Review of cp-2022-22: "Asymmetric changes of temperature in the Arctic during the Holocene based on a transient run with the CESM" by Zhang et al.

---------------------

Summary

This paper argues that there was an asymmetric temperature change between the Atlantic and Pacific sectors of the Arctic from the mid- to late-Holocene. The authors find this pattern in the temp12k global Holocene temperature reconstruction and also in transient climate model simulations with CESM. They argue that this is caused by orbital modulation of the Artic Dipole pattern and the Pacific Decadal Oscillation.

---------------------

Main comments:

The paper presents an interesting hypothesis with a lot of analyses to back up the main results. However, in places it seems like the results need to be better supported with evaluation of the uncertainties, while the link to the modes of variability may benefit from more elaboration. My main comments are as follows:

1) If I understood correctly, this Holocene simulation (AF) does not include changes in ice-sheet/sea-level forcing? It so, this might be an important caveat for the response in the Arctic. Although the global sea-level has stabilised by around 6 ka BP, this is right in the middle of the early-Holocene time window that you analyse throughout. I think some discussion of this is needed.

2) A more robust evaluation of the proxy-based signal is needed in section 3.1. The asymmetry is dependent on a relatively small number of points that show a stronger cooling in the Atlantic sector of Figure 1. If the coolest 2-3 of these were removed it looks like the asymmetry could likewise disappear. This makes one wonder whether the asymmetry is an artefact of the limited coverage by the proxies? Could you evaluate this in more detail? Perhaps add a histogram of the reconstructed temperature changes in the two regions? 

3)The reconstructed and simulated regional temperature anomalies are given to 2 decimal places which feels overly-precise. It would be more convincing if the estimated  uncertainties on these values were presented.

4) Assuming that the reconstructed asymmetry is robust to the choice of points it is not clear on first reading that the model actually replicates the 'asymmetric' temperature response in the annual mean as only the separate seasons are shown. Since the proxies are calibrated to reflect the annual mean signal I think it would be beneficial to show the annual-mean model result.

 

5) The analysis of the atmospheric dynamics is not easy to follow (see comments below) and it is difficult to understand precisely how the PDO/AD modes combine to produce the seasonal-mean signal in the sea-ice.

6) Changes in ocean circulation are not mentioned, but given they are important for the past 2000 years (Zhong et al 2018), it would be worth evaluating.

---------------------

Minor comments:

Line 102: Is the Glimmer ice sheet model used in this study or is it deactivated?

Line 103: I think you should cite Hurrell et al 2013, instead of this web link.

Line 109: It's not clear how the Gao et al reconstruction is used for the Holocene as in their paper they only discuss the last 1000 years. Please could you expand on this?

Line 113: I could not find Wan et al. (2020) in the reference list.

Line 138: This link does not appear to describe the Jonkers et al 2020 dataset or anything else that is mentioned in this manuscript.

Line 149: "... with red indicating an increase in temperature between the late and the early-mid Holocene (0-2 ka BP and 5-8 ka BP), while the blue indicating and decreasing."  This can be omitted.

Line 154-155: These values to 2 decimal places seem overly precise. Please could you estimate the uncertainty in these two values?

Line 173: again the regional average temperature anomalies should include uncertainties. I suspect 2 decimal places is overly-precise.

Line 206: This sentence starting "Many studies" makes it sound like these are all studies on the Holocene, but I believe that they are all focussed on the present-day. Please re-word to clarify this.

Line 223-227: "The difference in SLP between the two periods does show a similar dipole pattern, but combined with the stronger SLP in the late Holocene than in the early-mid Holocene shown above, it can be assumed that the stronger Arctic dipole in the late period had a greater role in influencing sea ice."

Perhaps I have missed something, but I don't follow this.

Lines 236-249: It's not clear how the regressed UV winds and sea-ice on PC2 are responsible for the climatological signal. I think this needs to be elaborated on.

Line 260: "The index indicates that negative PDO dominates the late Holocene, while the positive and negative PDO phases oscillate during the early-mid Holocene." 

This is not clear from the figure. Please can you provide a statistic that shows this.

Lines 265, 267: Please specify what you are comparing with this spatial correlation coefficient?

Line 280: Your results mirror findings of Zhong et al 2018. However, they invoked a significant role of the ocean circulation. Is that important in the present model results?

---------------------

Comments on the figures:

Throughout the labels on figures could be tailored for easier reading of the figures. As it is one has to read the caption carefully to understand what the multi-panelled figures are showing.

Figure 1: For clarity could you include in this caption whether this is late Holocene minus early Holocene?

