This study compares simulation output of the DGVM LPJ-GUESS using forcing data with different spatial resolution (approx.. 25 km² vs. approx.. 2500 km²). In particular, the authors emphasize on a comparison between two focal regions (one region with a high relief energy vs. another relatively flat region) as well as a pan-European simulation. The authors find that particularly in mountain regions (such as the Alps) the higher spatial resolution of the input data results in relatively large (up to almost 50%) differences in key output variables such as NEP, standing carbon mass, and LAI. Moreover, they emphasize on effects associated with coastal regions, where the coarse spatial resolution results in an overestimation of land-area and consequently related output variables, yet almost an order of magnitude lower as the effect reported for mountain regions. Based on this, the authors conclude that the biases introduced by coarse resolution should be taken into consideration when interpreting DGVM output since they rightfully claim this not to be a phenomenon specifically related to LPJ-GUESS.

As such, the study brings up an important aspect of dynamic vegetation modelling and consequently matches the scope of GMD very well. While I generally recommend publication of the study, the manuscript yet has to undergo substantial improvements regarding the overall structure and in particular the presentation of methods and results. In particular, I sometimes found the level of mathematical details overwhelming, whereas some textual parts of the manuscript lack sufficient detail to allow for reproduction of the approach. A general recommendation – in terms of readability – would therefore be to move mathematical deductions to the supplementary and elaborate textual descriptions. On a related note, I strongly recommend to transform the partly heavy tables into visual output (as done for Table 5 and Fig. 3) and present the tables in the supplementary. Finally, I wonder whether the effect of spatial resolution in coastal regions cannot be resolved more efficiently (see my specific comment on section 5.2 below).

In the following, I provide more specific suggestions on how to improve the manuscript. Once these issues have been resolved, the manuscript in my opinion is acceptable for publication. Please note, that since the line numbers are not continuous (only every 5^th line is indicated) I mostly based my comments on section numbers and not line numbers.

Section 1:

The introduction is relatively short and would benefit from elaborating in depth, why higher resolution climate input is required to more accurately simulate ecosystems. For instance, examples on topographic effects on temperature and precipitation can be mentioned, as well as their consequences for simulating impacts of extreme events such as late-spring frosts and droughts.

Also, some relevant studies which have previously used high-resolution climate-data input for DGVMs deserve a mention. For instance, (Meyer et al., 2024) used a 250 m x 250 m spatially resolved thin-plate spline interpolation for single-point simulations as well as a downscaled 5 km x 5 km set of forcing data for spatial simulations to better resolve the impact of late-spring frost which represents a phenomenon that requires high-resolution forcing data to account for small-scale variations in micro-climate as discussed in Meyer et al. (2024). Additionally, the work by (Levin, 1992) and (Müller and Lucht, 2007) deserve a brief mention in the introduction and a discussion when interpreting the results. Müler and Lucht (2007) do not simulate at an as high spatial resolution as you do here, but they discuss the impacts of spatial resolution on simulation output, which is the main point of your paper.

Levin, S.A., 1992. The Problem of Pattern and Scale in Ecology: The Robert H. MacArthur Award Lecture. Ecology 73, 1943–1967. https://doi.org/10.2307/1941447

Meyer, B.F., Buras, A., Gregor, K., Layritz, L.S., Principe, A., Kreyling, J., Rammig, A., Zang, C.S., 2024. Frost matters: incorporating late-spring frost into a dynamic vegetation model regulates regional productivity dynamics in European beech forests. Biogeosciences 21, 1355–1370. https://doi.org/10.5194/bg-21-1355-2024

Müller, C., Lucht, W., 2007. Robustness of terrestrial carbon and water cycle simulations against variations in spatial resolution. Journal of Geophysical Research: Atmospheres 112. https://doi.org/10.1029/2006JD007875

Based on such an elaboration you may want to consider to present specific questions/hypotheses that your work addresses, e.g. that higher resolved climatic input allows for more precisely mapping spatial heterogeneity of key model output variables in mountainous/coastal regions. Thereby, readers would already get a better glimpse of the topics the paper actually touches.

Section 2.1:

It is not clear whether this section describes a data source or an algorithm to process data (reading on, I understood it’s the latter). Please refine the section to make this clear. Recall, that CHELSA typically refers to a ready-to-use downscaled climate grid and most readers will likely initially interpret it as a data-set (as did I).

