First of all, we would like to thank the reviewer for their review of our manuscript, which has helped us to address issues and improve our conclusions. In this rebuttal, we would like to address their concerns. The reviewers’ comments are shown in bold, while our answers are shown in regular font-type.

Review of Scherrenberg et al., “CO₂ and summer insolation as drivers for the Mid-Pleistocene transition,” submitted to Climate of the Past.

The authors present modeling results of the Mid-Pleistocene Transition using a simple ice-climate model. This is a key approach for exploring the mechanisms behind this major climate transition, which remains an unsolved question in paleoclimatology. Their model is forced with caloric summer insolation, and an indirect atmospheric CO₂ reconstruction derived from leaf wax spanning the last 1.5 Ma. In addition to baseline simulations, the authors conducted additional simulations with either constant atmospheric CO₂ concentrations or constant insolation to disentangle the impact of each forcing on the climatic cycles of Pleistocene.

Overall, the manuscript is well-written and structured, although the second part is somewhat challenging to follow due to a mix of the results and discussion. The figures are well chosen and informative but would benefit from more specific description rather than broader summaries. I do have some questions about the way the authors interpret their results. Specifically, I find the boundary between the conclusions derived from their baseline simulations, the ones derived from their constant_ simulations, and the assumption based on paleoclimatic records very blurred. In addition, the use of the Yamamoto’s CO₂ record as a model forcing, that did not register any glacial CO₂ decrease through MPT, is not obvious to link with one of the main conclusions of the authors (i.e. the role of glacial CO₂ concentrations in the trigger of the MPT).

Major comments:

(1) The authors used the Yamamoto CO₂ record as forcing for their model. As it is the only continuous CO₂ record through the MPT, this choice is perfectly understandable. Nevertheless, the results derived from the model are per definition dependent on this record and this should be mentioned. A second point, probably the most important one, is the interpretation made of this record by the authors: “Yamamoto et al. (2022) (…) find a decrease in glacial CO₂ concentrations across the MPT”. I strongly disagree with this statement. Surprisingly, Yamamoto et al. found constant CO₂ glacial concentrations through the MPT, and a gradual increase in interglacial CO₂ concentrations. All along the paper, the authors suggests that a glacial decrease in CO₂ concentrations would have skipped some deglaciations during the MPT and the late Pleistocene: the baseline experiment performed using the Yamamoto’s record seems to demonstrate that it is not necessary to involve a decrease in glacial CO₂ concentrations to « reproduce » the MPT. It would be interesting if the authors could clarify the relation between the use of Yamamoto’s record as a forcing and their conclusion that a decrease in glacial CO₂ concentrations would have cause the MPT.

To address the first point:  We indeed chose the Yamamoto et al. (2022) record, as this is currently the only continuous CO₂ record that covers the past 1.5-million years (which will be explicitly mentioned in the introduction). The Yamamoto record does have lower Early Pleistocene glacial/interglacial CO₂ levels compared to both proxies and carbon cycle modelling (e.g., Chalk et al. 2017; Wileit et al., 2019). This will be addressed in the discussion, and it may also partially explain why we obtained too large Early Pleistocene ice volumes. In the discussion section, we will also explicitly address that our results are indeed dependent on the Yamamoto et al. CO₂ record.

To address the second point, we will make several changes to the manuscript to improve the support to the main conclusions:

In our simulations we found that if CO₂ and insolation are strong, it can trigger a termination. This threshold also depends on the ice sheet, as a larger ice sheet is more prone to collapse when temperatures rise. Insolation maxima fail to trigger terminations if low CO₂ levels are maintained throughout the interglacial period.

In the Early Pleistocene, obliquity maxima tend to coincide with relatively high CO₂ levels (~240-250 ppm), allowing for terminations. However, in the Late Pleistocene, low interstadial CO₂ levels are sometimes maintained through insolation maxima, preventing deglaciation. This phenomenon is why we referred to “decrease in glacial CO₂”, but we follow the reviewers’ suggestions to rephrase this line to “lowering interstadial CO₂”, or more specifically: “low CO₂ is maintained throughout the insolation maxima, thereby preventing deglaciation”. Additionally, the constant_CO2 experiments also show that maintaining a low CO₂ level throughout insolation maxima can increase glacial cycle periodicity.

We will also remove the erroneous statement “Yamamoto et al. (2022) (…) find a decrease in glacial CO₂ concentrations across the MPT”, and revise the discussion and results sections to clarify the aforementioned model behaviour.

