In their study, McGill et al. characterized the thermal sensitivity of streams in two watersheds of the Pacific Northwest of the United States which describes how changes in stream temperature track changes in air temperature. They characterized thermal sensitivity using the conventional method looking at the slope between air and stream temperatures. They also used a novel approach using time-varying coefficients to capture how thermal sensitivity varies through the year – this is a truly interesting contribution to the field to assess in a continuous way the seasonality of thermal sensitivity. McGill et al. then performed a clustering analysis on the annual average time series of thermal sensitivities and used classification and regression trees to identify drivers of thermal sensitivity.

Overall, the manuscript is well written and beautifully illustrated. Methods are well described and sound. While results per se are mainly of regional interest, their use of time-varying coefficients offers a methodological contribution of interest to the journal. With a few revisions, this would make a quality contribution to HESS. The main points to address are the following:

1) Clearly present hypotheses/predictions underlying the study

The manuscript identified three broad research questions (lines 82-85) and most of the results section then goes on to describe observed patterns found posteriori. Using a descriptive research approach is perfectly sound but I believe the study would be more informative if it used an explanatory research approach where the goal is to understand underlying causal mechanisms. In fact, at numerous places in the manuscript, authors talked of expected results: “we expected thermal sensitivity to increase with river size” (line 418), “we expected land cover characteristics such as open water and forest cover to be important predictors” (line 426) or “expectations of a negative relationship between thermal sensitivity and groundwater influence” (line 362). While not presented this way, it appears authors had hypotheses/predictions underlying their work. I believe framing the manuscript to more clearly present hypotheses/predictions would make conclusions of broader interest to the stream temperature community in comparison to the current presentation of results which can be difficult to interpret without regional knowledge. For example, presenting results for the Chiwawa, White and Little Wenatchee rivers (lines 270), the tributaries to the mainstem and Raging River (line 259) or the Chumstick Creek (line 415) is factually correct but bears little meaning to someone unfamiliar with the study region.

Along the same lines, authors performed three distinct cluster analyses (air temperature, water temperature, thermal sensitivity) and their goal was unclear until I reached the discussion and understood that they wished to show that thermal sensitivity clusters offers additional information to what we find when studying solely air/stream temperatures. If that was one of the goals of the study, I suggest it be clearly stated and communicated as a take-home message.

2) Conclusions need to be better supported by analyses and results.

A few statements in the results are not sufficiently supported by analyses. Moreover, the results section often lacks precision and statements are often little quantified.

For example, the abstract states that thermal sensitivity regimes “differed in both timing and magnitude of sensitivity” and while Figures 4-5 offer a nice illustration of regimes, no formal analysis clearly compared the timing/magnitude of clusters. There are a few other examples in the results section:

The manuscript states that “thermal sensitivity estimates were not entirely consistent” (line 230) although it is not clear what consistent refers to and if it was quantified. Similar wording regarding a “consistent seasonal signal” (line 243) should be revised.

The manuscript states that “only SWE displayed a relationship with thermal sensitivity” (line 233) while no formal analysis was done – a visual assessment is not sufficient to determine the presence/absence of a relationship. At minimum, R² values should be presented in Figure 3 to assess the strength of relationships. I also question here the presence of the “wedge-shaped pattern” (line 234) for SWE in Figure 3 which does not stand out clearly and may be simply due to fewer data points for large SWE values. Further analysis is required to assess the strength of this relationship.

Similarly, I question some of the relationships between variables that are discussed in Table 3. It is not clear to me that we can see a “consistent negative relationship between thermal sensitivity, distance upstream and MWE” (line 236).  Figures in Table 3 dot not present the correlation coefficient but my visual assessment is that it is likely close to 0 for MWE. The results section also points towards a “weakly positive and parabolic” (line 238) relationship between hydraulic conductivity and thermal sensitivity in the Snoqualmie basin, yet a linear regression is plotted in the figure in Table 3. Overall, many different relationships between thermal sensitivity and environmental variables appear to be weak and should be confirmed using statistical analyses.

Last, the paragraph from lines 256-275 should be more precise and quantify some statements such as “somewhat high mean thermal sensitivities” (line 263), “overall high thermal sensitivity and low variability” (line 267), “cluster 3 had the greatest variability through time” (line 271)

3) Better consideration of interannual variability in thermal sensitivity

The cluster analysis of thermal sensitivity relies on an annual average time series of thermal sensitivities. I suggest that the manuscript better lay out the implications of having sites with fewer years of data. Did this have a strong influence on the clustering? For example, was the clustering similar if performed using a single and most common year of data available?  Section 4.5 in the discussion does a very good job of discussing limitations in general terms but adding a more formal analysis would be more convincing.  

