Both reviews were very much favorable of ms bg-2016-365 and the author reply is reasonable (with maybe one exception detailed below) and the new figures will contribute towards improving an already excellent manuscript. I am thus inviting the authors to submit a revised version.

The one issue where the authors did not entirely convince me regards the comment on energy balance closure raised by reviewer #1. True that the approach the authors have taken does not assume energy balance closure, but as the energy balance is not closed at their site, this indicates that at least one of the energy balance terms is underestimated (H, LE, G) or overestimated (Rnet). Following the author's argumentation they implicitly assume that H and LE they use are measured correctly and as a consequence the energy imbalance at this site is due to errors in Rn and G (and other non-turbulent terms).

The revised manuscript nicely incorporates the changes recommended by the two reviewers and I am thus conditionally accepting it.

The issue that the authors in my view still do not get quite right is again the reasoning why they didn't adopt the Penman-Monteith (PM) combination equation (l. 21-36 on p. 7 or revised ms).

While the PM model was originally formulated in terms of LE (elimitating H), being based on the principle of energy balance closure, it can be expressed in many different ways. For example, the PM model can be written in terms of the Bowen-ration (see Wohlfahrt et al. 2009, 10.1016/j.agrformet.2009.03.015). The major difference, in my view, between the PM model and the approach the authors have taken is that the former makes an implicit assumption about energy balance closure. The original version assigns any missing energy implicitly to H (as the authors correctly point out), while the version based on the Bowen ration distributes any missing energy to both H and LE according to the Bowen ratio (which satisfies the author's reasoning that both H and LE are likely to be affected by the energy imbalance). In the latter case, LE and thus surface conductance will be larger.

So in summary I do not see why the PM model would systematically yield lower values for gs. One good reason for not choosing the PM model would be the desire not to make assumptions about energy balance closure, also because then one probably would need to think about adjusting the COS flux for a possible underestimation (if the cause for the energy imbalance causes an underestimation of the COS flux as well).