Thanks for your contructive responses to the reviewers. I am looking forward to your revised manuscript.

Thanks for your careful revisions. I only have a few minor technical issues before this piece can be published:
1) General comment: the structure is not always clear, there are methods in the intro and also in the result section (e.g., section 3.4), please check that the methods are all in the methods section and so on …
2) P1L26 and elsewhere: the term verification (establishing the truth) is not suitable here, validation would be better (or confirmation/testing)
3) P1L33: 102% and 123% would be clearer here (a few lines above you use 123% to refer to an increase)
4) Eq 1&2. Please clarify whether other functions had been considered in K2017b
5) Could you add a photo from each of the sites (also showing the surroundings)?
6) P3L29: were filtered (data is pl)
7) P3L36: why symmetric around zero degrees, the snow/rain threshold usually is above zero degree
8) There are a lot of abbreviations, are really all of them needed?
9) P5L11: isolate? Other word? Unclear what is meant
10) P11L17: “NSE has a higher sensitivity to bias and outliers than (r)”, note that r has no sensitivity at all to biases, and can you explain why NSE is less sensitive to outliers? What does an NSE=0 imply in this case (compare as good as mean runoff for runoff simulations)
11) P11L22: Unadjusted NSE: I assume you mean NSE for unadjusted precip measurements. As there are adjusted correlation coefficients and NSE, your formulation (also post-adjusted NSE in L25) are misleading

Thanks for your revisions. While I am happy with most of them, I am less convinced by the responses regarding NSE. Going back, I realize that NSE was included based on my initial comment. What I was concerned at that time, was that using only Pearson r could result in misleading results as a perfect value of one actually could be achieved with large errors. For streamflow this problem is usually solved by using NSE (or KGE) instead of Pearson and, thus, I suggested to do something similar in your study. However, precip data are different than streamflow and NSE might not be the best metric. If you keep using NSE you need to address the issues I raised in the previous round. Alternatively, you might use another metric. I am sorry if there was a misunderstanding, I did never mean to force you to use NSE, but I wanted you to use a metric where the perfect value (of one) ensures that the fit is perfect (and without any bias).

Thanks for these revisions. While I honestly saw here a chance for a more creative performance measure, exchanging the NSE for PE at least removes the issue of using NSE with precip as discussed before. Judging the values of PE might be a bit tricky and using some benchmark values to make results more comparable would be valuable, but I leave this to the authors to decide.
What I would strongly recommend, however, would be adding an equation to better describe the PE statistic.