The authors would like to thank both Reviewers and the Editor for insightful comments. These will be very helpful for improving the manuscript, the revised version of which will be submitted in the near future. The responses to the Reviewers' comments are provided below with bullet points, with some (preliminary) indications of how the manuscript will be revised.

Reviewer #1

General remarks

In this manuscript, the authors present the CLASH model. According to the authors, CLASH is "a biophysical land-use model that describes the allocation of land to different uses, forest growth, terrestrial carbon stocks, and the production of agricultural and forestry goods globally." CLASH is designed for hard-linking with IAMs but can also be run in standalone mode. 

According to the authors, CLASH "fills a critical, vacant niche in the model ecosystem" (line 49) by combining features from three model types: Dynamic Global Vegetation Models (DGVMs), economic partial equilibrium models of the land-use sector, and forest sector models.

Indeed, there is currently no such model available that combines features from these three model types in a simple and lightweight model with intertemporal optimization for hard-linking with IAMs. Therefore, the concept of CLASH is innovative. 

• We are glad that the Reviewer finds the model innovative and sees the vacant niche that we wanted to fill with this model. We are also grateful for the questions and observations by the Reviwer, which will help us improve the manuscript further.

The paper is very well structured and written. The language is fluent and precise.

• Thank you. We are happy if the text was clear to read.

In the abstract, the authors claim that CLASH "allows the inclusion of terrestrial carbon stocks, agriculture and forestry in global climate policy analyses."

This might be true from a technical point of view. But it seems that economic factors such as trade patterns and self-sufficiency ratios, protected land areas, and existing climate policies for the land-use sector are largely missing in CLASH (see my detailed comments below).

• Thank you for noting this, and also for the related observations below. It is important to clarify the scope of the model, so that readers or potential users of the model are not confused about what the model represents and what falls outside its scope.
• The scope of CLASH is exactly as the Reviewer has understood. The aim is that CLASH portrays only biophysical factors, whereas all economic or societal factors, such as those listed above, should come from the IAM to which CLASH would be linked; or given externally, if run as stand-alone.
• We will clarify this in the revised manuscript. For example, the cited text passage will be amended as follows: "allows the biophysical representation of terrestrial carbon stocks, agriculture and forestry in global climate policy analyses." We will also stress that economic and policy factors need to be considered outside CLASH (e.g., through an integration with and IAM) to realistically represent land-use.

The absence of these factors, which are highly relevant for near-term model realism, considerably limits the usefulness of CLASH for climate policy analyses. In my view, CLASH is rather a useful tool for idealized (biophysical) experiments but not for concrete climate policy analyses. This aspect should be clarified by the authors in a major revision of the manuscript. 

• Indeed, these are highly relevant factors for model realism, but – as stated above – outside the scope of CLASH. In this sense, CLASH alone does not provide a scope that is sufficient for analyses that would be realistic from the societal viewpoint. Instead, economic and policy factors should be provided through the linkage to an IAM; or e.g. as external constraints when running stand-alone.
• This matter was discussed more explicitly in the concluding section of the manuscript, but rather briefly in the introduction. We will extend the treatment in the introduction in the following way, so that the scope and role of different models will be more clear. For example:
  
  “In this role, CLASH represents the biophysical aspects of land-use, while the IAM needs to provide the rationale for why land is used and managed in a specific way, including economic, policy and other societal factors; and also how the climate changes over time.”

Detailed comments

1.)  It remains unclear which features from economic partial equilibrium models such as MAgPIE or GLOBIOM are represented in CLASH. The authors claim that "CLASH provides a comprehensive, climate-responsive depiction of global land-use" (line 68f). However, they also admit: "This comes at the expense of detail. Ecosystem characteristics, land-use competition, agriculture and forest management are described in less detail than in models focusing on each aspect individually" (line 69f).

It's fair that a model like CLASH that aims to integrate features from different modelling domains represents less details from each domain. However, the authors should clarify in the introduction which features typically included in economic partial equilibrium models are captured by CLASH, and which features are missing.

Key features of economic partial equilibrium models like MAgPIE or GLOBIOM include food demand, agricultural production, factor costs for agricultural production, dynamic land-use competition, yield increases, accounting for historic trade patterns and self-sufficiency ratios, and GHG emissions. 

