Summary

The authors compare pollen-based reconstructions of Holocene vegetation (REVEALS) with simulated vegetation from an Earth System model (ESM) and a dynamic global vegetation model (DGVM). They find that the simulated decline in Holocene forest cover is too abrupt and late in both models and attribute this to errors in the land-use history applied in the models. The general underestimation of forest fraction in the ESM is attributed to the parameterisation of disturbance.

Recommendation

This is interesting study with an important finding which is very relevant for this journal and which is strengthened by the use of two very different modelling approaches. This and the level of detail is a strength. My main comment (see below) is that the paper is overly long and the authors should aim to substantially shorten the text in several places in order to make it more accessible.

Main comments:

1) I appreciate that this is a technical paper probably mostly aimed at those within the field, but I found the text in places hard to digest. In the methods I would recommend that you include some summary paragraphs at the start of the section 2 to very briefly explain the differences between MPI-ESM, JSBACH, LPJ-GUESS and REVEALS. More generally I believe that you can make the text more concise, especially section 2 (currently ~ 6 pages), section 3.1 (2 1/2 pages), 4.1 (3 pages) and section 4.5 (2 pages). Some of the detail in section 2 could also optionally be moved to an appendix.

2) Can the authors comment on whether other parameterisations besides the ones explored (disturbance and windthrow) could contribute the biases in the forest cover simulations?

Minor Comments

Line 19: specify that LPJ-GUESS is the dynamic vegetation model.

Line 297: I could not see a figure with these results, please can you check this?

Figure 5: could you add a key to this figure to make it easier to read?

Lines 588-600: It is probably worth clarifying here that the KK10 reconstruction has substantially earlier and more widespread land-use than LUH2.

Line 615: I cannot comment on how topography is treated in REVEALS. As written this sounds a bit uncertain. Can you confirm that topography is not accounted for and how this would influence the model results. At the moment it sounds as if you’re unsure whether it does or not.

Line 705: This seems to contradict the sensitivity test with increased disturbance you present earlier, can you comment?

Line 717: This seems to echo results by Kaplan et al 2017 (Land). Is that correct? If so, can comment on this here and earlier in the text?

My comments are attchaed below in PDF.