic warming." Geophysical Research Letters 39.10 (2012).
Feldl, Nicole, et al. "Sea ice and atmospheric circulation shape the high-latitude lapse rate feedback." NPJ climate and atmospheric science 3.1 (2020): 41.
Cohen, Judah, et al. "Divergent consensuses on Arctic amplification influence on midlatitude severe winter weather." Nature Climate Change 10.1 (2020): 20-29.
Kaufman, Zachary S., and Nicole Feldl. "Causes of the Arctic’s Lower-Tropospheric Warming Structure." Journal of Climate 35.6 (2022): 1983-2002.
Citation: https://doi.org/10.5194/egusphere-2023-2741-RC2 -
AC2: 'Reply on RC2', Xavier Levine, 22 May 2024
We thank the referee for the thoughtful and constructive review of our work, which has helped improve the quality and readability of our manuscript. While we’ll update our manuscript to fully address the referee’s comments at a later date, we’re already summarizing below some of the changes we intend to make. The referee will find the cited figures and tables in the pdf supplement attached to our response.
Major comments:
L182-183: We agree with the reviewer that Arctic atmospheric warming shows strong seasonal and vertical contrast, and this is already noticeable in the climate change response observed in the past decades (Graversen et al., 2008). As shown in past studies (e.g., Fig. 1 in Graversen et al., 2008), warming rates generally peak in the mid-troposphere (~600 hPa) in summer, but in the boundary layer (<800 hPa) in winter. Likewise, we find that tropospheric warming is nearly uniform across levels in summer (Fig. 1a), unlike winter when it peaks in the boundary layer (Fig. 1b). In summer, we expect the marine boundary layer to be sensitive primarily to changes in ocean currents and state, while the terrestrial boundary layer to respond primarily to atmospheric processes. This is due to the fact that the troposphere is often found to be stably stratified over the Arctic ocean in summer, consistently with the Arctic ocean acting as a heat sink for the boundary layer (Tjernström and Graversen, 2009). The effect of this stable stratification on the climate response is made readily apparent when comparing separately over ocean and land regions how surface temperature changes co-vary with the Arctic temperature profile changes. Specifically, we find that the mid-tropospheric warming (between 850 and 500 hPa) covaries much more strongly with land surface warming than ocean surface warming (Fig. 2a,b), as would be expected from a thermal decoupling of the marine boundary layer from the free troposphere.
Overall, we select the predictors of our Arctic storylines such as to best explain the warming of the marine and terrestrial boundary layers in the Arctic. Our profile analysis suggests that free tropospheric warming (above 850 hPa) may be well suited for explaining changes in the terrestrial warming, due to its strong correlation with land warming (Fig. 2b). In our study, we select the Arctic-mean 850-hPa temperature change as our predictor for the Arctic free tropospheric warming, due to its strong performance for explaining 2-m temperature changes in the Arctic in a multivariate linear regression (MLR) framework in conjunction with our other predictor, the Barents-Kara sea warming. Compared with other vertical levels, the 850 hPa level maximizes the variance explained by the MLR (Table 1, top row) while being only weakly correlated with the Barents-Kara sea warming (Table 1, bottom row). The relevance of ArcAmp to land temperatures is confirmed by our pattern response analysis (Fig. 3a). We select the Barents-Kara sea warming as a predictor for the marine boundary layer. We picked the Barents-Kara sea as our reference region specifically because of its role as a climate gateway between the North Atlantic and the Arctic ocean (e.g. Smedsrud et al., 2013). Our pattern response analysis confirms the strong covariability of the Barents-Kara seas warming with both the Central Arctic and North Atlantic sea warming (Fig. 3b). However, our analysis does not seek to establish a causation between North Atlantic, Barents-Kara or Central Arctic warming, but rather to leverage the strong co-variability across all three regions with global warming. Hence, Central Arctic or North Atlantic warming could have been chosen as predictors instead of the Barents-Kara sea warming, with little difference in the overall fitness of the model (see Table 2).