(2) This second point is related to the first one: it is unclear which conclusion is derived from their baseline simulation, from their constant_ simulations or from literature. For instance, in the case of the CO₂, the two approaches are opposite: in the baseline simulation, the authors forced their model with a glacial-constant CO₂ record while the comparison of constant_simulation would indeed allow to discuss the decrease of glacial CO₂ concentrations (but not only), even if these experiments are theoretical.

The manuscript will benefit from a clearer distinction between the proxy-forced and theoretically-forced simulations. We will revise and restructure the discussion section to improve these conclusions.

First of all, we will make a clearer distinction whether a statement or conclusion is derived from the proxy CO₂ record, the constant_CO₂/insolation simulations or literature. (for example, we will use “low constant CO₂ levels” rather than “low CO₂ levels” when discussing the constant_CO2 experiments). Similarly, we will specify each (relevant) mention of CO₂ whether it concerns interglacial/glacial or interstadial levels.

Secondly, we will rewrite and restructure parts of the discussion section. Parts of the results section (e.g, line 221-224 and line 264-270; see our answers to major comment (5)) will be removed or moved to the discussion section and rewritten, as these contributed to the unclear support to our conclusions. We will explicitly state that the constant_CO2 experiments do not distinct between glacial and interglacial CO₂ levels. We will make a clear distinction between the conclusions derived from the baseline simulation, the constant_CO₂ and constant_insolation experiments in both abstract and discussion section. These three "types" of simulation will be discussed in three separate paragraphs in the discussion section.

(3) The results derived from the constant_insolation are interesting. Nevertheless, I wonder to what extent these results (the absence of climatic cycles in the pre-MPT world) depend on the chosen  insolation value (set as present-day by the authors). A comparison of different insolation values, similar to what was done for the CO2_constant experiments, would strengthen these conclusions.

For the revised manuscript, we will add three additional simulations. Firstly, the constant_insolation_10kyr_ago (a Northern Hemisphere insolation maximum), which has glacial cycles with small ice volume amplitude. Secondly, we will add the constant_insolation_25kyr_ago (an insolation minimum) which has very long glacial periods.

Lastly, we will add the constant_insolation_5kyr_ago. This simulation has a slight increase in constant caloric summer insolation compared to the present-day insolation, but captures all major termination events during the past 1.5 million years.

The constant_insolation_5kyr_ago generates the periodicity and amplitude of the Early Pleistocene. It does reasonably well for the Late Pleistocene as well, though has relatively long interglacial periods and a slightly reduced glacial-interglacial amplitude compared to sea level reconstructions.

We will therefore weaken the conclusion “Early Pleistocene are dominated by orbital cycles, while late Pleistocene is dominated by CO₂”, as it is possible to mostly capture the full glacial-interglacial periodicity of the past 1.5 million years with just prescribed CO₂. Though, note that past CO₂ levels were (indirectly) affected by long-term insolation changes, thus the ice volume evolution of CO₂-only-forced simulations can still match orbital cycle periodicity.

We will add these three simulations to figure 6 (ice volume vs climate) and figure 7 (constant_insolation time-series).

(4) The authors have chosen to use a constant sediment map throughout the MPT, thereby excluding the possibility of testing the regolith hypothesis. While I understand this decision, as testing all MPT hypotheses is challenging and time-consuming, I strongly suggest that the authors exercise greater caution in interpreting their results : “Using these driving forces, we are able to capture the MPT without any change in (…) basal friction”. Simulating the MPT is not a binary process where one either succeeds or fails. While I agree that their simulations effectively capture the change in frequency, there is no guarantee that adding a variable sediment mask wouldn't improve the simulations. In my view, if the study still does not include a varying sediment mask, it cannot address the regolith hypothesis.

We have recently conducted the “sediment_change” simulation, which has reduced “sediment” friction during the Early Pleistocene (1.5 – 0.8 Ma). In the sediment_change simulation, we replaced the basal friction map with a new map that treats the entire domain as if it is covered by sediment. This simulation has a ~10% reduction in amplitude during the Early Pleistocene, mostly due to a lower average thickness of the ice sheets. Additionally, the termination at MIS 21 (~865 ka), which is skipped in the baseline, is modelled when using lower friction. This is because the lower friction makes the ice sheet more likely to collapse at insolation maxima (which is of course the method behind the regolith hypothesis). This sediment_change simulation will be added to the supplementary information.

We do believe we should mention that we were able to model the frequency change with only CO₂ and insolation, though also highlight that the model results improve slightly when applying lower basal friction. Though this should be addressed in the discussion section. Line 205 and line 108 will be removed.