MINOR POINTS

line 52: Define thermal memory as it is not a widely accepted concept.

line 179: what is the dissimilarity matrix d^c xy ?

line 246: Although presented in Supplementary Material (Table S2), I suggest adding a sentence to give an idea of the variability in the number of clusters according to the method used.

line 249: Without regional knowledge, it is not clear from figures that “air and water temperature correspond closely with elevational gradients”.

line 302: thermal sensitivities varied substantially between sites? I suggest being more explicit as to what is being compared here.

line 306: non-redundant aspects relative to what? I suggest being more explicit as to what is being compared here.

line 323: This statement is a bit strong and little supported by results. For example, static thermal sensitivity (e.g. Table 2) may in fact align well with clusters defined using the time-varying approach, something the manuscript did not look into.

line 334: To what does the buffering refer to?

line 335: A comma is missing after “summer”

line 361: Do “summary metric regression” refer to Table 2?

line 435: Are there large dams in the two studied basins? If so, it should be clearly stated as this could explain why certain environmental variables had little influence.

line 457: What were the bandwidth and averaging periods used? I couldn’t find this information anywhere in the methodology.

Overall, I really like this study, from the conceptual development, to the data collection, to much of the analysis (especially continuous time series of stream thermal sensitivity), and discussion. I think there is great transferrable value of interest to HESS readership. I have some criticisms of the way the sensitivity metric data are visualized and discussed in Figs 1 and 3, but I really like the metric time series analysis is shown in Fig 4 and 5. It would be nice to show representative streamflow from those basins over the same time periods to help assess how thermal sensitivity may be driven by the volume of water in the channel at any one time (determines channel water thermal inertia to changes in net heat flux). Low stream discharge volume may be a primary driver of increased thermal sensitivity at many sites in late summer, though I do not see discharge included in any of your quantitative analysis of controlling parameters (though baseflow index is derived from stream discharge, and is included here in a general way). As mentioned by Reviewer 1, given the ‘expectations’ listed in Table 3 it would be nice to frame the study as hypothesis driven/testing, which would not be a major change to what you have now. Below I list some more major and minor points that could be considered during the revision process.

