Dear Authors, I have the comments back on your revised m/s which was reviewed very favorably by 2 of the 3 reviewers. One reviewer (report 3 included below) still has concerns about the way you have formed your conclusions, and I ask you to provide a detailed response for my consideration as this has a clear impact on the take home messages from your paper. Another reviewer noted a number of typos so please check these also.

Best regards, Ross Brown (ed.)

Report 3:
General considerations
The authors have satisfactorily addressed most of the comments of the first review. However, they still have a hard time to give up the idea that the forcing model should be the reason for the bad performance of their model chain. I have reformulated my former major comment 1 and still call it ‘major’. Apart from this I think there are only a few minor issues to be addressed.

Major comments
1) Still: the authors try to ‘explain’ WRF offsets (Section 3.1, point inter-comparison) by ‘wrong COMO input’. It is (still) argued that these offsets were (partly) ‘due to offsets in the COSMO-2 input’ (p11, l. 3/5; p.13, l. 32). As I pointed out in my first review, this is a very weak argument and must completely be omitted. Rather, the performance relative to COSMO – or to the coarser WRF simulations - should be used in the discussion of possible reasons. Fact is that for none of the variables, there is a clear trend to obtain better overall correspondence to the observations when increasing the model resolution. Sometimes, (e.g., 31.1., Dischma moraine, temperature) indeed the 50m grid spacing is ‘a little better’ (and it seems to be systematic from coarse to fine), sometimes (e.g., 5.3., Dischma moraine, RH) COSMO overestimates, and this is reduced the finer the resolution becomes (leading to an underestimation at 50 m grid spacing…), sometimes (31.1., 4.2., Moraine, wind speed) COSMO overestimates and the high resolution makes it worse, and sometimes (31.1., 4.2., FLU2), wind speed is astonishingly good in COSMO – and higher resolution makes it (much) worse. So, what one, first of all, may conclude is, that it NOT the input that determines the high-resolution performance. Second, one may state that T, RH and WD are ‘reasonably wrong’ (in all resolutions) and might be explainable to some degree (as the authors do). Wind speed, on the other hand, is sometimes unreasonably wrong and therefore deserves a somewhat more in depth discussion. Again (as in my first review), it first should be stated that only if the simulations are adequate (spin up time, distance from the (upwind) boundary, …) it makes sense to investigate the discrepancies (the authors do this quite briefly on p.13, l. 33). Second, it should be stated that the improvement is not systematic (see above), i.e. not ‘better with finer resolution’. Rather, the varying performance seems to indicate that we are dealing here with compensating errors. Another aspect which is clearly evident (and it also shines through in the ‘explanations’ by the authors) is that the scales of the model and the observations do not match. It is rather difficult to address this issue in the observations, but what can be done is to estimate the area of influence (footprint) to judge how this length scale relates to the grid resolution. As for the model, I recently came across an interesting ‘grid point ensemble’ approach (Goger et al. 2018), which could also be tried here.
I think the authors should abstain from the (quite often observed) desire to ‘have a good model performance’ (p14, l.8). Very often, the bad performances are much more conclusive and advance our knowledge much more than calling a bad result good. What we can learn from this exercise is that finer resolution does not necessarily lead to better ‘point-performance’ in complex terrain (certainly not if the necessary conditions are not respected); maybe another outcome can be what the scales of observations and simulations are, and finally that wind speed is much more challenging than other variables (since one can be in a completely different flow regime with the measurement than with the model) – and additional research is necessary to address these issues.

Minor comments
P3, l.26 the question should read: ‘To what degree is snow precipitation variability represented by…’
P7, l.4 ‘…are used for the correction’: I still consider this inappropriate to be called ‘a correction’. I think it would be appropriate to say that the meteorological information (wind speed, friction velocity) from the first model level is used for the interpolation to the actual height of observation.
P9, l. 24/26 is the semivariance the ‘gamma’ (as on l. 24) or the ‘gamma hat’ as in the equation?
P10, l.25 if the sentence introduces all variables, the reference to Fig. 2a-c, Fig 3a-c is wrong, since these panels only show temperature.
P13, l. 24 ‘for the two events…’: the same is true for the third event, for about 6 hours around noon at Flu2….
P17, l. 8 this supports the hypothesis (no confirmation at all)
Tab 2, Fig 6, 7, caption: WRF simulation at which grid spacing?
P25, l.6 this is likely not a good summary of why the wind speed is overestimated (but not always), see major comment 1.
P25, l.15 ‘likely due to the high model resolution’: this is a quite misleading argument. The high model resolution is more realistic. Thus, when the ‘high model resolution’ produces too much precipitation, this indicates that the lower model resolutions were likely wrongly tuned to overcome an underestimation due to coarse model resolution.

Reference
Goger et al: 2018, BL Met, 168 (1), 1-27, https://doi.org/10.1007/s10546-018-0341-y

Dear Authors: thank you for the detailed responses and I am now pleased to accept the paper for publication. One minor quibble, the last modification you provide "among other causes it may likely be due to the high model resolution and therefore more complex topography and higher mountain peaks compared to common high resolutions simulations.” (P. 26, l. 6) does not read very well and does not convey the reviewer's point that this may be related to lower resolution simulations being overly tuned to topographic influences. Something to think about when you do your proof reading of the final version.

Congratulations, and thank you for your support of The Cryosphere.

Ross Brown (ed.)