In line 45 there is an odd (3) behind the spatial resolution. I assume this is a LaTex typo.

Section 2.1.1:

The adiabatic lapse rate depends on the moisture content, with more humid air featuring a lower lapse rate compared to dry air (roughly 0.65K/100m vs. 1K/100m). From the description, it seems you did not take this into consideration but simply used elevation and pressure to derive lapse rates. I wonder how much error is introduced by this approach and I propose to at least mention the applied lapse rate (dry vs. moist) and discuss the potential implications of this or ideally - if feasible - resolve it. But I understand that this might be too labor intensive, so possibly a thorough description and discussion is sufficient at this point. In any case, since this effect is larger in mountainous regions, i.e. where you reported the largest effect of topography, it deserves a critical discussion and suggestions for solutions in future work.

Section 2.1.2:

The downscaling of precipitation is not reproducible. For instance, I wonder whether CMIP6 wind data is used to derive the wind effect index or whether this is a purely topographic measure. I guess the former, since otherwise luv and lee - which depend on wind direction - cannot be identified. So, this certainly needs to be better elaborated. Ideally, you add equations as for the previous section from which the actual data processing and input variables can be reproduced and refine the textual description of the processing.

Please note, that it is not recommended to use the same variable nomenclature for different variables. In section 2.1.1 ‘H’ refers to elevation, here ‘H’ refers to the wind effect index. Please revise.

Section 2.1.3:

I do not fully get whether slope aspect and inclination are considered in the downscaling of rsds. Since this can make quite a difference in mountainous regions - which is a focal aspect of the paper - it should to the least be discussed and ideally implemented. But from the description on the 'adjustment according to the surrounding topography' it is not clear whether slope and aspect are included, too. It rather reads as taking into consideration shadow effects but not slope aspect and inclination.

General question: what spatial resolution does the underlying soil information have? Was this adjusted to match the spatial resolution of the forcing data? If not, this might explain some weird patterns observable in Fig. 4 (see my specific comment below).

Section 2.2:

This section lacks a clear rationale/message. The level of detail to which bootstrapping is explained is comparably high (and I wonder whether bootstrapping – which is a commonly applied procedure really needs that level of detail in the main text) but the purpose for running a bootstrapped hypothesis test is not clear. What is the main aim of bootstrapping and which data are used? Is this to show agreement or disagreement between the data from different spatial resolutions? This does not become clear the way it currently is presented.

And I wonder whether a wilcoxon rank-sum test (also known as Mann-Whitney U-test) would not perform equally robust since it has been designed for non-normally distributed data with low sample size.

Section 2.3.1:

In contrast to the previous sections, this section stands out due to its clarity in describing LPJ-GUESS. I recommend to adopt the style of writing and presentation of methodological details from this section to the previous sections.

Section 3:

I wonder why this section deserves its own main header (3). Why not simply adding this to section 2 and term section 2 ‘material and methods’?

Section 3.1:

I don't understand why you used a different downscaling approach for wind and relative humidity. Wind-speed is spatially quite heterogeneous so a detailed discussion on possibly introduced artifacts is certainly required if using a bilinear interpolation of wind-speed. Ideally, the authors would make suggestions on how to improve the downscaling of wind and relative humidity.

Sections 4 and 5:

I understand, that the authors decided to present the methodological approach for each of their two experiments before presenting the experiment outcome. Yet, I wonder whether these methodological aspects should not go into section 2 (to which section 3 is added, see my comment above) and then emphasize on the main findings in section 3 – the results. I personally would find this way of presentation more intuitive than the current version.

Section 4.1 – line 168: ‘The latter condition was intended to prevent significant global differences in climate between the two areas’ - This statement does not make sense. The Pannonian basin features a very distinct climate than the Alpine Arc. Yet, I wonder whether this similarity is really required or even possible for your analyses.

Very last statement on page 9: Only now it becomes clear why you applied a bootstrapping. As above, I recommend to restructure the methods section to link all of this related information more clearly, possibly in a specific section termed statistical evaluation or alike. And again, I wonder whether Wilcoxon rank-sum test might not also do the job. But this is more a philosophic question.

Line 190: why not running the whole experiment with these data from the very beginning? Please clarify why two different experiments are needed.

