The manuscript “Exploring different dominance patterns in western Amazonian forests from a functional trait perspective” analyses the functional correlates of species dominance - what are the functional traits of dominant species in comparison with non-dominant species, how this differs by forest type and for dominance based on local abundance or regional frequency. The theme of this ms is very welcome, as the reasons for dominance across the Amazon have not yet been solved. Thus, I think that a revised version addressing the following points would make a great contribution to the audience of J Ecology. My main concerns with this ms lies on the data and the analyses.

Data. The trait database tends to be biased towards dominant species (in a few cases the opposite) and only 34% species had values for the main five traits (and even less for 6 traits). I understand that extensive traits databases are yet to be built, but given the limitations described above, the analyses need to be strongly improved, as discussed below.

Analyses. I see two potential problems in the analyses:
1) The unbalanced design. The null models were built to deal only with unbalanced distribution among forest types, but there is a potentially huge imbalance of the “species with traits” across phylogeny and the abundance rank distribution that would need to be taken into account. The number of dominant species with traits tends to be larger than that without traits, which likely shrinks the space of variation of traits across the non-dominant species. Also, if traits are disproportionally sampled for some lineages, this could also change the functional space.
2) The comparisons of dominant vs non-dominant within forest types. Authors have tested how traits of dominant species of one forest type differ from the traits of all the species across the pool of all forest types. This doesn´t seem to be the correct way to test how a species becomes dominant in a specific condition, which would require testing how traits of dominants differ from traits of non-dominants WITHIN the forest type. The test applied by authors test more broadly a “filtering effect” of forest types against the complete species pool. I guess that even non-dominant species of a certain forest type will differ from the complete species pool, as detected in several previous studies. Thus, the test presented so far does not seem adequate for the specific question of the causes of within-habitat dominance.

There is a need to improve the presentation of the rationale and hypotheses (page5), as there are a few inconsistencies.

Specific comments:
Title can be improved, for example “testing how species traits affect regional or local dominance across western Amazonian forests”. Bringing the functional perspective to the title increases the potential interested readers
L. 23 was seedling survival measured to allow the affirmation contained in this sentence¿
L46. species spatial turnover is not an ecosystem process
L.50 “terra firme forests with well-drained and relatively more fertile soils than other upland habitats” – this is definitely not true. Soil fertility varies enormously across “terra-firme” forests, and a large fraction of them are very poor when soils are derived from the old sediments of the Guianan and Brazilian Shields. If the sentence refers to all of the Amazonian region, then the description of environments must cover all of it, not only the western region.

L.69 “may be dominant in different ways”. I suggest “may become dominant in various ways”

L70. “on the one hand, greater dispersal, stature and establishment abilities” – I suggest to list the traits in the order of the potential mechanism – “greater stature providing greater dispersal capacity, and large seeds providing higher establishment ability”.

I wonder how high stature and seed mass could go together, since higher stature provides better capacity for dispersal of small seeds that go on the wind, but large seeds that depend on ground disperses for long distance travel don´t need to be produced in tall trees – and actually many are not!

L.72. “whereas on the other hand, rapid growth rates, competition and defense could confer higher abundances at local scales”. Well, this sounds strange to me. Rapid growth tends to be associated to the need of more light and thus to canopy gaps that are spread in the landscape, generating the expectation of higher frequency of fast-growers in the large scale, but small local abundances.

L88 to help to identify - to help identify

L95. This description doesn´t make clear what a large leaf area means for the plant economy, and if it is aligned with the conservative-acquisitive axis.

L. 96 again, not clear what these traits represent in relation to the conservative-acquisitive axis or to other ecological axes or strategies.

L. 99. Same as above, and also – what “both dimensions” are these¿

L 118-123. Basically the hypothesis is that environment filters traits and species with traits that better fit the environment become dominant. Would be good to state the hypothesis in a mechanistic way such as that.

But there is a problem, because there are many more specie than there are functional strategies, and only a couple of species with a specific strategy become dominant.

L 146. These plots are VERY small, I wonder how many individuals are contained on them and thus how representative of the species abundance distributions they can be.

L 173. Who are the “their” in this sentence¿

Table 2 can be in the supplementary material.

L181. How standardized where in fact the traits data¿ I strongly suggest to include specific descriptions of the filters applied to make the data comparable, since in my own experience there is a lot of non-comparable data in some large databases. e.g. SLA may derive from measurements of whole leaves or leaf punches, with or without petioles. and in the case of compound leaves with or without rachis – this all of these differences lead to strong differences in values for the same species, in some case the species might move from one extreme of the trait spectra to the other! Another source of error might come from traits measured in environments that are very different from those being evaluated – e.g. wood density measured in poor soils would not be the same for plants measured in fertile soils, for the same species.

