Thank you for the conscientious responses to Referee comments. Please upload a revised manuscript for further review and I will send it to Referees to ensure that their comments have been addressed adequately.

Sincerely,
Paul C. Stoy

The manuscript as written is publishable with minor additional improvements that can be addressed in a brief response. The referee has a few minor additional points copied below; addressing these will in my opinion make the manuscript publishable in Biogeosciences.

Sincerely,
Paul C. Stoy

The authors have taken my comments (and those of the other reviewers) into account and made changes that have improved the manuscript. In particular, I think the new biome-level analysis is a useful addition.

The authors have added additional discussion in response to my major comments. This has improved the manuscript, although I think some more consideration is needed of two points.

(1) Using the mean of a sample to scale-up to the continental mean
I am happy to concede that an analysis exploring how accounting for climate and soil covariates affects estimates of mean biomass is beyond the scope of the study, and it is good to see there is now some discussion of this issue. However, I am concerned that this will be missed by many readers and this paper will be used to argue that existing plot networks can’t be used to obtain the continental-level mean biomass, when it actually shows that plot networks can’t capture the overall mean if the non-random nature of sampling isn’t accounted for in the analysis (i.e. the comparison of sample mean and true mean (lines 133-140) is a fine way of assessing the relative performance of the sampling methods, but is a much simpler analysis than anyone trying to scale up from plot data would do in reality).

To make sure this important point is clear to the reader, I suggest highlighting it in the abstract. Change line 21 to “Results implied that current plots networks can have clustered structures that reduce the accuracy of largescale estimates of forest biomass if a simple mean is taken of the data”. And then maybe add “and ensure clustered sampling is accounted for in analysis of inventory plot data” to line 23. This also improves the practical use of this work – we are unlikely to be able to set up a randomly distributed inventory plot network (although it may be possible to select the locations of some new plots to reduce clustering), but we can make sure we analyse the data from the plot networks to account for the spatial configuration of plots.

I wonder how much the poorer performance of clustered plot networks relates to the detrimental effect of oversampling certain areas versus the failure to adequately sample some areas. If the former is important, then could the point-pattern approach used here to generate spatial patterns that emulate plot networks be reversed to help subsample clustered plot networks to remove the effect of oversampling. If the authors agree this could be important, then perhaps it could be mentioned in the new text (lines 299-304)?


(2) Input biomass maps
The additional consideration of fine-scale variation missed by biomass maps is useful. However, I think the authors missed my point about there being large discrepancies between some biomass maps (including the Baccini map they use) and spatial patterns revealed by field data (see Mitchard et al. 2014 Global Ecology and Biogeography). I’m not sure what effect this would have on the analysis, but I think this issue should be briefly discussed.


Detailed comments (abstract only)
Line 19 – Odd wording, suggest remove “Though”
Line 20 – Change to “also applied”
Line 24 – I’d remove “This should be considered” from the end of the abstract as it isn’t really necessary and is a bit unclear what it is referring to.

Thank you for completing the additional revisions; the revised manuscript is publishable in Biogeosciences.

Sincerely,
Paul C. Stoy