The knowledge of the characteristics and mechanism for the climate change over Australian regions in LGM is still not enough. This study investigated the climate changes at the LGM over the Australian region, in terms of temperature, precipitation, moisture balance and wind, based on the output from PMIP3 and PMIP4 simulations. The work might contribute to our understanding of the hydrological change of Australia in ice ages. The following are my comments and reviews for the authors’ consideration.

Comments:

1. The uncertainties of model simulations and reconstructions are important information for model-data comparison. It would be better to evaluate the model consistence since there was large model spread. Specifically, the authors could further provide the percentage of model ensembles consistent on of the signal of their multiple model ensembles mean value.

For the reconstruction, the background information of proxy used here and their uncertainty could be listed in a table. The information of the LGM climate getting wetter or drier and in which parts of Australia based on proxies is still not clear, even though the authors cited others’ work in line 459-463. It would be easier to read and make comparison if were there reconstructed data mapped on the plots of model results.

1. Usually, modeling community use surface air temperature (SAT) instead of surface temperature (ts) to investigate the temperature change, and to explain the related change of circulation and/or precipitation.
2. The climate proxy from 28 to 18 ka is compared with the LGM simulations at 21ka (Line 55-56). This may also contribute to the model-data inconsistence considering the extended date of proxy. For example, the variability of climate proxy during 28 and 18 ka may switch between drier or wetter condition than pre-industry and thus make the complexity of model-data comparison. This point could discussed further when necessary
3. As pointed out by the authors, the sst gradients and related circulation change could explain the precipitation change (Line 309-311). Thus the analysis of sst change (which roughly equals to the ts value over ocean) and model-data comparison of sst could improve the knowledge of the LGM climate change over Australia.

Line by line comments

Line 42. “Many regions”, could be pointed in details.

Line 92-100. The reconstructed evidence of moisture or hydrocliamte could be compared with model simulations in the section of discussions.

Line 155-157. There were three different ice sheet reconstructions. Thus it’s necessary to clarify the information of ice sheet configuration of the four models from PMIP4.

Table 2. In term of vegetation of PMIP4, were there any model using the dynamic vegetation? Please check and make it clear.

Line 180-182. Usually models use the last 100 years, instead of the first 100 years, to do analysis. Were there any big differences between those two choices?

Line 213-214. The difference between the analyses in the paper with Kageyama et al. may lies in the choice of ts, instead of SAT. Please check.

General comments

In this study by Du et al., the authors propose a modelling intercomparison study of the simulated surface temperature, precipitation, moisture balance and winds at the LGM, focusing on the Australian region.

Such a study would fit well within the scope of Climate of the Past. The case study of Australia is interesting for several reasons such as the impact of sea level change on coastlines, the specific location (SH, in proximity to the Maritime Continent and Southern Ocean), and the overall disagreement of models and proxies especially concerning the latitudinal shift and variation of strength of the westerly winds, which seem both poorly represented in PMIP models (Chavaillaz et al., 2013) and poorly constrained by conflicting paleodata records (Kohfeld et al., 2013). The manuscript is also clear and an easy read.

However, as such (and as is often the case with intercomparison studies), the paper reads as very descriptive and superficial, so we are struggling to learn something new. Knowledge gaps, uncertainties in both model and data and processes/mechanisms are either barely mentioned or simply not highlighted enough so when we reach the end, the impression is underwhelming.

I am providing more concrete illustrations as to what could be improved and how with points 1-3. I also have suggestions related to methodology in points 4-8. So I would like to recommend a number of improvements before publication, in the hope of helping give this study more weight.

1. Please elaborate on the reasons why a case study of the past climatic changes over Australia is interesting. It would be great to mention climate processes that are key in this region. An example since dust transport is mentioned (L93): does this aridity have the potential to significantly enhance iron fertilisation in the ocean?

2. Consider adding one or two last sentences to the abstract and a paragraph in the discussion/conclusion to give the reader a broader perspective and hindsight on what we have learned and how significant are these new findings. A few questions to help brainstorm: So what? In the basis of the existing litterature and these new findings, have we achieved a better understanding of the processes which influenced the past Australian regional climate? If not, what are we missing? Do we understand the model response to LGM forcings? What does it entail?

3. Please underline the knowledge gaps in the introduction. As such, the introduction is very descriptive (not impactful), with a structure (global changes / changes in different regions of Australia) that doesn‘t help guide the reader very logically towards understanding the knowledge gaps, their importance, the scientific question and the methodology used in this study. If the authors would like to keep this regional description structure, then it would be welcome to also point out the contrasts between these different regions, and also with the global climate, with a few short sentences to conclude this subsection.

Still, I feel like the sometimes conflicting proxy records and the ‘uncertainty about the drivers of the LGM climate changes‘ (L74-75), the ‘ambiguous results‘ of models (L117) should be arguments brought to the reader‘s attention in a more convincing order to justify the need of this particular intercomparison study and its methods. For exemple, it is not clear what is the advantage of using an intercomparison method, nor for which reasons modellers simulate the LGM period (L102-107). It is also not clear why the authors chose to examine these three specific climate variables. I believe there are ways to reinforce the visibility of the scientific reasoning behind this approach.