(5) I found the organization of the paper, particularly paragraph 3.2 in the results section, somewhat confusing. The discussion within the results section also contributed to this confusion. For example, the section concludes with one of the study's most significant findings: “A gradual decrease in glacial CO2 levels could therefore explain the MPT.” This conclusion is not a direct observation of the result. As a result, I believe the discussion section is somewhat brief and would have benefited from a more thorough comparison of the authors' results with recent studies that have also aimed to simulate the MPT. For instance, Willeit et al. (2019) discuss the role of CO₂ using multiple scenarios of unconstrained CO₂ forcing; Legrain et al. (2023) draw similar conclusions to the authors using conceptual modeling; and Verbitsky et al. (2018) present interesting findings based on a physical approach of the MPT.

We will rewrite, restructure and expand the discussion section. Several lines of paragraph 3.2 will be moved to / merged with the discussion section. To give a brief summary of the changes to the results section:

• Lines  221-224 will be removed. In these lines, we address our lack of an active carbon cycle, which is already addressed in the discussion section. Additionally, line 221 contains a confusing statement "CO₂ levels have continued to decrease", which will be removed.
• At the end of section 3.1 (see line 264-270), we explain how CO₂, insolation and ice sheets relate to glacial-interglacial periodicity. This will be moved to the discussion and revised.
• The two sentences at line 282-284 will be removed.
• Lines 303-304 will be removed.
• In section 3.2 (lines 318-323), we compare our precession cancelling findings to Tzedakis et al. (2017). This will be moved to the discussion section.

Lastly, we will add comparisons with Legrain et al (2023), Verbitsky et al (2018) and Willeit et al (2019) to the revised manuscript. The former two used conceptual models and have studied glacial-interglacial periodicity through a shift in climate / ice sheet sensitivity. Legrain et al (2023) were able to capture some characteristics of the MPT by driving the model with only orbital forcing, which will be a very interesting comparison to our findings. Willeit et al (2019) used an Earth system model of intermediate complexity, and thus modelled the interactions between ice, climate and the carbon cycle more explicitly.

Minor comments:

Line 8: Provide the approximate timing of the MPT.

The approximate timing (1.2–0.8 million years ago) will be added to line 8.

Line 17:  Here and after you only mention the frequency. Why don’t you discuss the amplitude change that should be observed during the MPT?

We mostly focused on capturing the periodicity and matching the δ¹⁸O record. However, our lack of substantial sea level amplitude change warrants a broader discussion. We propose a number of changes:

• Whenever we mention “capture glacial-interglacial variability” (e.g., line 16) we will instead replace it with “capture glacial-interglacial periodicity”, to reflect that we captured specifically the periodicity.
• We will add a 1.5-million-year sea level record to figure 4 by Rohling et al. (2021; Science Advances).
• In the discussion, we will suggest possible reasons that could partially explain our lack of amplitude change. These reasons are: (1) our relatively high friction in the Early Pleistocene and (2) the perhaps underestimated CO₂ levels in the Yamamoto et al. (2022) record.

Line 19: Here and after: when you are talking about low CO₂ levels, are you talking about low glacial or low interglacial CO₂levels ? Or both ? Please be more specific to help the reader to correctly understand your findings.

We will specify for each (relevant) occurrence of CO₂ whether they are interglacial, interstadial or glacial. Similarly, we will specify each time whether we refer to constant or proxy CO₂ levels.

Line 21: Considering a glacial climate as a “relatively stable climate” could be questioned.

This can indeed be phrased differently. This statement specifies that this ice sheet configuration (where Laurentide and Cordilleran merge) leads to relatively strong self-sustained growth. A larger (merged) or small ice sheet (separated) state are more sensitive towards increases in insolation or CO₂. We will replace the statement “generate a relatively stable climate” with “A medium sized ice sheet is less sensitive towards insolation or CO₂ increase through its location and the merger of the Laurentide and Cordilleran ice sheets”.

Line 26: I think this is not surprising as the CO₂ is part of the forcing index. From which simulation do you derived these results ? Baseline or the comparison between the constant_ CO2 ?

These results concern the different constant CO₂ concentrations. We will make sure that this distinction between baseline and constant_CO2 simulations are clearer throughout the manuscript. Line 26 will be removed (to shorten the abstract) and instead replaced by a statement that constant_CO₂ levels did not yield persistent 100 kyr periodicity.