Partly, the above point is addressed in section 5 (lines 520ff): "Economic factors (such as production costs, trade policies, security of supply concerns, and the value of ecosystem services) are not considered in this optimization problem." However, if economic factors are not accounted for in CLASH, which features of economic partial equilibrium models are then included in CLASH?

• Thank you for this observation. The model scope indeed needs to be communicated more clearly. Particularly the statement that ‘CLASH combines features from three model types’, to which the Reviewer also refers, is too vague to describe what CLASH covers and what is does not consider.
• As discussed above, CLASH covers only biophysical factors, i.e. vegetation growth, carbon stocks, crop yields and livestock (but only from a biophysical point-of-view). As compared to model focused on agriculture, the are no separate crops (only an aggregated crop) or specific management practices (although irrigation and fertilization are accounted implicitly through LPJ-GUESS); while in comparison to forestry models, tree species are not considered explicitly (but only implicitly, through LPJ-GUESS) and managed forests cannot be thinned, but only clear-cut. Economic and policy factors (e.g. costs, markets, prices, trade and self-sufficiency ratios) are outside its scope, but can be introduced externally or when linked to an IAM. These will be stated more clearly in the revised manuscript.

2.)  Related to the above point on economic partial equilibrium models. In line 61f, the authors claim: "Both of these features [dynamic forest age classes and intertemporal optimization] are needed to enable the dynamic optimization of forest management when, e.g., studying global resource use or climate policy questions."

This sentence needs clarification. In my view, it cannot be generalized that intertemporal optimization is a precondition for the dynamic optimization of forest management. Instead, the choice of intertemporal optimization vs. recursive-dynamic optimization depends on the underlying research question. Intertemporal optimization tells us how land should be allocated to different uses over time to meet a certain goal such as carbon stock maximization. However, policy-making in the real world, especially in the land-use sector with its many small holder farmers, happens rather incrementally and not in an intertemporal fashion. In this sense, projections based on recursive-dynamic optimization are likely closer to what will happen in the future, compared to their intertemporal counterpart. 

Moreover, the choice of intertemporal vs. recursive-dynamic optimization may also be informed by the properties of the sector under consideration. For instance, in the energy sector long-term planning is much more established compared to the land-use sector. 

For improved context and clarity, I suggest that the authors revise the above statement, and add some sentences on the relation and choice between intertemporal vs. recursive-dynamic optimization.

• Thank you for these insights. We agree completely that the different modeling paradigms can answer different questions, and thereby can represent different aspects of the real-world with different degrees of realism. Due to this, one is not inherently better than the other.
• Intertemporal optimization (with perfect foresight) might be a very idealized and optimistic approach, as all decision-makers are assumed to do long-term planning with perfect knowledge of future events. In this sense, recursive dynamics can maybe be considered a more realistic description to real-world decision-making. But with recursive dynamics, decision-making is reactive to the present state only, and thus disregards future foresight. This, we think, is not fully realistic either.
• But besides realism, the paradigm choice affects what can be answered with the model. Intertemporal optimization answers what would be an ideal long-term strategy. We think such analyses can be illuminating, even if the long-term strategy would be an unrealistic one due to the short-sightedness of real-world decision-makers. However, both approaches can remedy the ‘missing realism’ and emulate the behavior of the other approach through iterative procedures: running intertemporal optimization with myopia, or running recursive dynamics with appropriately set incentives that reach long-term targets, for example.
• Intertemporal optimization is computationally more challenging than recursive dynamics, as we discuss in the introduction. Exactly because of this, CLASH was designed to be simplified and computationally lightweight, so that it could be embedded with IAMs that employ intertemporal optimization. However, this does not prevent it from being integrated into an IAM using recursive dynamics.
• The difference between these two modeling paradigms might be less relevant for agriculture, but with long-rotation forestry it is more important. In forest economics, the classic Faustmann’s problem of optimal rotation length, already from 1849, is defined as an intertemporal optimization problem. Thereby, we stand by our statement that dynamic forest age classes and intertemporal optimization are necessary to enable the long-term optimization of forest management. We think this is the only logical conclusion if ‘optimization’ is understood in a strict, mathematical sense.
• However, the manuscript can be supplemented with some of the ideas from the above response, as to bring up a more nuanced and broad discussion of alternative modelling approaches, just as the Reviewer suggests.