1. L15: ‘…it is critical to both understand the underlying processes causing stream warming and identify the streams most and least sensitive to environmental change.’ Measurement of air-water temperature relations across the landscape provides an efficient way to address this important topic. However, it is a localized measurement that may not reflect general behavior across the stream system as other related studies have shown, especially when there is strong variability in groundwater discharge (eg Z. Johnson et al papers). This point is discussed somewhat in the body text, but still could be made more clear throughout. Local stream channel heat exchange process can dominate the local air-water temp sensitivity metrics, which speaks to collecting spatially distributed datasets, as you nicely did for this study.
2. Although stream thermal sensitivity is quantified relative to changes in air temperature, air temperature warming may not always be the primary driver of stream temperature warming. Sensible heat fluxes are often dwarfed by solar and latent heat fluxes along the stream corridor. L39 acknowledges this important point. However, climate warming as typically described is primarily driven by the impacts on the global long wave radiation budget by accumulation of greenhouse gasses, not changes in solar short wave radiation input. The point that air temperature itself may not be the primary driver of stream temperature change at the seasonal timescale should be more clear, throughout. For example there is this statement on L122: ‘The slope of this relationship, the thermal sensitivity, indicates how sensitive a given stream’s water temperature is to changes in air temperature.’ I am not sure that is true, more that air and stream temperature are sensitive to solar radiation in more or less coupled ways. This is kind of a nuanced point, but I have interacted with several people who interpret these type of metrics as air temperature often being the primary driver of stream temperature, presumably through sensible heat exchange.
3. L41 and elsewhere: Addition of water to the stream channel impacts thermal inertia and stream temperature sensitivity, even if that water is of the same temperature as the channel. How are these patterns impacted by variable stream discharge at locations over time and along the stream network continuum? For example, clusters 2,3, and 4 show substantial increases in thermal sensitivity in late summer during presumably the lowest flows.
4. I found the ‘Identification of environmental drivers in thermal sensitivity’ section most questionable given the relatively small sample size and lack of representation across varied types of watersheds. Also, hydrologic attributes downstream in a network are inherently influenced by physical attributes upgradient in the network, and your spatial sampling spans upstream to downstream. I think that statements such as: ‘Annual patterns in thermal sensitivity are largely controlled by underlying geology and climate across two Pacific Northwest river basins’ are too definitive given the sparse nature of the datasets across a range of geologic and climatic variables. It may be that stream network position is more important that some of the apparent shifts in the tested physical variables.
5. The air-water temp sensitivity metrics in Fig 1 are somewhat difficult to interpret, as data are plotted seasonally over years for individual sites all by elevation. Given some sites appear at quite similar elevation, its not possible to disentangle changes by site and changes by elevation, and which sites are upstream/downstream of each other. I do not have any great advice with how to deal with this, however. Different colors for all sites would be overwhelming. Apparent trends in thermal sensitivity with elevation in some seasons may be somewhat of an artifact of plotting both watershed datasets together. Taken alone, seasonal datasets from either watershed would not seem to show an increasing trend with elevation. Given the inherent hydrogeological and climate differences between the two study watersheds I am not sure it is appropriate to depict and analysis the season metrics together.
6. There are numerous places in the paper where a statistical test is inferred but it is not clear if a statistical test (along with p-value) was performed. For example: L233 ‘Overall, weak and inconsistent patterns emerge in summer between thermal sensitivity and landscape and climate variables’. While ‘patterns’ does not indicate a test, ‘weak’ does. Also, L230 ‘Thermal sensitivities for sites with consistent data coverage tended to covary,..’. Covariance is a statistical test and should be associated with a significance level. My biggest problem is with the fourth column of Table 4, where linear fits are shown to the datasets without significance levels being directly indicated. I am pretty sure that many of those fits are not significant, and therefore should certainly not be shown. Plotting the best fit lines tends to influence the reader’s perception of trends, and if they are not statistically significant, they do now exist according to those significance metrics (eg p value levels). Labeling the column ‘observed relationship’ indicates all linear fits shown are significant and I see that as highly problematic.
7. As mentioned above, plotting data from the two study watersheds together to assess apparent changes in the sensitivity metrics across elevation and other physical variables may be problematic given the inherent differences in settings. Essentially all of the apparent patterns shown in Fig 1 and 3 would not exist if either watershed dataset was plotted alone.
8. I am not sure I universally agree with this statement that leads the Discussion: ‘Thermal sensitivity varies throughout the year and reflects hydrologic conditions at a given time and place within a watershed; therefore, it should not be treated as a static value.’ Just because a parameter may show variability over time, does not mean the average value is not meaningful in assessing differences between sites. Daily temperature is one example, or anything else that varies diel or seasonally. I do agree there can be great value in inspecting short term to seasonal variation in air-water temp sensitivity metrics, but that is not a requirement of all studies to be useful.
9. It is typical to not assess air-water temp relations when stream temperature falls below some threshold close to freezing, as described by Ben Letcher’s work and others. Was a cutoff value used here (eg 0.5 or 1 deg C?) It does not appear so for some of the winter datasets, which may not make sense conceptually. Stream and air temperature must decouple as the water starts to freeze, though perhaps these streams do not freeze (or come close)?
10. What do you think may drive the super low thermal sensitivities observed at some sites (eg less than 0.01?) That would seem to be possible mismatch of air and water temp data or a spring run creek totally dominated by groundwater near to the discharge source.

Minor comments

L37: This statement could use a range of supporting citations

L41: addition of water to the stream channel impacts thermal inertia and stream temperature sensitivity, even if that water is of the same temperature as the channel.

L45: ‘diagnostic’ tool may be better here than ‘predictive’ tool

L65: what do you mean here by ‘insensitive data’? Do you mean difficulty in collecting appropriate data to calibrate/validate heat budget models or something else?

L72: You could pull this thought out of parenthesis.

L75: ‘along’ river networks?

L78: It is not clear here whether you are referring specifically to statistical cluster analysis or more qualitatively to spatial groupings of streams that show similar response across the landscape

L82: mention generally where the two experimental basins are regionally

L83: it is not clear what you mean here by ‘characteristic regimes’

L85: perhaps add ‘(decreased thermal sensitivity)’ after ‘decoupling between air and water temperature’ for clarity

L107: Can you clarify the subscripts for number of loggers in each basin, and also list what specific Tidbit model(s) was used?

L111: please clarify these are water years in North America

L117: Solar shields were also used for the Tidbit loggers deployed in the water?

L141: drop ‘original’

L141: when you say ‘continuous’ metric what is the realized timestep of the output? Is it calculated by season or over entire datasets?

L162 and elsewhere in this section: It would be helpful to have topical sentences explaining plainly why these various calculations were done before diving into the nuts and bolts of how they were done.

L199: Can you better explain ‘the stability of clusters’ concept? Again, these methods subsections tend to dive right into the details of the calculations without a clear explanation up top of why the calculations were performed. The ‘why’ can be gleaned, but may not be clear for readers from varied scientific backgrounds.

L220: you may want to reminder what years you are talking about.

L230: Are you assessing covariance by eye or statistically?

The subsection 3.2 title may be better posed not as a question

Table 1. Its probably OK, but a little odd to list Baseflow Index as a geologic variable, given the importance of groundwater levels in addition to geologic materials.