Line 29: Be cautious with this kind of assessments: “The MPT can be explained by”, especially in this case: I don’t see which simulations can led you to this conclusion, as Yamamoto’s record does not register any glacial CO₂ decrease and the constant_CO2 simulation does not make any difference between the interglacial and glacial CO₂ decrease. I understand that the abstract is not the appropriate for arguing this, but I think this is a crucial point that deserve more explanation in the discussion.

This sentence (line 29) will be removed, and we will avoid a statement such as “the MPT can be explained by”. Instead, the last sentences of the abstract will emphasise that the terminations are triggered if CO₂, insolation and ice volume are favourable for a termination. If low CO₂ is maintained during an insolation maximum, it may prevent a termination, thus increase glacial-interglacial periodicity. This is not the same as glacial CO₂ decline, and should not (and will not) be treated as such.

Lines 45-52: The three concepts (ice sheet, regolith, and CO₂) are nicely explained, but I wouldn’t put the first one at the same level as the two others: regolith and CO₂ would be a primary cause of MPT, and the ice sheets are more a secondary response to an initial shift, of what I understand.

The ice sheet threshold regime hypothesis should indeed be viewed not as the prime cause of the MPT, but rather as feedback processes affecting glacial-interglacial periodicity. We will state that the threshold regime hypothesis can “partially explain the MPT” and will also add the following sentence:

“While this theory requires another process (e.g., long term cooling) to prompt a change in ice volume, it can be a feedback process that facilitates the MPT.”

We hope that these changes reflect that the ice sheet is a feedback process, but is still very important towards the MPT.

Lines 78-80: Here and after: The most important thing to mention when comparing CO₂ record from ice core to other records is the fact that CO₂ is directly measured in the air trapped in the ice, while other CO₂ records used several transfer function, physical and chemical assumption to go from the initial proxy to the final CO₂ record.

We briefly discussed the difference in uncertainty, but indeed did not mention the challenges of proxy CO₂ reconstructions compared to ice cores.

We will add a few sentences to lines 78-80 to explain this difference between ice core (direct) and proxy records (indirect measurements).

Line 83: I would mention here that it is the only continuous CO₂ record across the MPT, and it is associated with a new and still discussed proxy of CO₂ concentrations. The apparently well correlation over the 800-0 ka with ice core record comes from the fact that the amplitude of CO₂ concentrations was calibrated on the ice core record itself.

Indeed, this is also the main reason why we chose the Yamamoto et al (2022) record, as our transient 1.5 Myr simulations require a transient CO₂ record.

We will add the statements “published the first continuous CO₂ record covering the past 1.5 million years” and “and calibrated to the 800 kyr ice core record” to this part (line 83) of the introduction.

Line 87: Why do you only focus on frequency and not on amplitude? The change of these two parameters is equally crucial in the definition of the MPT.

During the development of these simulations, we mostly focused on capturing the frequency, rather than capturing amplitude. Additionally, we tuned the model to match the δ¹⁸O and sea level of the past ~330 kyr, as sea level, δ¹⁸O and CO₂ record at that time are well known.

However, in the results (section 3.1) we only briefly mentioned our too large amplitude. We will thus address this in more detail and compare it to a sea level reconstruction (Rohling et al., 2021; Science Advances). This record will be added to figure 4 and figure 8.

There are also possible reasons that could (partially) explain our large Early Pleistocene amplitude. First of all, when applying lower “sediment” friction, the amplitude of Early Pleistocene glacial cycles is reduced by ~10% (the sediment_change simulation, which will be added to the supplementary information). Secondly, boron isotopes and carbon cycle modelling suggest that the Yamamoto et al (2022) glacial CO₂ may be underestimated. If that is the case, this would lead to an underestimation of temperatures, yielding larger ice sheet, and thus partially explaining the large Early Pleistocene amplitude. These two possible reasons will be addressed in the discussion section.

Line 107: I understand that everything could not be tested in a single study, and thus the authors choose to keep a constant sediment map. But I would not argue that this approach would help you to solve the answer: does the MPT can be captured without any changes in basal friction? Because the way you will capture the MPT is not a yes/no answer. Reversely, I would say that it would be very interesting to quantitatively described what are the results improvement using a variable sediment mask rather than a constant one (as the approach made by Willeit et al. 2019, science Advances). If there is no significant gain using a variable mask, then you can mention that your model suggests that a variable sediment mask is not relevant to better capture the MPT.

