Reply on RC1

(1) The authors could do a better job of motivating this study. It is not clear why it is necessary to simply compare these two time periods regarding Green Sahara (vegetationprecipitation transition in the Sahara). We understand some studies already modified to simulate Green Sahara reproducing intensification and geographical expansion of the West African monsoon (e.g., Pausata et al. 2016; Hopcroft and Valdes 2021), but many Paleomodelling still fails to simulate it (Tierney et al. 2017). Therefore, it would be better to have a clear motivation, for example, to obtain clues (regarding vegetation-climate interactions) to modify the Holocene simulation by comparing the two periods. Alternatively, it would be interesting to have a new fact (not known from the data alone) that can be obtained through the comparison.

REPLY: We thank the reviewer for this good suggestion and for pointing us to additional relevant literature. In a revised version, we will discuss more clearly our motivation for this study. Our main motivation comes from the fact that the abruptness of the termination of a 'Green Sahara' state is still debated and that factors contributing to such a termination are still uncertain. By performing transient climate model experiments on two different interglacials, and by applying two different vegetation models of different complexity, we are able to provide a thorough evaluation of the abruptness and the most important factors. Ideally, similar experiments should be conducted with a range of coupled climate models to extend such an analysis.
(2) It might be better to describe what is already known in the Introduction section and show more new results in this paper. For example, the spatial and temporal complexity of the termination of the African Humid Period (AHP) have already known (Shanahan et al. 2015;Tierney et al. 2017;Dallmeyer et al. 2020), and data regarding the abruptness of precipitation/vegetation decline is a local-or regional-scale feature (Brovkin and Claussen 2008), not the whole Sahara. In this study, one of the main analyses is to investigate changes in climate (particularly surface temperature and precipitation) and vegetation cover in the whole Sahara or North Africa, including the Sahel, during the LIG and Holocene. However, it is better to analyse the western and eastern parts of the Sahara separately rather than the whole Sahara.
REPLY: We thank the reviewer for highlighting the important temporal and spatial variations in the termination of the AHP. We are aware of such variations and in fact we already analyse in Figure 4 separately the responses of the western and eastern parts of the Sahara as suggested by the reviewer. However, we agree that the Introduction can be improved by discussing these spatial-temporal aspects more clearly and will do so in the revised version of the manuscript.
(3) It seems that the data already show that the Green Sahara happened in the two periods, but what else do we know from the data, especially about differences? Finally, is there a reason for no quantitative model-data comparison, in particular, the Holocene has been made at all? We cannot decide whether these simulations are good or bad at all.

REPLY:
We agree with the reviewer that quantitative model-data comparisons can be very powerful ways to evaluate model results. However, we would argue that such a quantitative model-data comparison for the two interglacials would imply a major additional effort that requires a paper of its own. The focus of our paper is on the comparison of different model results for the two interglacials. In our study, we do compare our Holocene results qualitatively with proxy-based evidence and conclude that our model results are consistent with the proxy data. For the LIG, less proxy data are available for a comparison of the nature of the Green Sahara termination. In our revised manuscript, we will discuss more clearly the importance of model-data comparisons.
Not directly related to this study, but compared to the data (e.g., Hoffman et al. 2017;Capron et al. 2017;Scussolini et al. 2019) how good do the LIG simulation (Li et al. 2020) reproduce global-scale surface temperature and precipitation?
REPLY: In our previous publication (Li et al. 2020) we discussed the global response to LIG forcings in our model and the agreement with proxy-based evidence. In the present manuscript, we would like to focus on the response in North Africa and the comparison to the Holocene. However, in the revised manuscript, we will add a few lines that summarize the global performance of the model relative to proxy data.
L.43: "the rate of this transition remains controversial" -The authors would be better to clarify whether these are differences among data or/and discrepancies between data and model.
REPLY: We will clarify as suggested. REPLY: We cited the original publications that first reported on these two different types of transitions. We will add more references in the revised manuscript. 60: "during the termination of the AHP" to "during the AHP" REPLY: We prefer to keep "during the termination of the AHP", as this sentence is about the "rate of vegetation change", so not about the vegetation during the AHP itself.
L.67 and L.71: I do not understand the mechanism well. Compared to the desert, the vegetated area has lower albedo and absorbs more SW radiation leading to a warmer surface. However, a vegetated surface produces more latent heat and cools the surface (thus warming the atmosphere above). Thus, changes in surface temperature should be determined by the balance between warming due to increased absorption of SW radiation and cooling due to increased surface latent heat.

