1. General comments
   • The study by Guignabert et al. provides a calibration and evaluation of the process-based forest growth model HETEROFOR for eastern North-American forests. The model is calibrated with data from the forest inventory of Québec and model output on tree and stand level growth is evaluated against an independent data set from the same source. Overall agreement between modelled and observed data on growth and forest dynamics is good, hence the authors suggest that the model is a reliable tool for providing accurate forest growth predictions for Québec, which could be used for testing forestry practices under a shifting climate.
   • The study fits well into the scope of GMD. The methods are mostly clearly described, but at some points reproducibility is difficult (see comments below). Title and abstract provide clear outlines of the content of the manuscript, while the language throughout the manuscript is fluent, but imprecise in a few places. The paper is well structured and easy to follow with straightforward figures that help understanding the results. Also, the thematic background and existing literature is presented in an appropriate manner.
   • One minor drawback of the study is that it is difficult to judge the exact model performance because the reference data did not contain observations of individual tree positions to be initialized in the model. Taking into account spatial information on tree positions is one of the main strengths of the model. The authors provided a workaround to amend this drawback regarding data gaps in the reference data but the question remains how the model would perform with information of single tree positions. Nevertheless, the study still provides a valuable assessment of model performance as it gives a trustworthy range in which the model performance is expected to be located.
   • In summary, the presented study provides a valuable evaluation of an important tool for assessing forest responses under a changing climate for heterogeneous forests. The group of process-based and hybrid models have been shown to represent responses under shifting external conditions better than their empirical counterparts. Additionally, spatially explicit models are needed to describe interactions among individuals in structurally and compositionally complex forests, which are foci of research as they will most likely be more widespread in the future as policy and practice seek to increase resilience to climate change and disturbances by increasing forest complexity. HETEROFOR combines these two approaches, and an evaluation of the model’s performance is therefore highly relevant.
2. Specific comments
   • The title should contain the term “forest” as it may be unclear for colleagues from other disciplines what a “stand” is
   • A very short description of the mortality module/process, i.e. when/why trees die, would be a useful addition to section 2.1, because understanding how mortality is modelled in HETEROFOR is relevant for the long-term evaluation analysis.
   • From 2.4 it was not clear whether CUE is a constant parameter or variable across time for a given individual with increasing dbh. Furthermore, it is not clear where the empirical formula for deriving the CUE comes from. Is there a reference for it or is this a specification in the HETEROFOR?
   • L. 186: “The NPP was obtained from the two inventories for each tree using the reconstruction mode in HETEROFOR (see Jonard et al. (2020) for detailed information) and then divided by the predicted GPP.” The section would be a clearer to the reader if “[…] to estimate CUE.” was added to the end of this sentence.
   • L. 197: “Tree positions were randomly generated considering the social status of the tree and/or the sun exposure class when available, as well as the size of the trees, ensuring that two trees with a large crown were not positioned too close to each other”: As this is a crucial step of preparing the data for a spatially explicit model, it would be good to have a little more information about the process. How exactly were these covariates considered? Was it possible to check the robustness of this location estimation algorithm with observed data on tree positions? It would also be possible to include this additional section in 2.5.1.2. and reference that section here.
   • L. 254: “We focused this evaluation on stands dominated by the two main broadleaved species of Québec's temperate forest, sugar maple (Acer saccharum) and red maple (Acer rubrum).” Why only broadleaved species? Do you assume similar model performances for coniferous and mixed stands?
   • “Predictions were greatly improved for all species when focusing solely on the best prediction for each tree.” (l. 306). “However, this does not mean that our predictions would match the best predictions if we had the initial tree positions, but we can assume that the predictions would probably be between the mean and the best predictions.” (l. 438): These sentences underline the need for an evaluation with observed stem positions. Is the approach presented for producing the tree positions the standard application procedure in HETEROFOR?
   • The language is imprecise in some places. I suggest that the authors go through the manuscript once more to check for words and phrases that are not specific, e.g. “not positioned too close to each other” (l. 199); the use of the terms “accurate” and “precise” (e.g. l. 416), which should not be used synonymously”; “which decreased around -20%” (l. 324): decreased to or by -20 %?
   • Potential reasons for observed-predicted offsets potentially resulting from the model structure are not well discussed, e.g. “However, the seedling height growth was slightly underestimated for the two species and the mortality initiated somewhat early for the sugar maple depending on the self-thinning curve considered” (l. 461), could this be traced back to any specific source of uncertainty, be it input data, model parameter or model structure uncertainty?
   • L. 462: is it really valid to say “results are thus promising regarding the suitability of the model to predict simulate regeneration dynamics and thereby to species composition” if the species composition was fixed at the beginning of the simulation without natural species assembly taken into account and no feedback from mature trees, as done here?
3. Technical corrections
   • L .85: the sentence “HETEROFOR includes various modules and options.” could be omitted as it does not provide any relevant information to the reader.
   • L. 205: “another important” can be skipped
   • L. 275: predicted -> predict
   • L. 292: to avoid confusion with model parameters, the term “performance parameters” could be exchanged with “performance metrics” as the term metric is already being used in the section on the methods and in the discussion
   • L. 342: “very” could be omitted
   • L. 364: “vy” -> “by”
   • L. 380: “quite” could be omitted

This paper describes the calibration and validation of an individual-based forest model. The authors thoroughly tuned the key parameters (a total of 28) for 23 North American tree species using data from around 100 plots of the forest inventory of Québec, and used another 100 plots to validate model predictions. This is a hard work (at least to me!), tedious, but informative. I have more critics to the model developers than to the model users. The topic falls in the scope of this journal (Geoscientific Model Development) and the paper is generally well written.

I just have some questions to ask the authors and hope they can clarify them in a revised version.

1. From the parameter table (Table 2), I see the growth of tree height, diameter, and crown are set as “parameters”. I am wondering that if the model use allometry equations to describe tree size and structure. It seems not. Otherwise, the authors should tune allometry parameters and then use allometry equations to calculate tree growth.

And, these growth rates should be age/size dependent. Do users need to tune them separately?

Another question, since the NPP calculation and tree size growth are estimated separately, does the model calculate carbon cost of the tree growth (i.e., if an individual tree has enough NPP to support its growth)?

2. Apparently, there are many phenomenological equations that are used to link different variables. Are these equations species and location specific?

3. For the long-term predictions (more than one hundred years), there are stochastic demographic processes that change the forest structure (e.g., the random dispersion of seeds, mortality, and regeneration and the positions where the events happen). How does the model choose one case in these random events? Does it affect predicted forest structure and dynamics? If yes, do model users need to make an ensemble of model runs and see how a forest develops in a stochastic system?

4. When trees grow, their crowns get larger and thus some of them must be left behind or crown shapes must change (usually we call it “crown packing”). Does this affect the parameters of a tree? Or the parameters for a species apply for all conditions and life stages?

5. There have been many efforts to reduce the complexity of this type of forest models, such as the model of Perfect Plasticity Approximation (Strigul et al. 2008). A little discussion of why we still need the tree spatial position-explicit model is necessary.

6. I think it would be valuable to publish the tuned parameters for these 23 species as appendix or supplementory material.

Maybe, I should not ask the authors of a model calibration paper question about the model development. I still think it is necessary to explain that, since the calibration processes are way complicated for me, and it is good to let the readers know the specific value of this model.

As for PPA model, please refer to:

Strigul et al., 2008. Scaling from trees to forests: tractable macroscopic equations for forest dynamics

Purves et al. 2008. Predicting and understanding forest dynamics using a simple tractable model.