General comments

In their study, Grégoire Blanchard et al. aim at using LiDAR-derived canopy metrics to unveil the drivers (i.e. canopy height and canopy gap fraction) behind the influence of edge effects on tropical biodiversity (both taxonomic and functional), biomass and microclimate. The authors’ main findings confirm the idea that the influence of forest edges on biodiversity, biomass and microclimate in the tropics is mediated by changes in canopy structures. To show that, the authors used a very elegant and powerful set of analyses, including structural equation modelling (SEM) which allows to highlight such mediating effects. The manuscript is very well written in general and I only have a few set of minor comments, suggestions and questions listed below for the authors’ interest.

Specific comments

Line 69: Here, you may also refer to Zellweger et al. (2019) who demonstrated that local canopy cover was a strong driver of understory microclimate in temperate forests during summer time. Alternatively, if you are looking for other tropical forest referencences, you may also have a look at Table A4.1 in Appendix S4 in Lenoir et al. (2017) for a list of references on the effect of canopy structure and density on understory microclimate (i.e. magnitude of the T°C offset), including references for tropical forests.

Line 148: “based on” should be deleted here, right? Sounds like the sentence suddenly stops and that something is missing otherwise.

Line 156: It would be nice for that section to add a figure or a supplementary figure explaining the study design. This may help the reader a lot to understand the underlying data. As usual, a drawing or a picture is worth a thousand words.

Line 165: Why a minimum distance of 100 m was chosen? Why not 50 m or 200 m? Is there a reason for choosing 100 m as a minimum distance between two sampling points? Did you do that to limit spatial autocorrelation? If so, does that mean that you preliminary studied the minimum distance to loose a spatial autocorrelation signal in your study area? Would this explain why you chose 100 m as a minimum (cf. the lag distance in a correlogram)?

Line 222: Why a 10-m radius plot? Why not a greater (or smaller) radius? Did you test (cf. sensitivity analysis) for other radius sizes (cf. different buffer areas to test which one makes more sense)?

Lines 319-320: Here you mention that best models were selected based on the lowest corrected AIC but you did not mention how many candidate models were tested. It would be nice to mention or display in a table in the supplementary materials which candidate models were tested. Besides, I am not sure you necessarily need that step of model selection based on AICc values prior to running your SEMs. SEMs are usually constructed from hypotheses rather than based on a preliminary model selection… Also, note that a recent paper suggested that model selection based on AIC values should be restricted to predictive models only and not to causal inference (see Arif & MacNeil 2022). As here you are rather focusing on causal inference, it might be worth thinking about dropping the whole part on model selection based on AIC values and jump straightforward to the SEM building based on hypotheses.

Lines 327-330: See my previous comment and have a look at the recent paper from Arif & MacNeil (2022) who recommend not to use model selection based on AIC values for causal inference. My suggestion here is to drop the model selection part and directly explain which SEM you built based on knowledge from the scientific literature and hypotheses testing.

Lines 341-342: Not necessarily appropriate for building SEMs (see my previous comments).

Line 380: What about the effect of elevation per se? I mean, you mention the weak influence of curvature and slope but nothing about elevation. Did you test for the effect of elevation?

Line 408: Be careful, figure 5 does not exist in the main text!!! This must be an error, right? Besides, what is S.M. XX?

Lines 408-409: How canopy height relates to canopy gap fraction? Are these two variables redundant or complementary? Might be worth checking the correlation matrix of the different variables that are used as covariates in the same path.

Line 415: Figure 5 is missing from the main text?

Line 590: True but how much topographic complexity did you encounter in your study area? I guess it is also important to remind the reader about that. Indeed, if topographic complexity is relatively small in your study area and not the main variable of interest here, it is not that surprising that it did not show up as a strong predictor in the models, as opposed to distance to the edge which is the main variable of interest here and thus you made sure to cover its entire gradient from the edge to the forest interior. Do you see what I mean?

I sincerely hope that my comments will help you revised your work.

Best,

Jonathan Lenoir (except if the journal does not allow it, I systematically sign all my reviews since 2011, when I got a permanent position in research)

Suggested references

Arif & MacNeil (2022) Predictive models aren’t for causal inference. Ecology Letters, Early View https://doi.org/10.1111/ele.14033

Lenoir et al. (2017) Climatic microrefugia under anthropogenic climate change: implications for species redistribution. Ecography, 40: 253-266

Zellweger et al. (2019) Seasonal drivers of understorey temperature buffering in temperate deciduous forests across Europe. Global Ecology and Biogeography, 28: 1774-1786