In general, I agree with your response, but for some of them you do not suggest to change the manuscript accordingly. I would urge you to address at least the following in a couple of sentences.
Response to reviewer 1.
Spatial distribution of uncertainty
I would urge you to at least include one or two sentences explaining why you chose not to present the spatial distribution of prediction uncertainty.
Legacy data
The same applies to the comment on legacy data correction. Again you do not have to correct the data for legacy effects, but it is a bit late to bring up this issue in the conclusions. One or two sentences in the discussion section would be appropriate.
Soil density
Regarding the comment on the soil density. I agree that this is not a well-defined term, and some confusion might arise with particle density. I would suggest using the term ‘fine fraction bulk density’. I fully understand that you, as many others, do not have measured rock fragment contents, and therefore use an approximation.

Reviewer 2
Large sets of co-variates
Could not you add a sentence on page 2 line 21 stressing the efficiency in terms of pre-processing the covariates. There will certainly be readers that reason in the same way as reviewer 2.
Separate model for sand
Reviewer 1 also asked this question. Could not you add one or two sentences to justify why you did not choose to predict the sand content?
Evaluation of overfitting
Please carefully review this section, as you suggested.

Soil function assessment
Please add a phrase and a reference justifying the choice of the soil properties.
Covariate interpretation
Please make sure that the readers understand that this is only an example to illustrate the interpretation of covariates.

I am impressed with the very thorough revision. The paper reads very well.

Dear Madlene,

Thank you very much for your continued support of SOIL and doing such a good job on the manuscripts! It is with pleasure that I accept your paper for publication in SOIL.
best,
Jo