Review of the Manuscript: “Storylines of extreme summer temperatures in southern South America” by Suli et al.

I have received and read the manuscript. The authors employ a storyline approach to verify the climate change responses of maximum summer temperatures in four regions of southern South America, aiming to better understand the drivers of structural uncertainties in GCM projections. They concluded that the dominant drivers of the projected warming in maximum summer temperatures vary substantially across regions and often reflect a combination of thermodynamic and dynamical aspects of climate change. Their results are promising, and their analysis is adequate. However, I recommend minor revisions based on the suggestions as follows:

• Line 63: Where is Iberia located?
• Lines 114-115: How did you interpolate the data? Have you considered interpolating all data to the coarser resolution? It should be explained as well.
• I recommend not abbreviating southern South America (SSA) as it may be confused with southern southern South America (SS).
• Section 2.3: Please specify which drivers are local and which are remote (and why, preferably) in this Section rather than in the Introduction.
• The manuscript has too many abbreviations. I recommend adding a table with all of them to support the reader.
• I suggest using grey dots in areas where changes are statistically significant in Figure 2.
• Lines 276-278. This sentence should be tied to the last paragraph.
• I am not sure if Table 2 is relevant to the manuscript. Results are not discussed in the text.
• Line 297: What does weak ZCAS intensification mean?
• The conclusions are quite long, and the writing is confusing. I recommend shortening it to something between 700 and 900 words.

Review of the Manuscript: “Storylines of extreme summer temperatures in southern South America” by Suli et al.

The authors provide a storyline analysis of summer absolute maximum temperature (TXx) over four regions of southern South America: northern, central-eastern, central Argentina, and southern areas. The storyline approach allows them to explain 56% of inter-model spread in projections of TXx.  I found the analysis very interesting and highly relevant to better understand future projections of climate in South America. However, I find that the authors could improve the robustness of their results if they considered the uncertainty in the remote drivers due to internal variability. Methodologies to account for this uncertainty in storyline studies have been presented (e.g. in an article cited here, Mindlin et al. (2020), based on which other analysis have estimated uncertainty in the responses). I suggest major revisions and that the authors consider the suggestions to ensure that the statistical methods of storyline analysis are performed as robustly as possible. It is fundamental to ensure that the regression captures differences across models that are representative of climatological changes and not artifacts of model biases or internal variability.  

The authors put their analysis in context and cite recently published literature illustrating there is uncertainty in extreme temperatures in the region of study (IPCC, 2023), justifying this analysis. In lines 55-70 previous storyline analysis are referenced. However, the authors do not reference Mindlin et al. (2024) who analysed climate impact drivers in the same region (https://doi.org/10.1016/j.cliser.2024.100480). It is particularly interesting that in that paper, the authors show that with a “low Pacific warming” storyline (which would correspond to “Low ΔN3.4 +”)  the maximum temperature in the extended warm season ONDJFM is high more extreme than under a high Pacific storyline. This is the case both in the north of SS, CCH and the west of CES. It would be interesting if the authors could put their results in the context of this other storyline analysis of temperature indicators for the same region. The authors mention Garrido-Perez et al. (2024) on Iberia, but I don’t understand how this has to do with their study.

105 Why did the authors use one ensemble member per model? It would be useful to understand the uncertainty in the local drivers. In particular, it would be good to know that the soil moisture response is robust and that the uncertainty is not associated with internal variability.

115-120 The authors use absolute values of maximum temperature as an index. What I understand from other storyline analysis based on precipitation changes, is that climatological changes are analysed and hence the regression analysis captures the difference across models that can be explained due to uncertain changes in each driver. In this case, the absolute value of maximum temperature in the models might differ due to model biases and hence the biases can be confounding the signal. Could you please explain how in this analysis the potential biases in models are addressed?

125 Regionalization is based on present climate. However, the storyline analysis allows the authors to better understand the uncertainty in the forced response. The signal of change could not overlap with the climatological regions. Could you please better justify the regionalization being based on the reference climatology and not on the regions where the projected change is most uncertain?

145 – 150

On the selection of remote and local drivers, the authors chose to characterize SACZ changes in terms of SACZ intensity. Zilli et al. (2019) and Zilli and Carvalho (2021) reported a poleward shift of the SACZ and methodologies of convergence zone shift have been proposed and used in other storyline analysis. In this analysis of storylines of for the South Pacific Convergence Zone the authors illustrate that the poleward shift and strength of such poleward shift explains a great part of the model uncertainty https://doi.org/10.1175/JCLI-D-21-0433.1 I am concerned that by defining and index based on the difference of OLR between the two proposed boxes as a way of capturing the intensity of the SACZ the authors in this study miss the most important component of the uncertainty in the SACZ chanages and its associated impacts. This would be captured by an index that captures spread in the poleward shift of the SACZ. Indeed in their results, only for CES the sensitivity pattern is significant in a region expected to be affected by the SACZ. Could you please illustrate what is the model spread of the SACZ change that this index captures?