REPLY:
We agree with the reviewer that the involved mechanisms can be explained more clearly. We will do so in the revised manuscript.
L.81: Which specific period is the LIG here? For example, from about 129 ka BP to 120 ka BP.

REPLY:
We thank the reviewer for this point. Indeed, the period with stronger insolation than the Holocene is the early part of the LIG, from about 129 to 120 ka. Since our experiments are focused on the green-desert transition, we start them at 127 ka BP and run until 116 ka BP. This is also consistent with the PMIP4 protocol. We will clarify this in the revised version.
L.82-93: After all, are those simulations quantitatively consistent with the data? Does the LIG data also show a fact of "nonlinear response of the African monsoon to orbital forcing" and "the spatial heterogeneity of the response" as well as the MH data? REPLY: We refer to both modelling and reconstruction studies in this paragraph, and both consistently show that there was a "Green Sahara state" also in the LIG. However, as mentioned previously, there exists to our knowledge no clear proxy-based evidence for an abrupt termination of this vegetated state. L.93-95: Since these two sentences are new topics, they could be moved to a new paragraph. Also, is the issue the authors point out here limited to iLOVECLIM, or does it involve other GCMs as well?
REPLY: We agree with the reviewer and will make a new paragraph in the revised version. This is likely to be applicable to other models as well.
L.103: Could the authors also use these two vegetation models under the same conditions? In other words, can VECODE also be simulated asynchronously with iLOCECLIM? Is it technically impossible?
REPLY: Yes, this would be technically possible.
L.109: The scientific significance of the first and second questions is a little unclear to me. Could the authors please elaborate a bit more on why these questions are important?

REPLY: Question 1 allows us to estimate the impacts of different vegetation components on both interglacials. By answering this question, we know the baseline of vegetation anomalies induced by the differences of vegetation components in iLOVECLIM.
We can then answer question 2 by comparing the two interglacials, aiming to understand how orbital forcing and internal feedbacks affect desertification in North Africa. Questions 1 and 2 are important because before we can analyse and compare the feedbacks during the LIG and Holocene, we first have to characterize what happens to the climate and vegetation in the different experiments. The first two questions may be combined into one in the revised manuscript.
L.119: The authors can describe a little more about cloudiness, humidity, and precipitation of ECBilt because ECBilt is somewhat different from AGCMs. I understand that ECBilt uses the prescribed/fixed cloud cover based on the modern condition throughout the paleosimulation.
REPLY: Yes, ECBilt uses prescribed monthly cloud cover based on observations. This will be added in the revised version.
L.130: LPJ-GUESS adopts a simple two-layer bucket model (with prescribed percolation rate and water holding capacity), but is VECODE the same/similar structure? If they differ, it would be better to describe the difference. L.167: This paragraph is a bit confusing. Does it mean that soil hydrology calculated in LPJ-GUESS does not directly affect ECBilt, but has some indirectly influence through vegetation type?

REPLY: ECBilt-VECODE uses a one-layer bucket model, so different from what is used in
REPLY: Yes, the procedure is as described by the reviewer. We explain this on lines 171-173. We will check the text again to see if we can make it even clearer.
L.185: At each time-slice simulation (HOL_LPJ, LIG_LPJ), how many model years did the authors run the model and how many years of the output were used in the analyses?
REPLY: LPJ was run for 1000 years per time slice, of which we used the last 30 yrs. This will be clarified in the revised manuscript.
L.213: What is the range of the target area (latitudes and longitudes) for North Africa or the Sahara here?
REPLY: We took 10W-35E, and 15N-30N as limits for our analysis, as in Li et al. (2020). This will be clarified in the revised version.
L.223-225: How about the recent (CMIP6/PMIP4) simulations about? Comparison with past simulations is important, but comparison with recent simulations as well as data is also important.

REPLY: We will update this part with most recent LIG simulations.
L.225: Why is the LIG_FIX temperature trend positive? (e.g., Bakker et al. 2014). We will clarify this in the revised paper.

