This paper aims to provide a comprehensive analysis of the interactions between edge effects and canopy structure, microclimate, diversity and topography using an exciting dataset collected using cutting edge drone LiDAR. The paper is ambitious in its aim to provide a holistic view of edge effects and their impact on forest functioning and also in disentangling the directionality of these effects. I congratulate the authors on collating this multi-faceted dataset. I enjoyed reading the article, it was well written and interesting and I particularly liked the discussion. This is a complex and relevant topic and this paper had many components, which is to be expected given how varied the data is, but also meant at times the methods and results were hard to follow.

There were moments where I found it difficult to understand the directionality in results between the results and discussion. For example, are edges and associated abiotic changes driving a different composition through environmental filtering as the authors mention, and therefore changing structure or are shifts in composition brought about by changing structure. I felt like the article sometimes lacked clarity here and from what I understand is a core finding of the study.

I also felt like the methods section for the drone data lacked the detail that was later evident in the statistical and field based sections. I think its important to cite the tools used and algorithms explicitly for both recognition and reproducibility. I also think a slightly different direction could have been taken in extracting the metrics, in particular the gap fraction but also in how canopy height and ground surfaces are calculated. I think the gap fraction measure could be improved as seen in my comment below.

Regarding topography, there was very little signal – is there little variation across the site and therefore does it warrant inclusion. I can see elevation in the figure but what about slope and curvature.

I am not an expert of structural equation modelling so I didn’t really have any comments here apart from a minor question below.

line by line comments

Title: I know this is highly subjective and therefore this is only a suggestion, but I think the title could be improved.

Apendix sm 4: Probably better to show delta AICc values rather than absolute values.

62: suggest changing “firsts” to “first.

170-172: How accurate do you think the GPS was given that 11.3 m radius plot isn't that big. How were data geo referenced – for example you fit models using UAV metrics and plot metrics, what kind of mismatch error do you expect given that you used a small portable GPS unit?

206: Maybe highlight how high above the canopy (on average) this is.

209: What processing did you have to do in LAStools?

211: Why 5m meters? How many ground returns did you have and did this differ with distance from edge? Which algorithm did you use?

208: Was the point density relatively consistent across the study area? For example were there any 1 m pixels where a small number of points caused issues in extracting a height measurement.

211 – 212: Not that I strongly believe this should be done or change the results in any significant way but I think more sophisticated and accurate surface models are available such as pit-free CHM. These use triangulated surfaces instead of point to raster techniques which are more sensitive to noise and gaps. (Khosravipour, A., Skidmore, A.K., Isenburg, M., Wang, T. and Hussin, Y.A., 2014. Generating pit-free canopy height models from airborne lidar. Photogrammetric Engineering & Remote Sensing, 80(9), pp.863-872.). Perhaps one to consider in the future.

215: 1.37 m seems quite low to define a canopy gap. I am aware these thresholds are difficult to set but why did you choose 1.37 – see reference below where a 20 m drop was used instead.

219: ‘thalwegs’? Maybe use something a bit more easy such as ‘valleys’ or along those lines. ‘ridges’ make intuitive sense for the positive curvature.

214-216: [Note I appreciate you highlight this late in discussion but have left here as relevant to methods] Is the aim here to quantify gaps within and between trees or both? I interpret canopy gap fraction as the probability of a sun ray entering the canopy not intercepted by plant elements but within the canopy gap dynamics literature it refers to gaps between trees associated with mortality. The former usually necessitates more complex modelling involving radiative transfer modelling which you mention later on. Your index at 20 m resolution is including a mix of within tree and between tree gaps without explicitly modelling the radiation regime.

A full 3D approach would either ruse radiative transfer or voxelise the cloud to determine the proportion of empty to filled voxels. For consistency I would suggest either using the fine scale CHM and classify gaps looking at sudden drop's in height (to a threshold) or using voxels (setting voxel size can be difficult - https://www.mdpi.com/2072-4292/14/5/1054/htm#B29-remotesensing-14-01054). 20 m spatial scale seems a bit arbitrary to me but looking at gaps using a fine scale chm like you have would be more consistent and minimise issues around sensor viewing geometry. The proportion of points potentially a problem if when flying the edges the scanning field of view has many ground first returns ‘from the side’ entering the canopy – using the chm would reduce this issue. For instance see https://onlinelibrary.wiley.com/doi/full/10.1111/ele.12964 – where a fine-scale grid was used with a 20 m height threshold to determine gaps. As above, 1.37 seems low and would mean small plants in an otherwise deep canopy gap would lead to misleading characterisation.

216-222: Please detail how these were calculated (i.e. what package and algorithm) – also how much topographical variation was there across your sites? It didn’t seem to have much explanatory power but could this be due to lack of variability?

310-321: What packages and functions did you use to fit these models?

310-348: I just wondered why you chose to fit multiple linear regression models prior to the SEM modelling? Especially seeing as you mention a hypothesis on lines 325-327. Why not build an SEM directly on what you hypothesise. Not really a criticism but curious.

361: Could diversity not influence canopy structure rather than the other way round? i.e. environmental filtering of species that then set the canopy structure.

Figure 4: There is quite a lot going on here which makes it a bit difficult to interpret.

623-627: I wonder whether this is really possible at the moment? I could be entirely wrong but I thought that given current resolutions, mixed pixels on the edges of forests would cause problems. I do agree drones are an exciting opportunity to look at these effects but height and diversity could be hard to quantify from space at this granularity.

644-651: Okay I can see you had this in mind – I have left my comment above as I still think its relevant in terms of the approach to quantifying gaps.

634-651: I think its a shame to end the paper with caveats of what could be better. Include it by all means but I think it would be stronger to end on the bigger findings of the paper and their relevance to understanding the effects of edges on forest functioning. Could then move the caveats part a little earlier maybe?

Thanks for all the hard work you have put into this, paper looks great. Congratulations.

Two minor points:

"thalwegs" was still used in text. I don't mind which term is used but in the response it was suggested that the authors changed this so more just to make sure its consistent.

Lastly, please proof read the in text figure citations looked incomplete to me (i.e. with XXX).