Regional soil moisture is a very hard variable to work with. Did the researchers explore the performance of the models in simulating the climatology and variability of soil moisture? A storyline analysis based on this variable can be hard to trust if the variable itself is not well represented. Is the spread in soil moisture change distinguishable between models (i.e. what is the spread in climatological soil moisture in the models? Is the spread significant with respect to internal variability?). The significance of remote driver changes was not evaluated in Zappa and Shepherd (2017) but has been reported in storyline analysis like Mindlin et al. (2020), appendix.

I am not convinced about how appropiate the index of local Z500 to be representative of model uncertainty in circulation change. Is Z500 particularly uncertain over the area where it is averaged? This driver is used to characterize changes around 50S. From a dynamics perspective, any local circulation driver in this region has to capture the uncertainty in regional circulation change. For example, as the author reference Mindlin et al. 2020, the changes there are understood in terms of changes in the westerlies, which influence this region (Figure 3b and 3f – if I can interpret this figure correctly, see below). What is the circulation feature that is projected to change and is captured by this index? Is  the climatological strength of Z500 locally a circulation feature that is chaning in the models?  

215 I don’t understand why the drivers cannot be correlated. Isn’t the multi-linear regression framework used precisely used to controll for any confounding effect? Drivers can be correlated in multi-linear regression frameworks. I recommend the authors this review article on how to interpret regression models: http://dx.doi.org/10.2139/ssrn.3689437

I don’t understand the discussion related to uncertainty in TXx changes over CES. If it is explained by soil moisture, why are storylines based on this driver presented?

230 A correlation analysis is useful to show that the target variable is correlated with changes in the region. However, the way that the results are presented the caption is wrong, as well as some parts of the interpretation of the maps in the text. The authors say “Figure 3 illustrates the corresponding sensitivity patterns, as obtained from a MLR of regional ΔTXx onto grid-point drivers’ changes” but the caption says: “Sensitivities of summer changes in absolute maximum temperature (ΔTxx, 2070–2099 minus 255 1979–2014) associated with uncertainties in the responses of key drivers for each region determined using a multi linear regression model (see Eq. 1)”. These are not the same analysis. What does “grid-point drivers’ changes “ mean? I understood that the drivers were indices, not fields. If these are the regression coefficients of Eq 1 bx and cx, then the coefficents show how much of Δ𝑇Xx/GW is explained by each of the drivers. If this is the case, I don’t understand what the authors are trying to interpret from the map. For example, N3.4 is correlated with the temperatures in the region where the index is computed. This is clearly to be expected, but it has nothing to do with the temperatures in the study area.  In any case, it would make sense if the authors correlated the changes Δ𝑇Xx/GW onto SST at each grid point , OLR at each grid point, etc. This is done in https://doi.org/10.1029/2023JD038712 to identify drivers of uncertainty in their study area.

In my understanding, the following statement “Regarding CA (Figs. 3 e-f), the results show that GCMs with larger decreases in SMCA or more pronounced poleward shifts of the SACZ display exacerbated warming of TXx.“ is a correct interpretation of the maps if the statement in the caption is correct. But then the description of the figure “Figure 3 illustrates the corresponding sensitivity patterns, as obtained from a MLR of regional ΔTXx onto grid-point drivers’ changes” is unclear.

It would be relevant to present the robustness of the responses under the storylines. Could the authors provide a reference of which is the mean or median absolute error of the regression model? The coefficient of determination is presented, and this is useful to understand how much of the model spread is explain. However, the absolute value of the errors is a better estimation of how robust the storyline estimates are, if they are developed using the regression framework. With this I mean, if you try to reconstruct each model’s response with the regression framework, how good is this estimate? I understand that this was not done in Zappa & Shepherd (2017) but since then, for example in Mindlin et al. (2020) who you cite, this was done. If errors are an order of magnitude smaller than the difference between storylines, one can ensure that using the regression model to represent the storylines is appropriate. It is important to show the robustness of the analysis, otherwise the conclusions can be based on statistical artifacts.

References

Cinelli, Carlos and Forney, Andrew and Pearl, Judea, A Crash Course in Good and Bad Controls (September 9, 2020). http://dx.doi.org/10.2139/ssrn.3689437

Monerie, P.-A., Biasutti, M., Mignot, J., Mohino, E., Pohl, B., & Zappa, G. (2023). Storylines of Sahel precipitation change: Roles of the North Atlantic and Euro-Mediterranean temperature. Journal of Geophysical Research: Atmospheres, 128, e2023JD038712. https://doi.org/10.1029/2023JD038712

Narsey, S., J. R. Brown, F. Delage, G. Boschat, M. Grose, R. Colman, and S. Power, 2022: Storylines of South Pacific Convergence Zone Changes in a Warmer World. J. Climate, 35, 6549–6567, https://doi.org/10.1175/JCLI-D-21-0433.1.