We will add a simulation (sediment_change) where we apply “homogenous sediment friction” from 1.5 to 0.8 million years ago. Essentially, we replace the friction map with a map that treats the entire domain as if it is filled with sediments. This simulation is similar to the baseline except for a reduction in glacial-interglacial amplitude of ~10%, and a full termination during MIS21 (~865 ka; which is not the case in our baseline). This simulation will be added to the supplementary information and briefly addressed in the results section (3.1). The sediment_change simulation thus yields a slight improvement.

Line 167: I appreciate the efforts of the authors to perform an additional baseline_icecore experiment but I am not sure if this experiment is relevant as the two records are similar over the 800-0 ka period, due to the tunning performed in Yamamoto et al. (2019) on the ice core record.

This simulation was mostly conducted to show that there is some (but not substantial) difference between the two simulations. But it is true that these provide little to the overall conclusions of the paper.

We have decided to move these experiments to the supplementary information and change the text in section 3.1 accordingly. The simulation will also be removed from figure 6 (climate forcing compared to ice volume at terminations).

Lines 173-175: There is a structure problem in the sentence.

We will fix this grammatical mistake.

Line 183: You go a bit too fast in the description of your experiment: The amplitude of climate cycles are not significantly different before and after the MPT. I think it is a limitation that deserve to be discussed further.

We will make two changes to address this issue:

• We will add the sea level reconstruction by Rohling et al. (2021; Science Advances) to figure 4. This also enables us to introduce this limitation in more detail in the results section.
• We will address two possible reasons behind our large modelled sea level amplitude in the discussion (the low CO₂levels in the proxy record, and no basal friction change).

Line 204 and after : The section 3.2 is a combination of results and discussion. Understanding what is deduced from a direct observation is difficult.

To address this issue, we will move parts of section 3.2 and section 3.3 to the discussion. The discussion section will be restructured and rewritten to accommodate these changes.  We have addressed a list of changes in major comment (5).

Line 205: Same remark as for line 107: it is not needed to insist on this point as an argument.

This line will be removed.

Line 219-220: I do not understand from which result this assumption comes from.

The “decrease in glacial CO₂ concentrations” should have been “interstadial CO₂ levels” (or even better, "maintaining low CO₂ levels during insolation maximum"). These levels can be most clearly seen in figure 4 at 737, 175, 50 kyr ago). We will change this sentence to reflect this.

Line 265: Are we still talking about glacial CO₂ levels ? Or is it an averaged CO₂ concentration level over an entire climate cycle ?

This concerns CO₂ levels during insolation maxima.

This entire paragraph (264-268) will be moved to the discussion section and rewritten. We will be more specific throughout the manuscript whether CO₂ levels concerns interglacial, glacial or interstadial.

Line 276: Why do you chose the present-day insolation value ? Intuitively I would have taken the average insolation value over the past 1.5 Ma.

We chose present-day for two reasons:

• It is more-or-less in the center of the insolation of the past 1.5 Ma. ~5.79 GJ/yr (0 ka) compared to ~5.84 GJ/yr (average).
• It provides a “realistic” monthly / latitudinal insolation (as this is relevant for the surface melt).

In the revised manuscript, we will also add simulations with constant “enhanced” (5 kyr ago), insolation minimum (25 kyr ago), and insolation maximum (10 kyr ago) summer insolation. This should provide a larger range of constant_insolation values.

Line 276 and after: This result is interesting, but how would sensitive is it from your chosen insolation value?

We have now conducted three additional experiments and found that the results are dependent on the insolation value: The constant_25ka_insolation (insolation minimum), which gave very few terminations. Constant_10ka_insolation (insolation maximum) generated small ice volumes throughout the simulation.

We also conducted a third simulation: constant_5kyr_ago_insolation, which has enhanced summer insolation compared to present-day. This simulation generates all termination events during the past 1.5 million years, captures the Early Pleistocene ice volume amplitude, but generates long interglacial periods during the Late Pleistocene. Whether the terminations are triggered depends on the interglacial CO₂ levels in the Yamamoto et al (2022) record. The record has low interglacial CO₂ levels in the Early Pleistocene (~250 ppm), so it needs relatively strong insolation values to trigger a termination. Only a small range of constant summer insolation strengths can capture all these termination events.

The Yamamoto et al (2022) CO₂ record alone can therefore generate the glacial-interglacial periodicity of the past 1.5 million (given a narrow band of summer insolation strengths). We will therefore make changes to the abstract, results and discussion section to include this new result.

Additionally, we will clearly mention in the discussion section that past CO₂ levels were affected by insolation. Thereby, the constant_insolation simulations can still be (indirectly) paced by the orbital cycles.