Section 4.2:

The tables presented in this section are difficult to digest and I wonder why tables 6 and 7 are not accompanied by figures as is table 5 with fig. 3. The authors may want to visualize tables 6 and 7 to then move the tables to the supplementary information and focus on the visual interpretation, which still can contain information on test-statistics if significance stars are added.

Table 5: While the table is quite informative, I personally find it to better fit into the supplementary information. Instead, I would add significance stars to Fig. 3 to make clear which variables showed a significant effect of the downscaling. In the text, I would also emphasize on the actual fractions observed, i.e. down to approx. -50% for the mountain region and only down to -10% for the Pannonian basin. This provides readers with a better relative impression on how much precision is gained for a given parameter when using finer-grained forcing data.

Section 5.1:

Line 242: please indicate clearly which domain you're referring to. If you would move section 5.1 to the methods you probably don’t have to make this link because you can generally describe your domain and then elaborate on the experiments.

Figure 4: I wonder why the authors have chosen to not show fractions of the mean value as in section 4/Fig 3. Moreover, it seems there are some weird pixels, e.g. in Norway or Finland, where a clear fingerprint of the LR data can be seen in between high delta values. I recommend the authors inspect these grid-cells to check for potential artifacts. Could this be related to the resolution of the underlying soil information in case this was not spatially downscaled? Did you downscale soil information?

Section 5.2:

Line 260: The climate effect alone is only 2.1%, i.e. much less compared to the topographic effect of mountains. Since the geographic effect seems to be dominant (3.4 % vs. 2.1 %) I wonder whether this bias cannot be accounted for by adjusting the values for coastal grid-cells according to actual land-mass. So, in your example of Fig. 5 the output of the northeastern LR-grid-cell could be weighed by a factor of 1-25/64 (25 out of 64 grid cells are water pixels) to better represent the actual land-mass contribution in coastal regions. This might be a more efficient way of treating spatial effects in coastal regions. So, for coastal regions there might be a relatively quick fix to improve simulation accuracy, since the remaining 2.1 % of climate effects probably are within the ballpark of general uncertainty of DGVMs. This aspect deserves more attention in the discussion, i.e. the current section 6 (which I would intuitively see as section 4). For mountain regions I however fully agree, that a spatial downscaling is required to improve accuracy given the comparably stronger effects.

Line 267: I do not fully understand why LAI and FPC cannot be quantified in a similar manner. Please elaborate.

Section 6:

I personally believe, that the topographic effect is more important than the coastlines based on your results shown above. In the Alps you showed fractions up to 50% deviation from the mean, whereas the effects of coastlines at most were 10.3 % which could partly be resolved by accounting for actual land-mass within the LR grid-cell (see my comment section 5.2 above). This aspect deserves more attention (see also my comment above).

Line 276: I don't get the implication of this sentence. Why should it not affect other models? And below you even state that other models should be affected, too. Please clarify.

Instead of ‘growth season’ I would refer to ‘growing season’

Line 283: Spatial PFT realization is likely affected, too, beyond productivity and vegetation cover in general. Please include this aspect into your discussion.

Line 293: See my comment above. It should be possible to weigh the output achieved for coastlines according to the actual land fraction of a coarse grid cell. This does not resolve topographic effects but for coastlines it should do the job. Please discuss.

Line 300-316: I wonder whether this level of mathematical detail is required for a hypothetical framework which is designed for a future study. It does not really harm to have it, but it distracts from the actual point of the current manuscript and the discussion of its findings. I therefore suggest to simplify this paragraph and omit the theoretical/mathematical framework.

Otryakhin et al. applied downscaling to the meteorological input data for LPJ-GUESS and assessed the impact of using high- versus low-resolution climate data on model outputs. They statistically evaluated the differences introduced by the orographic downscaling by comparing mountainous and relatively flat regions, demonstrating that differences in model outputs due to climate data resolutions are more pronounced in mountainous areas. The statistically robust approach presented in the manuscript provides valuable insights not only for researchers applying downscaling techniques, but also for those using coarse-resolution climate data. For example, it provides information for evaluating whether statistical errors arising from the coarseness of climate data fall within the range of uncertainties caused by other factors, such as model parameterization or variability in input data.

I recommend this manuscript for publication in GMD; however, I request a major revision due to several concerns. While the writing is mostly logical and clear, there are sections where the lack of detail makes it difficult to fully understand the methodological flow from approach to results. Although the statistical procedures are described in detail, the downscaling method, the core aspect of the study, is not explained clearly. Individual comments are provided below. I believe that the revised manuscript will be suitable for publication in GMD.