4. Only PMIP4 outputs available on the ESGF were included. This is a bit of a shame because it limits comparison with the Kageyama et al. (2021) results, and the model ensemble size (and therefore the robustness of the results). CESM2 is also excluded for very good reasons (L143-144), but I believe the authors have found the source of the exagerated cooling in a cloud microphysics parameterization and run a corrected simulation. This is of course up to the authors, but they could consider contacting both the Kageyama et al. (2021) and Zhu et al. (2021) authors to request the model outputs. This would also enable the authors to compare individual model versions (CMIP5 vs CMIP6) and discuss potential improvements between the two generations. For now, only MPI-ESM is in the two ensembles.

5. The authors mentions that all model outputs are regridded (L170), and it seems that all the following analysis use these regridded outputs. Of course, this is necessary to compute and plot the multimodel mean, but I am wondering whether it would be worth extending the use of the model native grids when plotting the individual maps and wind profiles. Is the latitudinal shift of westerly winds affected by the model resolution? Are there differences in the land-sea mask of each model which could impact the simulated temperature and precipitation patterns? Please consider plotting the PI or LGM land-sea mask (e.g. as a grey or dashed contour) on maps.

6. On maps, the authors should also represent the multimodel agreement significance as hashes (when >90%) and the proxy data as scatter points, whenever quantitative reconstructions can be found. I consider important that the reader is able to compare visually the performance of the models with the available proxy data, especially as the authors conclude that there is a general good agreement between model and data.

7. The authors mention using the first 100 years of simulation for the analysis. Why not the last 100 years? Please check that the simulations are in equilibrium, e.g. by computing the drifts.

8. The paper would deserve more quantifications. An example is in L273: ‘Some models show weakening and other model show strengthening but there are other instances (e.g. L372). It would be great to provide precise figures, e.g. phrasings like ‘5 models out of 12 show a weakening of at least 20%...‘

Specific comments

L2, L12 and L13: ‘changes at the LGM‘, ‘to cool by 2.6 at the LGM‘ and ‚‘decreased‘. Unlike in the rest of the paper, it is sometimes unclear in the abstract that we are mentioning changes relative to the PI. This has to be indicated in some way (e.g. LGM-PI anomaly, with respect to PI…) or else the verbs are inconsistent with the time direction.

L13-14: Why are the changes in temperature and precipitation indicated over two different defined regions?

L18-19: I find the sentence explaining the potential reasons for model-data disagreement to be rather vague and could be reformulated.

L23: The time window proposed for the LGM is unusual and only justified later in the text.

L31: Why such a gap between the Annan et al. (2022) and the Tierney et al. (2020a) estimates?

Fig 1 and 2: These figures do not bring a lot of information to the table. The authors could consider combining them, combining Fig 1 with e.g. Fig 3 (the land-sea mask could be indicated on another map with a grey contour), or enriching them with more information (e.g. the Sunda and Sahul shelves mentioned in L34 could be annoted on the map to show the reader their location). As for Fig 2, consider using a different style (than the red contour, e.g. hashes) for the southwest box.

L85: The authors could elaborate on the reasons why they used the word ‘possibly‘.

L110: ‘not fundamentally different‘ could be true for the variables examined in the study and not others. Please check if this is the case for all variables (including ocean circulation).

Table 2 does not feel very necessary. The authors could consider moving it to SI to save some space.

L168: Why use both ts and tas?

Fig 4 / Table 3: What is the GMST simulated by these models? Are the models which are cold on a global scale also the ones simulating cold temperatures over Australia?

Table 3: Tables are not great to visualize data (also true for Table 4 and 5). The authors could consider a different type of plot to show the reader the large intermodel differences in the seasonal amplitude (not commented in the text?) in a single glance.

L250/L269: Please consider using transitions between subsections (here as in other instances). It is the opportunity to remind the reader of your scientific reasoning (e.g. how these variables are linked).

L272: I am wondering whether Fig S1 which shows very large model disagreements should not be part of the main text. Also, please explain why you chose to plot the JJA season specifically.

L304-305: ‘In JJA the SH westerlies shift equatorward‘. Would it be worth it to also investigate the seasonal shift of the westerlies in Sect. 3.2?

L397-398: Does this correspondance hold for all models?

L412-413: Does this relationship hold if you use the change in strength of westerly winds over the same region?

L463-465: I will make a subjective comment here. While this is a valid reason to criticize the proxy records (well-explained in introduction), I feel like modellers should maybe not be too critical of proxy uncertainties when such large intermodel differences are observed. The primary reason why we are observing this model-data disagreement might be that, well, models are wrong. I will also point out here that the discussion and especially the conclusion seem lenient with models. I would expect the large intermodel difference observed to reflect a poorly-represented process.