Line 288 – 289: Do you perform several simulations (increasing by 10ppm CO₂ concentrations) and pick up  a posteriori three representative ones, or did you design a priori the experiments with these three specific values for any reason ?

Prior to the full 1.5-million-year simulations, we conducted the first few 100 thousand years for simulations with 210 – 250 ppm with 10 ppm increment. We selected three that showed strikingly different behaviour and continued these simulations until present-day. Only selecting three therefore provided the most amount of information with the lowest number of simulations.

Line 304: This assumption clearly could not be part of the result section but should be included in a broader discussion. Also, I would be careful with the use of “explain the MPT”.

Lines 300-304 will be removed. We will remove (and avoid) statements such as “explain the MPT”. 

Line 320: I am not sure to understand how you came to this conclusion.

While precession is part of the forcing, this signal is completely filtered out in the resulting ice volume. This is caused by non-linear response of the ice sheet towards climate forcing (i.e. the slow build-up and fast self-sustained melt when melt is initiated). This statement will be moved to the discussion, and we will clarify our explanation.

Discussion section: Some of the results are not contextualized with recent studies that have attempted to model the MPT (e.g. Willeit et al. 2019, Legrain et al. 2023, Verbistky et al. 2018). Some of these articles would reinforce and support the authors' conclusions, or else highlight interesting nuances and diversity in the MPT modeling results.

These studies are good suggestions. Legrain et al. (2023) has conducted (amongst others) orbital-only forced simulations. They also included ice-volume threshold behaviour to their conceptual model, thereby making this study relevant to compare to our results. Willeit et al. (2019) included active ice sheets, climate and carbon cycle in their simulations, while we use prescribed CO₂ changes instead. We will add a comparison to these studies to the discussion section.

Line 332: “CO₂ is high enough”: would you detailed what you are talking about specifically? It is not obvious that CO₂ levels of Early Pleistocene were higher than during late Pleistocene. Especially Yamamoto’s record proposes that interglacials CO₂levels are higher during late Pleistocene.

We will make sure that statements such as “CO₂ is high / low enough” will become more specific. In this case, it should have been “if the CO₂ during the insolation maximum is sufficiently high, it can lead to a termination”. This issue is also present in other parts of the manuscript (e.g., abstract, discussion), which will be resolved.

Line 333: “CO₂ is too low”. If this observation comes from one of your simulations, please specify which one (baseline or constant_ CO₂). If it comes from the paleodata record, please quote the paper from which you get this information.

This specific point refers to both the baseline and constant_CO2. In both cases, if interstadial CO₂ is low, it can prevent additional warming during the insolation maximum, thereby prevent deglaciation and increase the modelled glacial-interglacial periodicity.

As this is an issue throughout the paper, we will make sure that it becomes clearer which conclusion was obtained from the constant_CO2 or proxy record (Yamamoto et al., 2022). This will mostly involve changes in the abstract and discussion section.

Line 334-335: I would be careful about the conclusion coming from these CO₂ _constant simulations. A decreasing CO₂ trend throughout the MPT is not similar to successive simulations with constant CO₂ levels, but at decreasing values.

The main conclusions that should be drawn from the constant_CO2 is that there are fewer terminations when the constant CO₂ levels decrease. This causes increased glacial-interglacial periodicity, though no “true” 100 kyr periodicity is established. We will focus on this point, though also be more specific that this concerns constant CO₂ levels, where both interglacial and glacial levels are the same.

Line 336: I am lost: “CO₂ levels have continued to decrease”: what are you talking about ? Your simulations ? Which one? Or paleodata records ?

This refers to a CO₂ concentration that would allow a medium-sized ice sheet to survive during one insolation maximum, but could also yield a collapse of a large ice sheet at the next insolation maximum. The minimum CO₂ level in the record (and peak ice volume in reconstructions / our baseline simulation) is often reached just prior to the termination. Larger ice volume makes the ice sheet more prone to collapse when temperatures increase. Therefore, despite low CO₂ levels, the ice sheet may still collapse (which is also something we find in the 210-ppm simulation). As this sentence is confusing, we decided to remove it and instead dedicate a few sentences in the discussion section to explain this idea.

Line 364: Please mention the fact that it is an indirect method to reconstruct CO₂ concentrations and not a direct measurement.

We will modify the sentence to reflect that CO₂ is measured directly from ice cores, but reconstructed indirectly from leaf wax proxies. This will also be more elaborately explained in the introduction.