REPLY: In LIG_FIX, the vegetation is fixed to desert in the entire experiment, so there are no changes in albedo as in the experiments with dynamical vegetation. Even without the albedo effect, the precipitation in Northern Africa was still significantly higher in the early part of the interglacial due to the enhanced summer monsoon, forced by elevated insolation values. This high precipitation resulted in relatively humid soils and enhanced evaporation, leading to evaporative cooling in the first part of the LIG relative to the end of the LIG_FIX experiment. This created the positive temperature trend that is also seen in other LIG experiments without dynamical vegetation
L.228: Because LIG_LPJ does not show large changes in surface temperature in North Africa, are changes in surface temperature and desertification (vegetation cover) less relevant in this simulation? REPLY: The surface temperature changes in LIG_LPJ are less expressed than in LIG_VEC, so they seem less relevant for the desertification. However, the response in LPJ_LIG is more regional, with a relatively strong response in the western part of the Sahara (see Figure 3).

L.235: Fischer and Jungclaus (2010) analysed time-slice simulations, not transient ones.
So that may not be an appropriate reference here. Moreover, according to Brovkin and Claussen (2008), which is also cited in this paper, Francus et al. (2013) may not be an appropriate reference either, because the individual data represent local responses and are not representative of the whole North Africa.

REPLY:
We thank the reviewer for pointing this out. We will adjust the referencing accordingly.
L.235~237: Figure 1f shows that magnitude of precipitation decline in HOL_LPJ is similar to one in HOL_FIX, and this sentence may not be appropriate.

REPLY: We agree and will correct this in the revision.
L.245: Change Fig. 2f to Fig. 1f or Fig. 2b(?) Anyway, we cannot consider "the simulated vegetation distribution and spatial divergence in North Africa" from this figure, I think.
REPLY: We thank the reviewer for pointing this out. We will refer to Figure 3 which shows spatial responses in vegetation.
L.253: It seems that surface temperature trend in HOL_LPJ is similar to one in HOL_FIX.
REPLY: Acknowledged. We propose to add "similar to HOL_FIX". L.257: Can we check the ratio of trees to grasses in North Africa? Vegetation-induced changes in albedo and surface evaporation may also depend on the surface conditions between trees and grasses.
REPLY: Yes, we can technically calculate the ratio of trees to grasses in North Africa. In fact, on the one hand, tree-cover was very limited in North Africa (the maximum tree cover of 16% at 127 ka BP was simulated in LIG-VEC, while the tree cover was less than 5% in the meantime in LIG_LPJ) and the changes of grass and desert cover are in phase; on the other hand, we can see declines in both tree and grass cover during the two interglacials, and neither trees nor grasses exist in North Africa after desertification. Therefore, we considered vegetation-induced changes through total vegetation cover rather than tree-induced and grass-induced separately.
L.262: What is the reason for the sharp decline in vegetation cover, especially from 123ka o 121 ka in the LIG_VEC simulation? Moreover, why is that trend not seen in HOL_VEC? (Fig. 1b), but also with stronger cooling (Fig. 1c), showing that feedbacks between vegetation and climate are behind the enhanced desertification. We will clarify this in the revised text.

REPLY: The sharp decline in vegetation in LIG_VEC is simultaneous with a strong reduction in precipitation
L.263: Can the authors check how much the ratio of trees to grass in North Africa varies from model to model? Looking at the vegetation area fraction anomalies (Fig. 3), there may be considerable differences between the two models in terms of the proportion of trees and grasses.

REPLY: Yes, in both LIG_LPJ and HOL_LPJ tree-cover remains less than 5%, and the main declines of vegetation cover were contributed by grass. Compared to these LPJexperiments, tree cover was somewhat higher in both LIG-VEC and HOL-VEC during the early periods of both interglacials although tree-cover was also very limited. During the desertification, both trees and grass decline until they disappear.
About the different vegetation diversity between the two models, unlike Claussen et al. (2013) VECODE and LPJ-GUESS are completely different process-based DGVMs, and there must be many differences besides diversity.
REPLY: We agree with the reviewer. More general, it is related to a difference in model complexity. In an earlier paper (Li et al. 2019a), we have tested the impacts of DGVMs with different complexity on vegetation simulations under different climate conditions. Based on our conclusion, the complexity of VECODE and LPJ-GUESS affects vegetation simulations mainly through diversity when the atmospheric CO 2 level is around preindustrial level (280 ppmv), while the difference in complexity affects vegetation simulations mainly through ecophysiological processes when the atmospheric CO 2 level is largely different from PI level. We will clarify this in the text. REPLY: There are three main biogeophysical feedbacks in this process, which are the positive vegetation-albedo-temperature feedback, the negative vegetation-evaporationprecipitation feedback (Liu et al., 2007;Notaro et al., 2008;Wang et al., 2008) and the positive vegetation-evaporation-precipitation feedback (Yu et al., 2017;Messori et al., 2018). Both the negative and positive feedback to precipitation are not seen in our experiments, which could be related to the fixed surface albedo of bare soil that is not a function of water content in both our dynamical vegetation experiments. This could lead to an overestimation of the positive vegetation-albedo feedback in our simulations, and we have discussed this in the manuscript.  Figures 1a and 1e, the summer insolation decreases more strongly in the LIG, with a value well below that of the Holocene between 120 and 118 ka. We will explain this more clearly in the revision.