L467-470: Could you discuss the potential reasons why you can find a displacement of the boundary line but no consistent latitudinal shift in westerly winds? Do you have any idea?

Fig. 11: Consider different marker styles or colors for individual models or model generation (CMIP5/6).

Fig. 6: It would make sense to put MPI-ESM-P (CMIP5) and MPI-ESM-LR (CMIP6) in the same column on Fig 6 so that it is easier to compare the two generations visually.

Technical corrections

L16-17: ‘many regions‘ is unclear

L26: ‘glaciers‘ instead of ‘glaciation‘

L43: ‘reflect‘ or ‘are associated with‘?

L44: Consider replacing ‘related to combinations‘ with something like ‘caused by a combination‘

L45: 190 ppm in Table 2

L110: ‘drier conditions‘ would work but not ‘drier changes‘

L121: ‘differences‘ or ‘gaps‘ would fit better than ‘variations‘

L124: ‘more recent‘ instead of ‘newer‘

L144-145: ‘Furthermore, the PMIP4 protocol highlighted…‘

L170: 10 m

L328: What is the Top-End region?

General comments

Climate change in the Southern Hemisphere is poorly understood, and large model biases are known to exist. Studying how climate has changed at the LGM may provide unique insights into the climate dynamics of this region. This manuscript investigates changes in temperature, precipitation, and wind over Australia at the LGM using a subset of PMIP3 and PMIP4 models and compares these changes to existing proxy data. Such a study could be helpful in improving our understanding of Australian climate.  

The analysis is generally okay: the authors looked at the climate response in individual models, ensemble mean, and seasonality. However, I think the authors could have added some more in-depth analysis or discussion. One thing they can do is to expand the inter-model agreement (hatching the maps of ensemble mean could be helpful), and consider how model disagreement may affect the ensemble mean values.

I also think that the mechanisms for changes in temperature, precipitation, and wind are not adequately discussed. Please see my specific comments.

In addition, I think the authors should do their due diligence to acquire model output from all PMIP4 models.

In terms of presentation, the manuscript is structured logically. But the color scales for showing hydroclimatic anomalies could be improved such that the map colors are not overwhelmed by the changes at the coast to make it easier to see changes over the continent. And a better integration of data-model comparison could be achieved by showing the proxy-reconstructed changes in the map of simulated changes.

Specific comments

The Abstract ends abruptly by describing changes in winds, whereas here it should provide the readers with some key implications or take-home message of this paper.

47: Ujvari et al 2018 is not an appropriate reference, as it does not talk about changes in dust at the LGM.

61: Many of these referenced papers did not use PMIP4.

66: You did not mark these regions discussed here in Figure 2. Maybe use consistent terminology here as the rest of the paper.

74: Reference for the fire study?

77: You cited a wrong Denniston et al (2013) paper. The correct one is:

Denniston, R. F., Wyrwoll, K. H., Asmerom, Y., Polyak, V. J., Humphreys, W. F., Cugley, J., ... & Greaves, E. (2013). North Atlantic forcing of millennial-scale Indo-Australian monsoon dynamics during the Last Glacial period. Quaternary Science Reviews, 72, 159-168.

Note that in the paper you cited, the C126 speleothem shows more positive d18O and d13C values at LGM than the late Holocene, which might suggest drier glacial conditions.

143: This statement is incorrect: Zhu et al. (2021) only assessed CESM2-CAM6, the “low top” version of CESM2, not the WACCM version.

156: Do these different ice sheet configurations affect the Australian climate at LGM? Did you use them in your study?

180: Why do you choose the first 100 years? Models need time to reach new climate equilibrations in response to external forcings. I would use the last 100 years if possible at all.

185: specify it is austral summer/winter. I also think this is where you can describe the regional climate systems in more detail. i.e., winter precipitation in the south is associated with the westerlies, summer precipitation in the north is associated with the monsoon.

241: If “land areas warm more than surrounding oceans” during DJF and SON is the case, why DJF and SON show opposite signs in temperature change over Sahul? Are there other mechanisms that could cause this change in temperature?

245-250: How do these analyses relate to your results in Figure 5? If there is enhanced cooling in SON and reduced cooling in MAM, why Fig 5 shows more cooling in MAM and less cooling in SON?

311: What is this “SST gradient”?

395-396: This statement does not make sense. Fig 5 shows DJF cooling and SON warming over northern Australia, why does it case wetting in both seasons? What is the “response to changes in seasonal heating” and “changes in atmospheric circulation” here?

414: p = 0.082 suggests that the correlation is not significant or “moderate” – it is insignificant. By the way, I wonder how do changes in precipitation and the northward displacement of easterly-westerly boundary correlate.

According to your findings, what is the mechanism for changes in winds?

Technical corrections

268: You don’t need a 3.2.1 subsection here

323: Figure S4 is MMM seasonal anomalies for LGM - PI evapotranspiration, not precipitation.

397: to the => to the

403: should be 3.3.2.2