L.189. Just one measurement for species was accepted to have a trait estimate¿ I hope not, and this needs to be thoroughly described. At the very minimum 5 samples should have been used to derive a trait estimate.

L 192. “We log transformed all trait values”. Why¿ There should be a strong reason to transform any data. Actually, the most accepted approach nowadays is to use models that deal with non-normal distributions instead of transforming data upfront.

L. 204. I hope the PCA was based on standardized traits (i.e. using correlation as the base for similarity calculations), but this is not stated. Please indicate the settings used on this PCA.

L.211. Here it is not clear if a different PCA per habitat type was run, and then the MANOVA applied over that. If not, how the variance partition was constrained to that within habitats¿

L.212-13. This is very concerning. What is the distribution of missing data among dominant and non dominant species¿ and what is the distribution of present and missing traits among lineages¿ The number of species with traits is very small, and this severely limits any conclusion from this analysis
The null models were built to deal only with unbalanced distribution among habitats, but there is a potentially huge imbalance of the “species with traits” across phylogeny and the abundance rank distribution that would need to be taken into account.

L. 268. The figures based on the sampling correction should be the main one, we can´t tell what the other pattern means. And still, as explained above, other corrections must be applied. I recommend that the main figures should be those based on all the necessary corrections for the unbalanced design.

L 277-281. Are these the results based on the sampling correction? Neither the text or the figure legend make it clear.

The ms has certainly improved from the previous version. The authors changed some analyses according to the suggestions (which were kind of similar among reviewers!), what has simplified and strengthened the msg. However, there are still points to work:

1. The second block of analyses uses mixed models with a random factor for species. I suppose the repetition of a species is across habitats, but this is not clear. If so, there would be many instances were the species is present in only 1 habitat, what would pose a problem for the mixed models. This needs to be clarified

2. The first block f analysis is based on the relative abundance of species at the regional level. The scale therefore is regional, but authors many times equate this analysis as if it represents a “general dominance metric” not linked to scale. This is problematic because it makes the interpretations and conclusions confuse. Need to rebuild some parts of the text to better reflect that

3. Since part of the second block of analysis also deals with the regional scale, there are two assessments of that but the discussion does not merge them, and ends up being repetitive and somewhat confuse

4. There are some statements about environments that are not correct, and some misuses of the literature, both of which need to be corrected.
   I have pointed all the necessary correction points along the marked pdf file.

I congratulate authors for the good job done on improving several points indicated before. Now that the technical problems pointed before were solved, and also the text description of the methods has improved, all that is left now is to improve is the text writing. I think authors can improve the text in these main points:
1) The hypothesis #2 says that they do not expect a group of traits (DBH, SM) to affect local dominance and a group of other traits (SLA, LA, LN, WD) to affect regional dominance. This is depicted in Table 1 (there are some signals there that need to be corrected according to the text). But then in the results, they present tests of relationships between all the traits and each scale dominance. It seems kind of a fishing expedition to find significant results, which I think is unnecessary, dangerous (too many tests lead to a high type I error rate) and lead the text to a lot of repetition, i.e. lack of focus. I suggest they focus on reporting the results of the original hypothesis test, which is strongly grounded on ecological theory.
2) Since there are 3 groups of tests (on each of 3 dominance metrics), it is easy to get lost in the results and discussion sections. I strongly suggest that authors: a) organize the presentation of results from the PCA focusing on the strategies, instead of in the traits, since there will be also description of univariate trait descriptions after that; b) present the local dominance results in one paragraph, and regional dominance in another; c) discuss the results invoking first the proposed hypotheses, and showing if results agree or not with it, and evaluating the strength of the results across all forest types at once. In the present format the discussion is fragmented by forest type and it is hard to see the overall picture. Would be good to have a table that summarizes the results comparing them to what was expected on table 1; d) discuss the strategies more than specific traits, since the fact that one or another trait has been found significantly related to a dominance metric may depend on the amount of trait data and its distribution, and may not mean that the specific trait is fundamental…
3) There is too much repetition in the discussion, would be good to focus and shorten. Hopefully the suggestions above help that.
I have also included some detailed suggestions along the text, please see the pdf. I tried to help in improving some sentences, but didn´t have the time to go over all the text. Would be good to ask that help from someone else.