3.)  It seems that real world trade patterns and self-sufficiency ratios are not considered at all in CLASH, which explains the following result (lines 520ff): "Notably, the maximization of global carbon storage subject to the global demand constraints leads to a strongly polarized land allocation between the biomes." 

Disregarding trade and self-sufficiency ratios considerably limits the usefulness of CLASH for climate policy analysis. Without any near-term realism of land-use at regional / biome level, meaningful climate policy analysis is not possible. Therefore, the corresponding statements in the abstract (line 9) and in the conclusion (line 570) should be revised.

• As discussed above, CLASH in itself does not cover any societal factors, such as trade or policies for self-sufficiency. These could be given externally to the model, however.
• In the demonstration that the Reviewer cites above, we deliberately wanted to keep the optimization problem simple, so that it is easy to see howand why the results arise from the optimization. The purpose of the demonstration was not to portray a realistic analysis for climate policy (for example: there is neither any linkage between crop and animal product demands), but to test the model in a simple setting that would be physically possible.
• In more elaborate uses, CLASH is intended to be embedded into an IAM, which would be responsible for depicting the optimization problem more realistically.
• We will express these aspects more explicitly in the revised manuscript.

4.)  CLASH represents land use at an aggregate level of ten biomes. However, policy-making happens the level of countries or geopolitical regions and not at the level of biomes. This aspect further limits the usefulness of CLASH for climate policy analysis. This limitation should be reflected and discussed in the manuscript. 

• We agree. The current version of CLASH is designed for global, single-region IAMs, and by design, such models cannot address questions on the national level. We still think that such aggregate-level analyses can provide important insights on the long-term pathways for reaching global climate targets, for example.
• We will add a clear reflection of this to the revised version, along with a mention that the model could be re-parametrized for different geographical divisions in the future.

5.)  The authors validate the results from CLASH (mostly carbon) against results from LPJ-GUESS. This is fine as such. However, results from CLASH are not compared at all to the results from economic partial equilibrium models such as MAgPIE or GLOBIOM. If , then the corresponding variables should also be validated against data from these models (e.g. from the IPCC AR6 database) or from FAOSTAT. The same holds true for forest sector models. If CLASH includes features from forest sector models, the corresponding variables should be validated.  

• We plan on including such a comparison in the revised manuscript. However, please note that due to the difference in model scopes, this comparison cannot be fully comprehensive. Partial equilibrium models involve a large array of assumptions and factors about policies and markets, which are outside the scope of CLASH. We can only do comparisons of physical quantities while fixing other physical quantities to other models’ results.
• For agriculture, we can e.g. fix land-use areas in CLASH to a certain SSP scenario and observe the production levels. For forestry, the setting is more complicated, as land-areas do not directly imply a certain production level (for example, the rotation length affects this). We will think of a suitable setting and a comparison point for forestry products.

6.)  under current management? How do yields in CLASH compare to FAO?

To what extent are yield increases due to technological change considered - if at all?

• The yields in CLASH are based on LPJ-GUESS, which models actual yields, assuming some predetermined management practices, like sowing/harvest dates and nitrogen fertilization (see e.g. biogeosciences.net/12/2489/2015/). However, we did not specify these explicitly in our experiments. The model seems to be more sensitive to CO2 fertilization than the average of DVGMs, according to the comparison presented by Franke et al. (https://doi.org/10.5194/gmd-13-3995-2020), and this is the main driver behind the yield changes, along with the warming effect for some biomes. This was discussed in section 2.7, but we will clarify the text further.
• Relating to the previous question, we will also include a comparison of model results with FAO statistics.

7.)  It seems that existing national climate policies for the land-use sector (e.g. reduced deforestation) are not accounted for in CLASH. Disregarding so-called Nationally Determined Contribution (NDCs) in support of the Paris Agreement further limits the usefulness of CLASH for climate policy analysis. 