Line 368: I strongly disagree with the statement that Yamamoto et al. (2022) and Hönisch et al. (2009) find a decrease in glacial CO₂ concentrations. Yamamoto et al. (2022) find constant glacial CO₂ concentrations and gradually increasing interglacial CO₂concentrations through MPT. Regarding Hönisch et al. (2009), the authors conclude their abstract as following: “atmospheric CO₂ did not decrease gradually as would be expected were it to be the driver of the transition.”. Nevertheless, it is true that more recent boron isotopes CO₂ reconstructions propose a gradual decline of glacial CO₂ through the MPT: you would refer to Chalk et al. 2017 (see Fig. 4) rather than Hönisch et al. (2009). The fact that Yamamoto’s CO₂ record does not evidence any decline of glacial CO₂ concentrations is quite problematic for one of the conclusion of this study (a decline of glacial CO₂concentrations would have trigger the MPT), as it is the record used as a forcing in the baseline experiment of the authors.

Yes, this should have been explained better, and we made a mistake here. To address these issues, we will make changes throughout the paper.

First of all, the erroneous statement in line 368 will be removed.

Secondly, we will improve our explanation that we obtain longer glacial cycles because low (interstadial) CO₂ levels are maintained through the insolation maxima. As such, these low CO₂ levels prevent additional warming during the insolation maxima, thus preventing deglaciation.

Thirdly, whenever relevant, we will clearly indicate if we refer to interglacial, glacial, or interstadial CO₂ levels. And we will specify whether we derive a conclusion from proxy or constant CO₂ levels.

Fourthly, we will rewrite and expand the discussion section to improve the support towards our conclusions, and comparisons to literature (including referring to Chalk et al. (2017) instead of Hönisch et al. (2009) when discussing proxy-based glacial CO₂ decline). We will also address that the Yamamoto et al. (2022) record has low CO₂ levels compared to recent boron reconstructions and carbon cycle modelling.

Line 369: It is relevant to compare your results to Watanabe, you could do the same for your CO₂ results with other modelling studies (e.g. Willeit et al. 2019).

We will add more comparisons to Willeit et al. 2019, but also Legrain et al., 2023 and Verbitsky et al., 2018. Willeit et al. 2019 conducted simulations with an intermediate complexity model that simulates the interactions between ice, climate and carbon cycle. They have also conducted an interesting “orbital forced” only simulations (with freely evolving carbon cycle, but without gradual regolith or CO₂ removal), which yielded 100-kyr periodicity. These are relevant comparisons that would improve the discussion section.

Line 372: I think there is a grammatical problem in the sentence.

This grammatical mistake will be solved.

Line 390: I would split the last sentence into two for ease the readability.

The sentence will be split into two.

Figs. 4 and 8: Why not use choose a sea level reconstruction that spanned the last 1.5 Ma ? Here we can not compare your modelled sea level with other reconstructions in the early Pleistocene.

We will add the sea level reconstruction by Rohling et al., (2021; Science Advances) to both figure 4 and 8, and introduce the differences between our modelled and the reconstructed sea level in section 3.1.

Fig. 6: Adding a quantitative x-axis would enhance the readability of the figure.



We will add a quantitative x-axis, but at the same time also keep the “glacial climate” and “interglacial climate” on the x-axis for easy readability.

Since we conducted additional constant_insolation simulations for the revised manuscript, we will add these to figure 6. We have removed the baseline_ice_core experiment, as this became part of the supplementary information.

Additionally, since we added three simulations (and removed only one), we have many additional “onset termination” points. It therefore became more difficult to see which volumes correlate to the largest/least number of terminations. As such, we will add a histogram to the figure’s background. This histogram indicates the number of “onset of terminations” per ice volume bin.

References:

Chalk, T. B., Hain, M. P., Foster, G. L., Rohling, E. J., Sexton, P. F., Badger, M. P., ... & Wilson, P. A. (2017). Causes of ice age intensification across the Mid-Pleistocene Transition. Proceedings of the National Academy of Sciences, 114(50), 13114-13119.

Legrain, E., Parrenin, F., & Capron, E. (2023). A gradual change is more likely to have caused the mid-pleistocene transition than an abrupt event. Communications Earth & Environment, 4(1), 90.

Verbitsky, M. Y., Crucifix, M., & Volobuev, D. M. (2018). A theory of Pleistocene glacial rhythmicity. Earth System Dynamics, 9(3), 1025-1043.



Willeit, M., Ganopolski, A., Calov, R., & Brovkin, V. (2019). Mid-Pleistocene transition in glacial cycles explained by declining CO2 and regolith removal. Science Advances, 5(4), eaav7337.