Figure 3: I would like to see the annual-mean model result as the proxies are calibrated to this if I understand correctly?

Figure 6: It would probably be helpful to have the same y-axis limits on panels (c) and (d). Also, are the timeseries of the PC 2 smoothed?

Figure 9: is this the AF or the ORBIT-only simulation? Do they both look similar?

---------------------

Technical corrections:

Line 148: "while the blue indicating and decreasing."  Typo here.

Figure 10: The captions says EOF1 but the figure labels say EOF2. I assume they should both same EOF1?

---------------------

References:

Hurrell, J et al (2013). The Community Earth System Model: A Framework for Collaborative Research, Bull Am Met Soc, 94,9, https://doi.org/10.1175/BAMS-D-12-00121.1.

Review for the manuscript

Asymmetric changes of temperature in the Arctic during the Holocene based on a transient run with the CESM

by Hongyue Zhang et al.

Submitted for publication in Climate of the Past

General

The manuscript investigates Arctic temperature changes in an accelerated earth system model (ESM) simulation for the Holocene with CESM. The authors present asymmetric temperature changes between the Pacific and Atlantic parts of the Arctic and attribute those changes to varying pattern of atmospheric circulation and sea ice concentrations. Moreover, authors suggest that those asymmetric changes are especially pronounced in a simulation that is only driven with changes in orbital forcings.

The manuscript is unfortunately not representing the state-of-the-art literature and more important, lacks of simulations that are currently available for the Holocene in a transient sense. Accelerated simulations for the Holocene were expedient because of a lack of computing capacities some 20 years ago. Therefore according conclusions, especially on long term changes such as ocean-related sea ice processes can be afflicted with high uncertainties, also in the context of the interpretation with proxy data.

As such I cannot suggest publication of the manuscript in the present form. Below I list a number of suggestions and more recent studies including non-accelerated simulations that can be used for a substantially revised version of the manuscript.

Specific

In the following I will just point to the main concerns and how authors might extent and update their investigations taking into account more recent studies and adapting their hypothesis to more ESM/GCM-relevant questions.

Introduction:

The introduction lacks at least one paragraph motivating recent modeling studies over the Holocene, the challenges and implications e.g. of accelerated simulations vs. non-accelerated and the uncertainties involved in reconstructing external drivers (specifically solar and volcanic) for decadal-to-multi- decadal variability (cf. also studies listed as additional references below)

Another crucial and yet missing part is on the potential drivers giving rise to an asymmetric temperature response. Some mechanisms such as changes in equator-to-pole temperature gradient and/or changes in overall sea ice concentrations are presented. But no hypothesis or guiding question in how those general changes should result in regionally different responses are discussed.

2 Method and data

2.1 The CESM model and the transient simulations

ll. 106 ff: The authors describe their acceleration technique, also using changes in solar and volcanic output. I was wondering how those changes, reconstructed on yearly time scales can be implemented in a simulation with an acceleration factor of 10. (e.g. typically more than 2 volcanic eruptions happen per decade). How is this temporal discrepancy between annual reconstructions for accelerated simulations accounted for, also considering the post-volcanic effects on the simulated climate.

ll. 116 ff: There are new, and non-accelerated comprehensive Earth System model simulations available (cf. references) that should be used as additional source of information to back-up results based on the accelerated simulations with CESM.

Another general comment relates to the questions why the authors did not at least use an ensemble approach for their simulations to estimate the amount of long-term (centennial-to-millennial scale) climate variability.

2.2 Reconstructing Paleo Proxy data

This paragraph just lists the proxy data sets used for comparison without any information on potential uncertainties involved in the reconstructions, e.g. related to the uncertainties in the proxy archives towards their meteorological/climate variables, dating uncertainties, regional sparseness of proxy data, especially in the Arctic domain.

Since the authors investigate changes in ocean-related sea ice variability, also a paragraph on proxies representing changes in sea-ice concentrations including their uncertainties would be helpful.