REPLY: This difference between the later parts of the interglacials is related to the difference in insolation forcing. As can be seen in
L.278: What do the authors think caused the decline with large error bars at 5 ka and 4 ka in the HOL_LPJ simulation? Moreover, why does that feature not catch in the LIG_LPJ simulation?
REPLY: In the HOL_LPJ snapshot experiments for 5 and 4 ka there was a strong increase in the interannual variability in the simulation of PFT cover. This is likely related to one or more PFTs being very close to their climatic limit, such that relatively small variations in precipitation cause large shifts between the area covered by the involved PFT. In this case, also bare ground was involved. In the LIG_LPJ experiments the climatic input was different, meaning that the PFT response was also different. We will add a comment on this feature in the revised version.
L.279~284: The authors should describe the spatial heterogeneity in the Introduction section, not here because this is a known fact. Furthermore, based on this, from the beginning, the region should be divided into East and West for analysis, I think.
REPLY: We agree with the reviewer that this spatial heterogeneity should be treated in the introduction. We will do so in the revised version. However, we would argue that it is also useful for the reader to discuss it briefly here in 3.2.2, as it relates to the model results discussed in this section. L.298: It's hard to see the differences between West and East Sahara from Figure 4. Moreover, what exactly is "A spatial and temporal complexity of the termination of the AHP"? REPLY: In our opinion, the results for West and East Sahara are similar, but still differences can be clearly seen. For instance, precipitation and vegetation cover was clearly lower in the early Holocene in the Eastern Sahara. Spatial complexity refers to this difference between east and west. In relation to the lower precipitation in the east, the AHP ends earlier here than in the west. The temporal complexity refers to this difference in timing. We will rephrase the text to clarify this.
L.302: About the sentence "the magnitude of our vegetation decline is much weaker than in their study", which study/value matches the available data?

REPLY:
We are not sure if we understand this comment. The mentioned sentence refers to Liu et al. (2007). We were wondering if the reviewer had some specific other study in mind?
L 302: Is "the differences in model complexity" simply about the vegetation models between Liu et al. (2007) and this study?
REPLY: Yes, that was meant here.
L.321-323: Does any data also support the changes in climate and vegetation in the LIG are stronger than ones in the Holocene?
REPLY: To our knowledge, data on the LIG are too sparsely to be conclusive on this point. L.328: Is around 125 ka BP and around 8.5 ka BP each peak of insolation at 20N for the LIG and Holocene respectively? REPLY: During the LIG, the summer insolation at 20N peaks at 125ka as can be seen in Figure 1, but during the Holocene the peak was at 10 ka, so a bit earlier than 8.5 ka.
REPLY: Shanahan et al. (2015) compared proxy data with the results from the TraCE simulations (see their Fig. 1 and Supplementary Figures S8 for example).
L.341-344: Does any data also support the changes in climate and vegetation in the LIG are stronger than ones in the Holocene? Or, will the results of the LIG simulation help to improve Holocene simulation? REPLY: As mentioned before, to our knowledge there are not sufficient data on the LIG to be conclusive on this point. L.361: In this study, "the fractional surface albedo of trees, grassland, and desert are seasonally fixed". Could this setting also be relevant?
REPLY: Yes, we agree with the reviewer that this could play a role. We will add this in the revised version.
L.380: It seems that the section 3.4 is not what the authors found out through comparison between the LIG and Holocene simulations. How about discussing at least one issue that arose through comparison? REPLY: We are not sure what the reviewer means here. Is the suggestion to discuss an uncertainty issue that is related to the comparison? We could for instance add that, compared to the Holocene, it is difficult to evaluate the LIG experiments because of the sparsity of appropriate proxy data.