First of all, we would like to thank the reviewer for their comments on our manuscript. Here we would like to address these concerns. The reviewers’ comments are shown in bold; our answers use regular font type instead.

This manuscript studies the causes of the Mid-Pleistocene transition (MPT) with an ice sheet / climate coupled model forced by CO2 variations and orbital variations. The ice sheet model used is IMAU-ICE version 2.1, a vertically integrated model. The transient climate forcing of the ice sheet model is derived using a matrix method by interpolating snapshots of global climate simulations. This method allows us to provide transient climate forcing at a significantly reduced computational time compared to GCM’s or intermediate complexity models, albeit without all involved interactions between climate components taken into account. The CO2 forcing used is the one from Yamamoto et al. (2022) based on a leaf-wax indicator. Both sea level and benthic oxygen-18 variations are simulated.

The authors are able to simulate the change of frequency of climate variations during the MPT with this setup. They argue that, because CO2 is the only forcing which has long term variations, it should be the cause of the MPT in the model. They explain that there are 3 regimes of ice sheet variations. Small ice sheet tend to disappear with a relatively small climate forcing. Middle-size ice sheets tend to grow. And large ice sheets are unstable because of several feedbacks that are discussed. They then perform several sensitivity experiments with constant CO2 levels or constant orbital forcing.

Overall, I find the manuscript interesting and well written. The conclusion that a gradual decrease in CO2 is responsible for the MPT coincides with the conclusion of Legrain et al. (Earth. Com. Env., 2023) from a conceptual model. The three regimes of ice sheet variation correspond to what was first proposed by Paillard (1998) and later by Parrenin and Paillard (2003, 2011) with their conceptual models. The model used is quite complex and the work is quite impressive, but the results are presented in a accessible way.

Specific comments

207:  It should be noted that the model does simulate quite well the change in frequency of sea level variations, but not so well the change in amplitude.

Indeed, this issue was only briefly addresses in section 3.1. Our main focus was to capture the frequency change during the MPT. Therefore, we propose several changes to the manuscript to address the lack of amplitude change:

• In the abstract we will replace the line "capture glacial-interglacial variability" with “capture glacial-interglacial periodicity". As we simulated the frequency change, but did not obtain substantial amplitude change. Similarly, in line 85, we will change the sentence to: "Our main goal is to explore if we can simulate the frequency change during the MPT“.
• In section 3.1, we briefly mentioned our lack of amplitude change. We will add a comparison to a sea level reconstruction by Rohling et al., (2021; Science Advances). This record will be added to figure 4 (time-series of the baseline simulation) and figure 8, and covers the entire 1.5-million-year simulation.

• In the discussion section, we will propose two reasons that could partially explain our too-large amplitude in the Early Pleistocene: Firstly, we have now completed a simulation with low (sediment) friction during the Early Pleistocene, which shows a ~10% reduction in amplitude. This run will be added to the supplementary information, and briefly introduced in the results section. Secondly, the CO₂ concentrations in the Yamamoto et al., (2022) reconstruction are low compared to boron isotope and carbon cycle modelling studies (e.g., Chalk et al., 2017; Willeit et al., 2019). If the Yamamoto et al., (2022) CO₂ levels are indeed underestimated in the Early Pleistocene, this could also partially explain our large ice volume during that period.

209: There are also long term amplitude modulations in the orbital forcing. Legrain et al. (2023) have a simulation of the MPT with only orbital forcing, which is less realistic than the simulation with a long-term forcing, but which still contain some sort of frequency change.

Legrain et al. (2023) has conducted conceptual model simulations and obtained a change in glacial-cycle frequency in the ORB (only orbital forcing) experiment. This shows that they were able to capture some characteristics of the MPT, without any additional drivers. Legrain et al. (2023) will be compared to our results in the discussion section.

The claim "since orbital cycles cannot explain the MPT" is therefore too strong. We will replace the sentence with “Since orbital cycles alone cannot capture all characteristics of the MPT".

252: "50 m.s.l.e."

This number refers to the “gap” where for a certain range of climate forcing and ice volumes we obtained few deglaciations. To prevent confusion, we will replace the line "between 40 and 55 m.s.l.e."  (meter sea level equivalent) with "around 50 m.s.l.e.". Additionally, we will add a histogram in the background of figure 6a, which should make this lack of terminations at 50 m.s.l.e. clearer.

372: "Simulating 1.5 million years requires..."

This grammar mistake will be fixed.

381: it is a bit surprising to have a discussion section but no conclusion.

We will add a brief conclusion section after the discussion.