Major comments:

L4-9

The abstract seems too simple. It should elaborate more on the unique aspects of this study, the insights gained, and its advantages. Specifically, the differences in climate variables caused by elevation gradients and their effects on the model should be clearly described. The introduction is similar in this regard. It would benefit from more detailed information that is linked to the experimental design. Since the analysis to investigate the effects of elevation differences is well-executed, it would be better to explicitly explain how high-resolution climate data influences dynamic vegetation models.

Section 2.1

It would be better to include a justification for the selection of CHELSA. Clarifying the differences from dynamic downscaling methods would help make the objectives of this study clearer.

The version of CHELSA used in the study should be specified.

L16

The authors mention local extreme weather events, but is the downscaling approach used in this study capable of reproducing such events? For instance, how accurately can CHELSA represent localized extreme precipitation caused by topographic effects, and what specific types of events can it capture?

Precipitation in methods

What is the spatial resolution of the satellite data? In the manuscript, some information such as climate variables is summarized in tables. It may be helpful to include this information in a table as well. Overall, the description of the downscaling methods is ambiguous. In particular, for precipitation and shortwave radiation, additional details are needed to ensure reproducibility. It is necessary to include a clear explanation of how low-resolution data are distributed across the high-resolution grid cells (e.g., Eq. 24 in Karger et al., 2023).

L68-76

It is difficult to understand from the presented equations how the downscaling from low to high resolution is actually performed.

L88-89

Is Equation (6) essential? The statistical testing is described in detail, whereas the downscaling method lacks sufficient explanation, leading to an imbalance in the presentation.

L101

It is unclear whether the “50-100 observations” refer to the number of grid cells at the downscaled or raw resolution. This should be stated more explicitly. Also, is this number limited by computational constraints? In Fig. 8, for instance, a simulation is performed at the European scale, so a more detailed explanation would be helpful.

L118-119

Since the manuscript includes fire on–off experiments, it should include a more detailed explanation of the fire-related processes to enhance clarity and reproducibility.

Fig. 2

It might be helpful to provide more information, such as what i represents and the sample size.

4.2 Results

It would be helpful to illustrate the characteristics of both the high-resolution and low-resolution climate data, for example using maps. This would make it easier to understand how downscaling affects climate variables, especially in regions with significant elevation differences.

Overall, the results are presented primarily as statistical information, but it would be helpful to also show the spatial differences visually using graphs or maps.

The statistical explanation of the errors arising from differences in resolution was very clear. Has the study examined whether using downscaled climate data improves the agreement between model simulations and observed fluxes?

If so, a brief description of this result would help strengthen the justification for using downscaled climate data in the modeling framework.

Table 3

Aren’t the units of fluxes kgC m⁻² yr⁻¹? Isn’t stored carbon expressed on an area basis?

Are the characteristic outputs of a DGVM, such as vegetation transitions, not evaluated in this study?

Fig3:

Roff showed remarkable difference between experiments in Fig. 3(b). Roff exhibited a notable difference between experiments in Fig. 3(b). Could you clarify the cause of this discrepancy?

L234

The discussion on the contribution of fire appears somewhat abrupt. Could you clarify why fire is considered to have a significant impact? Additionally, if fire events are infrequent, wouldn't ensemble averaging tend to smooth out their influence? Is geographical bias a particularly important and non-negligible source of uncertainty for the processes simulated by LPJ-GUESS?

L265:

How were delta(cli) and (geo) calculated?

In carbon budget calculations, the proportion of land cover within each grid cell is usually taken into account, so the error in the climate response would appear to be the more important factor.

L297 “correlations”

While I can infer the intended meaning, it would be better to explain it in more concrete terms.

L278-290

The discussion lacks sufficient consideration of the model processes. While nonlinear responses are mentioned, it remains unclear how the model processes and the downscaled climate inputs interact and what specifically leads to the nonlinear responses. Is the influence of climate variables other than temperature not addressed in the discussion?

L300-315

The proposed testing protocol in this section lacks specificity and its necessity is questionable. The statistical tests already presented in methods are sufficient to serve as reference information for other future studies. If a new approach is to be proposed, it would be better presented in text rather than as equations.

Minor comments:

L45 “(3)”

That is likely a typographical error.