Dear authors,

Your revised manuscript is sent for another round of reviews (to the same referees who reviewed your first draft) as the referees suggested major revisions. I will consult the reviewers on whether you addressed all their key concerns and revised the manuscript accordingly to the point that the revised manuscript is now acceptable for publication.

Best regards,
Valentina

Dear authors,

Your revised manuscript was sent for another round of reviews (to the same referees who reviewed your first draft) as one of the referees asked for major revisions. I've now heard back from this referee (see the comments below). Based on these comments, I ask you to implement the minor revisions and, subsequently, upload the revised manuscript which will then be accepted for publication.

Best regards,
Valentina

This is my second assessment of Bolibar et al, 2019. I had several main concerns about the first version of the manuscript. Some of them were addressed by the authors, and the revised manuscript is much easier to follow and understand. I now also better understand the methods used in this study. 

Some of my other concerns have been argued against, and resulted in no changes in the manuscript: mostly, my concern about targeting the prediction of glacier wide SMBs with the model. The argument used by the authors is based on "leave one year out" validation results and other predictor importance considerations which, I believe, do not allow to estimate the true effect of changing geometry on long (i.e. > decades) time scales. Furthermore, it must be noted that the "full SMB model" (predicting SMB based on *modelled* geometries) is only validated in Fig. 11, and only indirectly (via the area) i.e. without comparison with the reference SMB data (Figs. 6 to 10 still use known changing geometries, giving an advantage to the SMB model that it won't have outside the validation period). 

Since this model is going to be used "as is" to provide future projections (and without starting an unnecessary flame war, this is still a nice paper), I would recommend the authors to mention these two points in the discussion section (maybe section in 4.3).

Finally, I spotted a few typos/errors which, I believe, will be corrected by Copernicus' language editing service. 

Looking forward to see this paper published soon!

Dear authors,

congratulations on the great job in revising the paper and for introducing the deep learning into research on glacier mass balance.

Best regards,
Valentina