There's some logic with the methods that needs to be resolved - like, why we discuss throwing out some of the variables from the Random Forest model (GPP and Rs) on the grounds that they are modeled products, but yet we keep in SoilGrids and the mycorrhizae rasters, which are all modeled products themselves. There is a risk that future reviewers will not understand this and may ask why SoilGrids and the mycorrhizae rasters were included. What was the original rationale behind introducing these data layers, if they were not later used in the analysis? You might think about removing all mention of these data if they are not used in the paper at all.
Clarification is needed on data sources for the simple linear regression models versus the random forest models. I believe the former use data that are all from the SRDB, right? Whereas of course the RF models use raster data derived at the lat/lon of the soil respiration measurement. I think if this is the case, the methods can be re-organized just a bit to make them easier to follow.
Related to the second point, including information on Koppen climate zones does not seem to provide any new insight, and I wonder why we are including this information at all, especially when MAP/MAT is known for the sites as well as ecosystem. The regressions that are broken out by climate zone in the supplementary info seem to just confuse the big picture of the results. Someone might wonder why the regression with EVI and MAP was separated out by climate zone, because climate zone is probably the driver of the variability in EVI and MAP – and you need that variability to fit the regression.
Global C
Reviewer recommended
SRDB
Two summer intern students are working on this now?
I can involving as a data checher now?
I see there are a lot is going on there, wan to talk and know more.
Does those intern student could help updating the equation section? I want to update this part of data, and a following up analysis could combine to the "Time for warm" analysis.
Point 1: It is easy to remove all content about why not including Rs and GPP as a variable, however, how about if the reviewer asking one of our main focus here is the relationship between Rs and RC? And I believe we do not discuss this point in the first version of this manuscript, but one reviewer have a comment on why we those variables were selected as predictor in the random forest modeling.
Then how about we move those to SI? Or we cound slightly change the reason we do not including those two: 1) for Rs, it is because the global Rs data used a similar variables in the random forest modeling as here; 2) for GPP, we could just remove it.
Point 2: This is a great point, and help improve the readibility of the method section.
Point 3: I could re-analyze without the Climate as a dommy variable in the linear regression.
Rs partitioning
Due next week
There's some logic with the methods that needs to be resolved - like, why we discuss throwing out some of the variables from the Random Forest model (GPP and Rs) on the grounds that they are modeled products, but yet we keep in SoilGrids and the mycorrhizae rasters, which are all modeled products themselves. There is a risk that future reviewers will not understand this and may ask why SoilGrids and the mycorrhizae rasters were included. What was the original rationale behind introducing these data layers, if they were not later used in the analysis? You might think about removing all mention of these data if they are not used in the paper at all.
Clarification is needed on data sources for the simple linear regression models versus the random forest models. I believe the former use data that are all from the SRDB, right? Whereas of course the RF models use raster data derived at the lat/lon of the soil respiration measurement. I think if this is the case, the methods can be re-organized just a bit to make them easier to follow.
Related to the second point, including information on Koppen climate zones does not seem to provide any new insight, and I wonder why we are including this information at all, especially when MAP/MAT is known for the sites as well as ecosystem. The regressions that are broken out by climate zone in the supplementary info seem to just confuse the big picture of the results. Someone might wonder why the regression with EVI and MAP was separated out by climate zone, because climate zone is probably the driver of the variability in EVI and MAP – and you need that variability to fit the regression.
Global C
SRDB