Scheiter et al present an interesting approach to delineate biomes based on traits information. They use a combination of a number of extensive databases and interesting approaches. Even so, I have a number of major and minor comments.

Major comments:

- A main concern is that some components of the methods do not logically link to each other, which does not help the story line. This specifically applies to two analyses; analyses described in section 2.4 and 2.6.

a. The aim of the analysis in 2.4 and how it contributes to the story line remains unclear to me. You create, for each of the four cases (?) a trait set to work with. First of all, I don’t understand why that is needed because you already tested the performance of different complete trait sets in 2.3. So, why would you continue working with subsets of traits if you already know how the entire combination of traits works? Now, you seem to throw away most of the information presented in 3.1 instead of building upon them. That does not only seem unnecessary, but also creates biases (because subsets of traits are used; an argument you make yourself in l. 275). Moreover, given the different selection criteria and procedure for different trait sets/cases, interpretation of these results is made even harder (not possible). Also, I don’t see the need for doing it. You use it in 2.5 (but then only for the case 4), but I think that also could have been done directly on the best performing trait sets from 2.3. To me, that would have been a much more direct way to test it and with fewer biases.

b: I don't understand why the analysis in 2.6 is useful to do given that you started with using trait maps to fit biome distributions based on complete biome maps. Why would you then derive additional trait maps, using a different procedure with other input data than used elsewhere in the paper. Moreover, you already have 31 biome maps to test performance against. So, why create another one? I would say the study is about understanding biome distributions based on traits (a story already told by the kappa values) and not about predicting/extrapolating biome distributions. In other words, the aim and position of this analysis in the story line is not clear to me.

- Another methodological issue is that I do not understand why the authors thought it important to first use the traits in combination with species distribution models to make trait maps and then to couple those to the biome maps instead of using the locations of the trait observations (g. using the original TRY data, possibly aggregated to the 0.5 degree pixels of the locations) directly? Coupling individual locations can be done to calibrate and validate biome models and avoid the major uncertainties involved in creating/ extrapolating the trait maps, i.e. i don't see that necessity. I wonder to which extent these uncertainties contributed to the low Kappa-values of the predictions.

- I had hoped that the authors would have focused more on the ecological interpretation, rather than on the methodological aspects.

a.: This already starts from the presentation of the data. With slightly different analyses and visualization, potentially a lot more ecological insights might have been gained. For instance, fig 7 presents the mean trait values for some biomes. However, these are individual traits while you tested how different trait combinations allow distinguishing among biomes. However, nowhere we learn about those combinations (their synergies and their redundancies). For instance, I would have liked a so-called confusion matrix to see how for a given trait set, biomes were predicted properly or not and whether this is consistent among the different (best-performing) trait sets: which trait combinations (in addition to single traits) lead to the best predictions? Is that consistent between trait sets? Which trait combinations allow distinguishing e.g. ‘’subtropical forest’’ (just to name one) from other forests? Is that consistent through different trait sets? I think the analysis has the potential to bring such answer and thus ecological understanding, but that remains unexplored.

b: Also the discussion section is now almost entirely focused on the methods instead of the question why the analysis is a useful thing to do and what we learn from it: I would be interested in seeing an ecological interpretation for how and why the results of the analysis help global ecological understanding of biome distribution and functioning. That would have led to a much more interesting story. In the current set-up, the discussion section does not provide much insight.

Minor comments:

1. The abstract does not help to tell the story. The method applied is not clearly explained, i.e. which steps were taken. Also the role of the 31 biome maps was unclear (in the abstract it seemed an input, while it is used to train the data. Also, a clear take home message is missing: ‘’we can make biome maps’’, but it is not explained why that is important or what is the added value of this study.
2. I would have appreciated an explanation/argumentation on why choosing these trait maps instead of other maps, i.e. what is the conceptual advantage to these maps? (Also in light of the discussion section where it Is mentioned that different trait maps could have led to different outcomes)
3. With some self-advertisement; Verheijen et al. NewPhytol 209: 563-575 also evaluated which traits could be used best to distinguish among biomes, albeit which is much smaller dataset and fewer traits (And a different method). Interestingly, similar traits popped up as important.
4. At some parts in the methods section it is unclear whether all individual 31 biome maps were tested independently or whether an aggregate was used. It seems that you tested each map individually, but a better explanation throughout the methods on this would have been appreciated.
5. The role and use of PFTs is confusing to me. You derive different trait maps depending on growth form/PFT and clustered them into four cases in 2.2. Then, it seems that trait clusters were made for each of those four cases (if I understand the methods correctly) in 2.3. Then, later in 2.3 you seem to continue with the outcomes of case4 only. This is not clearly described and also the reason why you do this, remains unclear to me. Also the role and use of PFTs is confusing to me. In general, i don't think you tested different combinations of PFTs, but trait maps of woody vs non-woody vegetation.
6. I understand you have a personal interest to compare to aDGVM2 results, but to me those comparisons do not add much to the story.
7. The title of section 4.2 is strange and seems wrong.
8. I am surprised that the kappa does not go beyond 0.6. I had hoped that higher kappas would be feasible. This is not discussed, but I would be interested (e.g. instead of the current 4.3) if such aspects of model performance would be discussed and interpreted
9. With respect to figure 3; what does it tell us if ranks of certain traits vary with number of traits (while ranks of other don't)? i.e. what is the message/understanding this figure gives?

