The paper is an interesting and valuable contribution to the literature. Its unique value lies in connecting field-based measurements with remote sensing measurements, and then using this approach to link gap dynamics with biomass losses. As pointed out in previous reviews, the main underlying issue is one of sample size, but the authors have (mostly) addressed this now in their analysis and discussion. Similarly, earlier points on gap definitions precipitation have been addressed. I also highly value the authors’ efforts in compiling and analysing the data as well as the detailed responses.
However, there are still two main issues that need to be addressed.
Interpretation/discussion of results: I found the interpretation of the results and the discussion sections difficult to follow and sometimes misleading. The clearest instance of a misleading claim was Discussion section 4.4, which states “Extreme rainfall-events control gap formation”. This is a causal claim that is astonishing given both the study’s setup (small spatial and temporal scales) and the study’s results (at best weak correlations between gap patterns and precipitation). Similarly, section 4.3 states “Small-scale disturbances dominate canopy dynamics and associated biomass losses in Central Amazonia”. This was also surprising, because the authors never correlate gap size and biomass losses (which they absolutely should, cf. below!). From the available materials, it seems to me rather that tree snapping, i.e., the mechanism responsible for the largest gaps, accounted for the largest biomass losses, so the opposite of what the paragraph claims. I may be mistaken, but then the authors need to show this. In contrast, sections 4.1 and 4.2, while not misleading, seem to summarize a wide variety of issues without a coherent storyline. Some of the points, such as detection sensitivity in the first paragraph need to be explained, other points should be summarized much more so that readers have clear takeaways. Part of my confusion may be due to the authors attributing properties of methods to the forest ecosystem itself. E.g., in stating “4.1 The mechanism of gap formation is related to the sensitivity of detection” they likely mean that the “classification of gap formation mechanisms” is related to detection sensitivity, not the actual mechanism. Another aspect may be that the authors tend to focus a lot on p-values/significance both in discussion/results, but not on effect sizes (e.g. how large are differences), and sometimes report with a precision that is not warranted for the small sample size (e.g. differences between 6, 7 or 8 gaps between mortality modes are not important, but seem more important when rendered as 2-digit precision percentages). I have made suggestions in the detailed comments to improve this.
Biomass: I thank the authors for putting the overall biomass losses into the context of the carbon stocks at the site, but I still find that the analysis does not fulfill its potential here. I strongly recommend adding one or two scatterplots/correlation plots to Figure 6. This would be one plot c) that plots biomass losses against gap size, and one plot d) that plots biomass losses against gap perimeter or GSCI. In each case, the data points should be coloured by gap formation mechanism. That would quickly tell readers how easy (or difficult) it is to infer biomass losses from gap formation, and would provide an interesting result + directly relate to Discussion Section 4.3.
Finally, I appreciate that it is hard to track all changes in revisions, but there are a quite a few paragraphs where words or descriptions are missing or where the authors have only responded to the reviewers, but not included the respective text in the article (e.g., log-transformation of variables, when performed, needs to be mentioned in the methods).
Detailed comments:
Title: “in a Central Amazon forest”?
l.14: “In addition to detecting” could be removed, i.e., just write: “We measured the size …”
l.17: “corresponding” instead of “associated”?
l. 17: “either … or” not “either … and”
l. 22: “Regardless of”
l. 23: “lognormal” should probably be lower case, as it is not a name (also in the rest of the text)
l. 22-26: not sure whether I would put the different fits of lognormal/Weibull into the abstract. As stated in earlier reviews, given the small sample size such fitting exercises should not be over-emphasized, especially when there is no clear theoretical justification why one should be better than the other.
l. 26: I would reframe: “The main modes of tree mortality were not related to gap size, but to losses in biomass.” The problem is, however: what does this have to do with gap patterns? Can we detect differences in mortality/biomass losses from the gaps themselves (through size/perimeter)? This is one of the main issues referred to above.
l. 27: Again, not sure whether this is such a strong result.
l. 28: Does this not contradict l.26?
l. 32: This sentence is too harsh in my opinion. It’s not like these results don’t tell us anything about biomass/gap dynamics, it’s just that we have to be careful about extrapolating. I would drop the sentence about extrapolation, or merge it with the following: “Future investigations combining remote sensing with field data are needed to confirm these relationships at landscape scale.”
l. 53: Maybe combine the sentences: “The size of gaps can vary … and defines the amount of light …”
l. 54: “Apart from related” – there is something missing
l.58: Remove “Although with some surviving trees”
l.57-70: I would merge these lines into one paragraph on wind and rain, and link them to the previous paragraph, e.g. by starting after “and related functions (Jucker, 2022)” with “In the Amazon, one of the key disturbances are extreme wind and rainfall events….” The following sentences need a bit of reworking/reordering.
l. 74-75: maybe make it a bit simpler “from small number of plots and infrequent surveys is a challenging task”
l.76-77: I would focus only on optical methods here or rewrite this and the following with a clear separation of lidar/optical or satellite/airborne, otherwise it gets confusing for the reader. I.e., first you mention lidar, e.g. as in Greg Asner’s studies, but then you move on to Landsat, which operates in a very different realm, both in terms of resolution and data type. Also, as an aside, while I agree that airborne lidar generally provides high resolution and accuracy for e.g., canopy height, I would be more careful about this statement in the context of gap dynamics. Gap patterns depend a lot on CHM generation algorithms and the resulting surface roughness, so accuracy is probably not well-defined.
l.88: Again, it makes much more sense to only focus on optical data in the previous paragraph if the focus is on optical data here. Otherwise, a comparison with lidar would be needed. E.g., optical UAV are probably much cheaper to acquire and easier to handle, so practical (but they probably also come with a few disadvantages, such as not penetrating to the ground, being more sensitive to low clouds/mist).