• As discussed above, such analyses fall outside the scope of CLASH in two respects: it does not consider policies in itself, as it’s a biophysical model (although some policies could be introduced externally); and that it does not cover countries separately (although this could be remedied with a new parametrization with a different geographic split between regions).
• An extended discussion and a clarification about the model’s scope in this regard will be added to the revised manuscript.

8.) Likewise, it seems that existing protected areas (IUCN WDPA, https://www.protectedplanet.net/en/thematic-areas/wdpa?tab=WDPA) are not accounted for in CLASH but would be needed for meaningful climate policy analysis.

• As above, such policies are not an inherent part of CLASH, but could be added to the model as external constraints. This will also be discussed in the revised manuscript.

9.)  correct typos: "DVGMs" -> DGVMs

• Thank you for pointing out this typo. Which will check the text more carefully in the revised manuscript.

Reviewer #2

Overall I agree with the authors that this paper presents a modeling system that addresses an open area in integrated assessment modeling. The descriptions are concise and informative, and the demonstrations all look reasonable. I had a number of questions and minor requests for revisions that I'll organize in the order that I encountered them in the manuscript.

• Thank you. We are glad the Reviewer found the model’s focus relevant, descriptions informative and demonstrations reasonable. We are grateful for the questions and suggestions, which will help us improve the manuscript further.

Line 100: "carbon concentration, as these variables standard outputs of many IAMs" please replace with "carbon dioxide concentration, standard outputs of IAMs"

• Indeed, thank you for noting this. The text will be updated as suggested by the Reviewer.

Line 101: "changes in these variables serve as a proxy for the changes on local climatic factors" - I thought that the opposite was true, that realized climatic conditions, precipitation in particular, are regionally heterogeneous and not linearly related to the global average surface temperature and CO2 concentrations. Can this statement be further clarified, or sourced in the literature?

• Yes, this needs to be stated more clearly. The LPJ-GUESS runs use gridded data of temperature and precipitation that comes from different Earth system models. Thus, as CLASH is parametrized from these runs, the regional heterogeneity is considered, although some detail is lost when the gridded results are aggregated into the level of the ten biomes used in CLASH.

One of the weaknesses I see in the existing IAM information flow right now is that the CO2 concentrations are estimated in simple climate models (e.g., MAGICC) using an equation that estimates enhanced CO2 uptake by the terrestrial biosphere as a function of ambient CO2 concentrations (i.e., the CO2 fertilization response). The additional uptake in turn reduces the estimated CO2 concentrations. However this additional carbon, which is understood to be taken up into the biomass of vegetation and soils, isn't actually represented in the IAMs' land use modules. It looks like this model explicitly represents that aspect of the dynamics, but also it seems that the CO2 concentrations in CLASH are exogenous, and that the dynamics represented within the model aren't passed to a simple climate model, where they could replace the existing CO2 fertilization response function. Can the authors comment on whether the outputs of CLASH are used to provide an endogenous and bottom-up estimate of the global terrestrial biosphere's CO2 fertilization response which can then be used to revise the estimated atmospheric CO2 concentrations?

• Thank you, this comment is right on point. The CO2 concentration is given exogenously to CLASH, as you wrote. The vegetation growth in CLASH then reacts to this, so that one can derive the net exchange of carbon between the atmosphere and biosphere, also accounting for the fertilization effect.
• Some IAMs (such as those we have linked or plan to link CLASH with) have a simple climate module that can calculate CO2 concentration and pass this to CLASH (c.f. the DICE climate module or other similar approaches, e.g. the GAMS implementation of FaIR). And if CLASH is linked to such IAM, CLASH can represent the net exchange of carbon between the atmosphere and the terrestrial biosphere. But, as you correctly point out, the IAM’s climate module then should not consider this exchange (but only the exchange between the atmosphere and oceans), as CLASH accounts for it already. Although this point concerns more the climate modules of IAMs, it should be mentioned in the manuscript to avoid possible errors if some other researchers decide to link CLASH with their IAMs.
• We will, therefore, describe these linkages more extensively in the revised manuscript.

Section 2.7 - in general, when referring to masses and densities, please distinguish between carbon, total vegetation, and dry vegetation. I generally found the terminology to be a bit vague. For instance, in Table 1's caption and conversion factor unit, I'd recomment using "kgC/m2" instead of the currently written "kg/m2".