=> We will expand section 2.4 ("Choice of predictor indices”) to better explain those points, as follows: “In this study, we select Arctic atmospheric amplification and Barents-Kara Sea warming as our predictors (...). Both ‘ArcAmp’ and ‘BKWarm’ are defined over the extended summer season (May to October). We choose those two predictors owing to their ability to explain a large fraction of the inter-model variability in climate change projections in the Arctic, specifically the warming of the boundary layer over marine and terrestrial regions. Indeed, comparing the vertical profile of temperature against surface temperature in the Arctic regions shows a strong covariability over land but weak covariability over marine areas (see Fig 2a,b), consistent with the thermal decoupling of the marine boundary layer from the free troposphere in summer (e.g., Tjernström and Graversen, 2009). Over ocean regions, the warming of the marine boundary layer is found to warm coherently across the Central Arctic, Barents-Kara, and North Atlantic regions, in agreement with a coherent increase in sea surface temperature across those regions. Due to its role as a climate gateway between the North Atlantic and the Arctic ocean (e.g. Smedsrud et al., 2013), we select the Barents-Kara sea as our reference region for defining our ocean warming predictor in the Arctic. Conversely, we select the 850 hPa Arctic mean temperature warming as our second predictor due to its high degree of covariability with the warming of the terrestrial boundary layer and low degree of covariability with the marine boundary layer warming (see Table 1). ” We will be adding an appendix B in the manuscript featuring our vertical profile analysis (Fig. 2, Tables 1 & 2).
L185-193: We’ve addressed some of those points in our previous comment about the choice of predictors. To briefly summarize, we find a strong covariability between free tropospheric warming and land temperatures, while the marine boundary layer is found to warm coherently across the Arctic ocean and North Atlantic. We expect surface temperature over terrestrial areas to be more sensitive to atmospheric processes than ocean regions, due to the strong thermal inertia of the ocean and the stable thermal stratification of the troposphere in summer over the Arctic ocean. The processes tying temperature anomalies in the free troposphere to that of the surface over land likely involve multiple atmospheric feedbacks, such as cloud and clear sky radiative feedbacks and boundary layer mixing changes; those feedbacks are beyond the scope of this study, however, although we would like be analyze them in a future study. Regarding our selection of the Barents-Kara sea warming (BKWarm), we motivate it primarily based on its known importance as a mediator of climate change between the North Atlantic and Central Arctic. As mentioned in our previous comments, our study does not seek to outline a mechanism connecting those 3 regions, as it would require an in-depth analysis of changes in ocean current, upper-ocean mixing, and surface fluxes, which are beyond the scope of this study.
=> We will expand section 2.4 ("Choice of predictor indices”) to explicitly state the limitations of our study. “In this study, we select Arctic atmospheric amplification and Barents-Kara Sea warming as our predictors (...). Conversely, we select the 850 hPa Arctic mean temperature warming (...) marine boundary layer warming (see Table 1). The processes tying temperature anomalies in the free troposphere to that of the surface over land likely involve multiple atmospheric feedbacks, such as radiative or boundary layer mixing changes, which is beyond the scope of this study. Likewise, while our study leverages the connection between the North Atlantic, Barents-Kara and Central Arctic warming to produce a predictor for the marine boundary layer warming, it does not seek to outline a mechanism connecting those three regions as it would require an in-depth analysis of changes in ocean current, upper-ocean mixing, and surface fluxes, which is beyond the scope of this study.”
L336-343: Our study does not address the mechanism behind the links of our predictors with regional and global changes in the atmospheric circulation, which is beyond the scope of this study. Yet, our predictors are connected to changing surface climate on a global scale, tying our predictors to possible global circulation changes with global warming. For instance, the response of the 2-m temperature to changes in ArcAmp and BKWarm show distinctively different spatial patterns (Fig. 4). ArcAmp is associated with a bipolar climate change signal, with warming over the Northern Hemisphere land regions and over the Pacific ocean, but a cooling over the Southern ocean (Fig. 4a). This is likely linked to changes in sensible and latent poleward heat fluxes (Kauffman and Feldl, 2022), reaching far across hemispheres. BKWarm is associated with a strong warming of the northern hemisphere high latitude marine regions and slight cooling of the deep tropics (Fig. 4b). We intend to explore more in depth those teleconnections in a future study.