l.114: This is not a crucial comment and I am not familiar with the INVENTA plot, but out of interest: this seems an extremely narrow height range for canopy trees (only ~50cm standard deviation in height?). From most height allometries at plot level that I know, tree height variation for trees > 50cm DBH is usually massive (several meters). Are canopy trees very narrowly defined here?
l.198: could not find Magnabosco Marra et al. 2016 in references (was it 2014?)
l.230: “three” should be “four” now
l.245: “two-tailed” what? In your responses you mention the log-transformations, but this needs to be also part of the methods!
l.268-289: This goes back to my questions about your gap definition in the previous round. It is important to differentiate here between what one of the methods registers (UAV) and what has happened in the actual forest (as seen, e.g., from all combined methods). Since the field measurements found three gaps that fulfill Brokaw’s definition, it’s not correct to say that “there was no traceable change in the upper canopy of the forest”. There clearly was a traceable change, it was just not traceable by UAV. This should be carefully phrased throughout the article.
l. 274-277: I would rephrase this (cf. also my comments from last time). A non-significant p-value is not the same as a small effect size (e.g., “no evidence for strong differences”). I would suggest: “UAV-inferred gap size exceeded the field-inferred gap size substantially for the largest gap (254 m2), and on average, the difference was 11.5 m2 (17% of mean field gap size), but non-significant (p = 0.85).”
l.277-278: Make sure to check whether normal distributions actually are appropriate for those metrics as well, or if data were log-transformed or otherwise transformed, this needs to be part of the methods/tests descriptions. If possible, focus on effect sizes.
l.289: I do not understand what “The two most discrepant” refers to here
l.303-305: Please provide the number of gaps used for each analysis, and maybe use “best described” instead of “better described”. As per the more general comments, it’s fine to keep these results, but I would not put too much emphasis on them.
l.307-313: I really like this analysis, but I would not overemphasize the differences gap numbers. Differences between 8,7,6 (and even 11) gaps are hardly important. Also, I would not report percentages in Table 2, or at least not with 2-digit precision! With 32 gaps, a single gap results in a change of ca. 3%, so a precision < 1% is meaningless. You can always summarize as 34%, 25%, 22% and 19%, but probably 35%, 25%, 20% and 20% would be the most appropriate precision. I would write: “All mechanisms of gap formation accounted for a substantial share of the gaps created (from ~20% for standing dead trees to ~35% for branch falls). In contrast, contribution to total gap area was highly asymmetric, with ~60% accounted for by tree snapping, and <10% by standing dead trees.
l.317-321: As before, I would focus much less on the p-values. P-values usually tell us very little (for large sample sizes they are always < 0.05, for small samples, very rarely). How about effect sizes? Why not simply describe what you see in Figure 6a and 6b? From the graph it is clear that most gaps are very similar in size, irrespective of gap formation mode, but that snapping in this case has much more variance. So I would describe that. And from 6b, we see that all three tree fall modes have higher biomass loss than branch-fall, which also makes a lot of sense, so I would describe this! It would be important to add one or two panels showing the correlation between gap area (or gap perimeter) and biomass loss as a scatter plot. I.e., x axis is gap area/gap perimeter, y axis is biomass loss, and dots are coloured by mortality mode. I think what most people would be interested in is: if I measure the gap area and gap perimeter, how well can I predict the biomass loss?
348: The header for this whole paragraph is a bit odd, because this point (sensitivity of detection, mechanism of gap formation) is barely explained and probably factually wrong. The mechanism of gap formation itself is likely not dependent on detection, because it is a property of the forest ecosystem itself. I assume that the authors mean “The classification of gap formation methods is related to detection sensitivity”. But again, this is also not explained, and the section as a whole lacks in coherence.
348-350: By definition, the approach does not only underestimate the frequency smaller than the size threshold, it does not quantify it at all, so I would remove this.
351=354: The caveat about needing more studies across different landscapes might make more sense later in the discussion, but not an urgent alteration.
357-358: This needs to be explained more.
385: Again, the header is difficult to understand and does not summarize all the points that are discussed in the paragraph (GSCI, power law vs. lognormal). I would shorten these paragraphs a lot and clearly say what is essential. E.g., lines 425-430 seem to describe how the lognormal function is fitting the data, but it is very hard to follow. Also, as stated before, I would not focus too much on these results. I would just state the simple things: Lognormal was a bit better. This may reflect underlying processes, but should not be overinterpreted due to methods and small sample size. I don’t think the results allow for much more inference.
431: This header also does not make much sense – is this really what your data show? A scatterplot of biomass loss vs. gap size would go a long way in explaining this, so I strongly recommend to make this clearer both in Results and Discussion! If I look at Figure 6b, some of the strongest biomass losses are recorded for tree snapping, whereas branch losses predictably have a lot of very small biomass losses.
431-448: The whole paragraph seems to summarize more the literature than relate to the results.
449: The final header seems misleading. First, as outlined before, the sample size of the data makes this analysis problematic, but more importantly, if I understand the results correctly, there were no strong correlations in the data, so how do we come to the conclusion that “Extreme rainfall-events control gap formation”?
474-475: As stated in your abstract and in line with previous comments on sample size, the study cannot “reliably assess landscape patterns”. It can only assess local patterns and indicate what might hold elsewhere.
475: “Mechanisms of gap formation could only be distinguished in the field.” To my knowledge this has not been shown in the study. If this is an important result, then it needs to be shown, e.g. by plotting not only gap area against gap formation mechanism, but all gap properties. |