3. Result

3.1 Arctic temperature change

ll. 152 ff: How robust are the temperature changes ? Are they statistically significantly different to internal changes. Therefore, applying a statistical test is helpful to estimate the amount of internal variability between the two different periods, preferentially taking into account the serial correlations within the proxy-based estimations of temperature variability.

ll. 172 ff: How significant are the changes between the Arctic and the Pacific region ? (i.e. -0.67 vs. +0.09.) Especially the Pacific trend seems to be statistically indistinguishable from a zero trend).

ll. 191 ff: Also for the model-based differences of the sea ice a local statistical test on the spatial pattern including the effect of serial correlation is important to test the robustness and statistical significance of the according changes.

ll. 202 ff: Changes in atmospheric circulation are also influenced to a high degree to internal variability – as such it is very important to use additional model simulations to back-up those changes, resulting from the CESM accelerated simulation. Moreover, why are the results of the orbital simulation are “more significant” than the one for the all forcings ? On Holocene time scales changes in orbital forcing on seasonal time scales exert a larger impact than the decadal-and sub-decadal changes caused by solar and volcanic activity. Therefore it is important to describe in greater detail how changes in solar and volcanic forcings are implemented into the accelerated CESM simulation.

3.3 EOF of SLP and UV wind regression and 3.4 The connection between Arctic dipole pattern and PDO

The whole sections lack a more thorough motivation on i) how the statistical concepts are used/defined and the ii) the robustness and statistical significance of the according regression patterns between the PCs and the underlying wind/sea ice fields. For instance, the PCs presented in Fig. 6 are (obviously) filtered with a low-pass filter. This should be accounted for when discussing and presenting the results.

Further, in addition to the UV regression, a Canonical correlation analysis would be better suited for this kind of investigation in section 3.3, since the rationale is to compare the common behavior of patterns (in this case the spatially resolved SLP and wind/sea ice fields.)

A last point is again on the validity and model-dependence of the results based only on the accelerated simulation with CESM. This is in my opinion the weakest but most crucial point of the study.

4. Discussion

l. 291: Authors should formulate more nuanced that in this very version of the manuscript, results only apply to their few accelerated simulations with CESM that need to be compared with more recent, non-accelerated studies.

l. 293: How should GHG changes, only changing very moderately in the pre-industrial period of the Holocene counteract any changes in orbital forcing ? If any, volcanic (and maybe in parts) negative periods of solar activity could counteract the negative trend in orbital forcing during the JJA season over the Arctic.

l. 284: The authors state that additional simulations should be used for investigations. Since those simulations are yet available authors should use them as an integral part of their revised study and thoroughly test their hypotheses with non-accelerated simulations and those carried out with different CMIP4-types of models.

Figures:

Fig 3.1: How does the Proxy (z-score) and the Model (°C) compare on the same axis ? In my opinion it would be necessary to show both on the same scale for an appropriate comparison.

Fig. 5: Please use units of hPa when presenting changes of sea level pressure fields.

Fig 6, 9a and 10a: In this form of the presentation, the EOF pattern seem to carry normalized values (i.e. z-scores). In order to re-normalize the EOFs (i.e. eigenvectors), the patterns should be multiplied with the square root of their eigenvalue. Then the EOF patterns carry the units (in this case Pa(hPa) for SLP and K for SSTs, respectively). Eventually the according (original) PCs should be divided by the square root of the eigenvalue in order to show consistent patterns between EOFs and PCs. In addition, the temporal filtering should be indicated for the time series.

Additional references / State-of-the art Holocene ESM simulations:

Transient Holocene simulation (6ka BP - 2ka BP) with interactive vegetation and phenology: https://vesg.ipsl.upmc.fr/thredds/catalog/work/p86mart/IPSLCM6/PROD/Holocene/TR6AV-Sr02/catalog.html

Braconnot, P., Zhu, D., Marti, O. and Servonnat, J.: Strengths and challenges for transient Mid- to Late Holocene simulations with dynamical vegetation, Clim. Past, 15(3), 997–1024, doi:10.5194/cp-15-997-2019, 2019

Braconnot, P., Marti, O., Crétat, J., Zhu, D., Sanogo, S., Balkanski, Y., Caubel, A., Cozic, A., Foujols, M.-A. and Servonnat, J.: Transient simulations of the last 6000 years with the IPSL model, in PMIP Workshop group P2FVAR., 2019.

Bader, J., Jungclaus, J., Krivova, N., Lorenz, S., Maycock, A., Raddatz, T., Schmidt, H., Toohey, M., Wu, C.-J. & Claussen, M., 2020: Global temperature modes shed light on the Holocene temperature conundrum. Nature Communications, 11: 4726. doi:10.1038/s41467-020-18478-6.

Dallmeyer, A., Claussen, M., Lorenz, S. J., Sigl, M., Toohey, M., and Herzschuh, U.: Holocene vegetation transitions and their climatic drivers in MPI-ESM1.2, Clim. Past Discuss. Clim. Past, 17, 2481–2513, https://doi.org/10.5194/cp-17-2481-2021, 2021.