Minor comments:
L60-61: We agree with the reviewer that our original statement did not adequately reflect the degree of (un)certainty in the proposed causation mechanisms between AA and mid-latitudes weather extremes. We will modify line 60-61 to highlight that this is still a topic of active research: “The changes to the Arctic climate system have also been suggested to have caused an increase in the frequency and intensity of certain extreme weather over the Northern Hemisphere mid-latitudes (Cohen et al., 2014), even if the mechanism of action and broader importance of such polar-to-midlatitude teleconnection remains controversial (Vavrus, 2018)”.
L120-122: This is a fair point made by the reviewer. Ideally, we would have liked to use multiple members for all models, and the same number of members for all models. Initially, we were concerned that having only one member for all members would make it challenging to disentangle natural variability from climate change in our storyline analysis. Fortunately, we show that individual members of the same model are more alike than the ensemble-mean of different models, as shown by our 2 predictors (Fig. 1b in the manuscript). This provides strong evidence for the climate change response to dominate over natural variability, and suggests that using only one member would provide similar storylines. We will add a point to this effect in section 2.1 (“Model data”): “Our analysis uses a set of 42 climate models from CMIP6 (...). As most models only have a few members, setting a maximum of 15 members seems a reasonable trade-off for reducing internal variability while including as many models as possible. We find little difference in using only a single member or an ensemble-mean of members, as the climate projections are dominated by the effect of the climate forcing with only a small contribution from natural variability (see Fig. 1b)”.
L155-158: Orthogonality is not a necessity for constructing storylines since one can build a meaningful MLR model with covarying predictors, yet orthogonality remains desirable for interpreting changes. However, we do not offer any specific guidance on what may constitute a ‘good enough’ orthogonality (i.e., what may be the maximum value of a correlation coefficient), as this may be dependent on the context and situation. It is also possible to use a modified linear regression analysis when predictors are more strongly correlated, which preserves some of the essential properties of interpretability for the storylines (see ‘sequential regression’ in Mindlin et al., 2020).
Fig 3-6: We will add significance stippling on the storyline plots (Figs. 3, 4, 5, 6), which will consist of all regions in which at least one of the response coefficients shows significance (as defined and shown in the normalized response coefficients, Fig. 2 ).
References:
Graversen, R.G., Mauritsen, T., Tjernström, M., Källén, E. and Svensson, G., 2008. Vertical structure of recent Arctic warming. Nature, 451, 53-56.
Tjernström, M. and Graversen, R.G., 2009. The vertical structure of the lower Arctic troposphere analysed from observations and the ERA‐40 reanalysis. Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography, 135, 431-443.
Smedsrud, L.H., Esau, I., Ingvaldsen, R.B., Eldevik, T., Haugan, P.M., Li, C., Lien, V.S., Olsen, A., Omar, A.M., Otterå, O.H. and Risebrobakken, B., 2013. The role of the Barents Sea in the Arctic climate system. Reviews of Geophysics, 51, 415-449.
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 451 | 161 | 39 | 651 | 30 | 30 |
- HTML: 451
- PDF: 161
- XML: 39
- Total: 651
- BibTeX: 30
- EndNote: 30
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 213 | 33 |
| Germany | 2 | 70 | 10 |
| Norway | 3 | 58 | 9 |
| United Kingdom | 4 | 46 | 7 |
| Netherlands | 5 | 33 | 5 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 213
Cited
Ryan Williams
Gareth Marshall
Andrew Orr
Lise Seland Graff
Dörthe Handorf
Alexey Karpechko
Raphael Köhler
René Wijngaard
Nadine Johnston
Hanna Lee
Lars Nieradzik
Priscilla Mooney
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(4982 KB) - Metadata XML