See attached comments.

This paper presents a thorough set of measurements and calculations to estimate contributions of permanent snow and ice (PSI) features versus seasonal snowpack contributions to late season streamflow. It is a long paper, more monograph in scope than is common in recent years.  Kudos are deserved for assembling so much information bearing on an important question for this region.  There are some novel ideas and analyses presented as well. 

While it is great to see all the information brought together to be able to more thoroughly understand it, the document did not feel particularly coherent at times, with deep dives into details (agreeing here that they are important details) making it difficult to see where the arguments and calculations are going.  It’s all there, but the thread that keeps the reader’s attention clear on why they are reading the particular paragraphs and sentences is lacking.  This is only offered as feedback for the authors and whether they want potentially many other readers to experience the article in this same way. 

There may be some choices that add to the challenge of the long paper with many details that the authors may wish to reconsider.  For example, in at least two places, there is presentation of data using two different sets of PSI features.  One set is just larger features with relative permanence, e.g. buried ice, rock glaciers, perennial snowfields, and glaciers (as shown in in Figure 2).  Then there are additional smaller features grouped as “small snow patches” and “semi-annual” features.  Choosing one of these might be helpful as it can become difficult to know which PSI set is being invoked in any given analysis/figure later in the paper, although sometimes both are.  It’s just one less thing that the reader needs to track. There are layers of classification schemes, so it might be helpful to think about how many are necessary for the primary messages of this document.

In section 4.3 a key question that seemed to be approached from multiple angles, but not quite answered directly, is whether the PSI and deep snow are essentially in the same places.  I could see this being addressed in a straightforward way either by giving a CDF of SWE across the whole area in a watershed compared to a CDF for the PSI areas or by plotting a logistic regression of probability of being a PSI pixel versus SWE.  Maybe this would be appropriate to do by type of PSI, as only certain types may be correlated with greater depth. Closely related, is that there is some discussion on lines 566 and 567 where sentences are made about differences of snow in classes of all, >1 m, > 2m.  It might be worthwhile just to plot the cumulative density function of SWE in the 5 watersheds and indicate which are west and east.

Figure 10 in section 4.4 seems to be attempting to demonstrate that streamflow differences reflect a similar message as in section 4.3, and I’m wondering how strong the evidence from this analysis is.  Most of that argument seems to be encapsulated in Figures 10B and 10C and it is a little difficult to follow the details of the argument.  10B just looks to be a daily recentering and rescaling of flow in each watershed relative to the mean for each watershed during Aug and Sep, where the rank is Dinwoody, Torrey, Bull Lake, Pine, and upper Green, which matches the ranking in percentage in PSI area.  Important to the argument, this ranking changes little if the loss from the PSI is removed from the Aug-Sep flow totals.  What is left unclear is what produces the difference in runoff across watersheds.  For instance, unit-area runoff from Dinwoody is the second highest rank annually, and unit area runoff from the Upper Green is lowest annually as well.  Only figure 10A lets you know that there is a pattern in timing difference (essentially pdfs of flow timing), but the ranking in Aug-Sep in 10A does not match the ranking in 10B.  There is a decent negative correlation between unit-area runoff and basin area for Yearly, Jul-Sep, and Aug-Sep periods.  There is a similar but slightly weaker negative correlation between PSI fraction and basin area.  A story that one could tell is that most of the snow accumulates in the higher elevation areas, and as basin size increases, the runoff generated from those higher elevation areas is diluted when reporting unit-area runoff, whatever time of year you are looking at.  That is clearly not the whole story, but it is one more dimension of watershed difference that could be brought up to explain differences besides focused SWE accumulation and PSI. I would suggest a tighter argument for this section.  It is also not clear what the reader gets from seeing a time series in 10B and 10C, when the seasonal average could be used (all lines are slightly curved but nominally flat relative to the interbasin variability), either in a table, or in a single bivariate plot of the mean values for each stream from 10B and 10C plotted against each other (to show that there is at least rank correlation).

In a related vein, the word “groundwater” does not appear in the paper.  I’ve spent some time in the wind rivers, and there is indeed a great deal of old crystalline rock, but there is also a fair bit of sedimentary rock within the basins outlined by the gaged watersheds, substantial fracturing of the Precambrian batholith, and some famous and productive springs. The subject of within-season timing delays driven by transit time through groundwater has been popular in the hydrology literature in recent years (see e.g. Somers and McKenzie 2020 for some review in the context of mountains, snow, and glaciers). Is there any evidence or signal from the hydrographs that could clearly separate groundwater storage effects from snowpack storage? Again, I’m not doubting that differences in snowpack heterogeneity/PSI are a primary driver, but in the spirit of thoroughness and honoring a prevalent hypothesis within the literature, a short paragraph to acknowledge this perspective might be useful.

Figure 11 may address some of the problem seen in figures 10B and 10C, as the area is in both numerator and denominator, so it just describes the ratio of runoff on a particular day to total amount of snow above some threshold.  The problem is that the only interpretation that we draw from it is that differences across the basins in the metric plotted (full snowpack and three other thresholds) shift in time as the threshold SWE divisor shifts.  The initial broad grey bar in the “Full Snowpack” plot is not convincing, as +/- 50% is not all that equal in value.  Consequently, the notion that “total snowpack volume controls freshet magnitude” (line 749) is not well supported. The freshet magnitude may be more related to the elevational distribution of snow covered area and how quickly the snow covered area drops with melt.  More importantly, reserving interpretation of these graphs to places where all of the lines cross for different threshold values seems just to say that the deepest snow accumulations (>90^th to 95^th percentile) are a control on the differences between the 5 basins. Couldn’t we just take the top left plot of Figure 11 and plot a mean August “runoff efficiency relative to initial snowpack” versus metrics of snow heterogeneity, like the volume of snow above the 95^th percentile?  That seems like it would be a more direct illustration of what you are trying to say.  It also looks like you would get a similar answer but with less pronounced slope when considering the counterfactual (lower left).  This is an approach, where seeing a great deal of variability in outcome, one can use covariates to explain that variation.

This manuscript represents an impressive effort in data reduction to generate variables useful in a number of analyses that are threaded together.  It is exemplary in some respects in terms of serving up multiple perspectives on the snow to flow problem.  For the curious and persistent, there is much to see and it is relatively transparent.  That being said, there is so much presented, some with much detail (maybe more than needed in some places), that it is a bit overwhelming to take in and be able to understand the effects of choices made during the analysis.  I’m wondering if there are opportunities that could just use simpler bits of data from Tables 1 and 4, and Figure 9 for bivariate plots using the 5 basins to articulate the differences among the basins, and better pin down the contribution of snowpack variability and abundance of PSI.

Thank you for the opportunity to share some thoughts about my experience reading the paper with the authors.  Hopefully they can draw on the reflections to help them better communicate what they wanted to communicate.

Reference:

Somers, Lauren D., and Jeffrey M. McKenzie. "A review of groundwater in high mountain environments." Wiley Interdisciplinary Reviews: Water 7, no. 6 (2020): e1475. doi.org/10.1002/wat2.1475