• Thank you for pointing this out. One can often be blind to these things in one’s own text, while the text can be ambiguous for an outsider. We will clarify the unit notation, as suggested.

Section 2.8 - this is much more detail on the livestock side than I'd have expected for a land/carbon model, particularly given that process-based IAMs already have pretty detailed representations of the inputs and outputs of livestock production. It is noted in the text that the carbon in animal biomass is trivial and as such is omitted. So then, can the authors clarify in this section the purpose of the livestock module, and how it would interface with the livestock representation in an IAM in order to ensure consistency? It just seems strange right now that livestock production, which is economic in nature, would be represented in this model which doesn't represent economic considerations.

• The main reason is that livestock uses vast amounts of land area, both directly and indirectly through feed production, and also produces notable amounts of CH4 and N2O emissions. Although the scope of CLASH is biophysical, thereby excluding the economic and policy aspects of land-use, we wanted to represent the land-use and GHG emission impacts of livestock. Otherwise, the model would not be able to represent why such large areas are allocated for pastures. As an example for the relevance of this, the demonstration of the model provided in section 5 analyzed how land-use, carbon stocks and agricultural non-CO2 emissions react to different demand scenarios of livestock products. Although this is a purely biophysical consideration, it already can provide insight into the role of livestock for climate.
• Should CLASH be integrated into an IAM that already has a bottom-up representation of livestock, one has the options of disabling the livestock representation from CLASH, the IAM, or otherwise connecting the two.

278 - please move the disambiguation of PFT up from line 278 to 277.

• Thank you for noting this. We actually introduced PFTs already on line 210, so the abbreviation will be used from thereon.

Figure 7 - do the crop yields take management practices into consideration? These are the main drivers of historical and future projected yield change, and I'd think they should be an exogenous input from an IAM or other model. The yield changes are so large it looks like they might be considered, but the text suggests otherwise.

• LPJ-GUESS includes some management practices, like sowing/harvest dates and nitrogen fertilization (see e.g. biogeosciences.net/12/2489/2015/). However, we did not specify these explicitly in our experiments – and at the moment CLASH does not enable the optimization of agricultural practices to improve yields. The model seems to be more sensitive to CO2 fertilization than the average of DVGMs, according to the comparison presented by Franke et al. (https://doi.org/10.5194/gmd-13-3995-2020), and this is the main driver behind the yield changes, along with the warming effect for some biomes. This was discussed in section 2.7, but we will clarify the text further.

In the end I'm unclear about what is envisioned for linking with IAMs. This is claimed to be the main value added in the introduction, but then all of the details are left vague in the subsequent sections. Here are some of the questions I have, which don't all need to be answered directly and completely in a revised manuscript, but some information would be helpful.

Which IAM(s) is (are) CLASH intended to be "hard linked" to?

Which variables would be exogenous, imported from the IAM?

What would be the information flow from CLASH to the IAM?

Would it be used to revise the CO2 concentrations of the simple climate model?

• Indeed, this might have been bit vague, partly because the question over IAMs is not inherently a feature of CLASH itself. We have integrated this with the SuCCESs IAM (a new bottom-up IAM we have recently developed) and one of the authors is carrying out the linkage with ICICLE, a more simple top-down IAM. Neither of the works have yet been fully published, however. Also, the potential uses of CLASH are not limited to these models. Due to these factors, we thought that mentioning any IAMs explicitly is not relevant for the scope of this paper.
• The general information flow between the IAM and CLASH needs to be clarified, in any case. The basic idea, and what happens e.g. in the SuCCESs-CLASH implementation, is that CLASH gives the IAM the quantities of produced land-use commodities (food, biomass for energy and materials) and the GHG emissions from land-use. The IAM would provide the demand for these products (determined exogenously in SuCCESs), any policy measures that affect land-use, and CO2 concentration. And as discussed earlier, the climate module of the IAM should only consider the exchange of carbon between the atmosphere and oceans (i.e., not with the terrestrial biosphere, as this is already represented by CLASH). We will clarify these aspects in the revised manuscript.