Scheiter et al. delve into the utilization of crowd-sourced trait data for defining global biomes, investigating the suitability of trait data for biome classification and identifying the most relevant traits for this purpose. While the study is intriguing and significant, I have some comments to offer.

Major comments:

• The findings presented in this study offer valuable insights into the fundamental ecological mechanisms driving plant biogeography and global distribution strategies. However, while the results are intriguing and significant, the interpretation from an ecological perspective could be further emphasized. Additionally, it would be beneficial for the authors to underscore the importance of their study and the implications of their findings, particularly in informing our understanding of global biome classification and its ecological implications. Strengthening the ecological context and emphasizing the practical implications of the findings would significantly enhance the impact and relevance of the study.

• The findings regarding the differences in trait distribution and occurrence between biomes are indeed intriguing, but they have been explored somewhat superficially in the manuscript. While I understand that this topic is complex, questions 2 and 4 require a more in-depth exploration throughout the manuscript. As a reader, I was expecting these interesting questions to be thoroughly addressed. For instance, the observed tendency of tropical forests to have plants with greater height could be explored in more detail in the discussion section. One possible explanation for this phenomenon could be light competition. 

• It would be valuable to delve deeper into why the identified distinguishing traits are significant and what they represent in terms of plant strategies and ecological functions. Hence, you can offer insights into their ecological relevance and their role in shaping biome characteristics. Specifically, you could discuss how these traits contribute to plant adaptation to specific environmental conditions, resource acquisition strategies, and ecosystem functioning. While you have begun to address this topic, emphasizing the ecological significance of the identified traits and highlighting avenues for future research would strengthen the overall argument and underscore the importance of your findings.

• It's essential to clarify the methodology, especially in section 2.4, to improve understanding for readers.

• The conclusion and a significant portion of the discussion focusing on models may seem disconnected from the main purpose of the paper as outlined in the abstract and manuscript title. As a reader, I felt somewhat puzzled. Notably, none of the four questions posed at the end of the introduction mention the theme of modeling. To mitigate this discrepancy, you could consider clarifying from the outset that modeling will be a significant aspect of the study. Alternatively, you may choose to shift the focus of the discussion to emphasize other aspects that were highlighted earlier in the manuscript, aligning more closely with the stated objectives and questions.

Minor comments:



In the abstract the use of the “31 different biome maps” is not clear, is it used to perform the supervised cluster analyses or it is used to evaluate the results from these analyses?

Enhancing the fluidity of the introduction by providing more ecological context and emphasizing the significance of generating this type of map would improve the transition to the research questions.

l58 and l59 are more appropriate for Materials and Methods section



It's not clear in the methods why you “created three different maps for each trait”  and why you filtered the observations according to PFTs before spatially aggregating the trait values. 

In section 2.2 I didn’t understand the difference between 3) and 4).



In section 2.3, specifying the total number of traits considered would enhance clarity. If I don’t miss anything, it is only specified in the abstract.

It is unclear why the analysis described in section 2.4 was performed if the most important traits were identified in section 2.3.

Fig. 3: Parallel coordinates plots are already complex visualizations to comprehend, and as it stands, I don't think the figure adds much value, unless it is thoroughly explored and contextualized in the text.

Fig. 5: The background color (gray) on the map is too similar to the light blue, making it difficult to interpret the figure.

Why is there such an extensive comparison with the results from aDGVM2? This was not anticipated based on the main questions posed in the Introduction section.

In Section 4.2, you briefly touch on the importance of examining trait covariation, highlighting its significance compared to focusing solely on individual traits. However, given that some of your results seem to directly address this aspect, it would be beneficial to delve deeper into this topic and explicitly connect it with the obtained results. 

While I acknowledge that analyzing the occurrence and abundance of traits in each biome may not be the primary focus of your manuscript, it presents an opportunity for a valuable complementary analysis. By examining the diversity of traits within each biome, specifically by assessing the variance in their distribution, you can provide additional insights into the functional composition and ecological characteristics of these biomes. This analysis could help elucidate patterns of trait diversity and provide a deeper understanding of the ecological processes driving biome differentiation.

Writing errors:

• I believe a parenthesis is missing for the reference in the following sentences: 

1. ”Biomes are commonly used to represent major vegetation formations and to map their biogeographic distributions. Multiple biome maps were developed based on a variety of different data sources Beierkuhnlein and Fischer (2021).”
2. “Despite the increasing availability of trait data in databases such TRY (Kattge et al., 2020) and extrapolated global biome maps Wolf et al. (2022); Boonman et al. (2020), a systematic assessment of the performance of traits for biome classification and an identification of the most appropriate traits remain elusive.”