The study from Canham et al. explores the time-varying hydrological controls on 5042 rainfall-runoff events from 9 western US watersheds, with aiming to untangle the influence of wildfire on streamflow. The paper is well-crafted, with the supporting data and text well recorded in the supplementary file. However, my main concern is that I think the focus of this paper should be on exploring the influence of wildfire on streamflow. Given that there are already lots of studies focus on rainfall-runoff event separation method or large sample events temporal-spatial controls investigation. Thus, the novelty of this paper should be exploring the wildfire influence on streamflow. Yet, the current paper structure contains large proportion of text describing event separation and also the controls for undisturbed events. So I think the structure of the paper should be adjusted to highlight your contributions on untangling the wildfire impact on streamflow. My detailed comments can be found below. By the way, I was accidently uploaded my review comments as community comment before. Just ignore the community comment.  

1. Table1: It would be better to add hydrologic characteristics in this table for these catchments, i.e., mean annual precipitation, mean annual potential evapotranspiration, mean annual streamflow and also maybe the streamflow regimes that you mentioned in the line 120-123.
2. Line 132: Can you explain what PRISM means?
3. Line 240-242: why it needs to use two different statistical tests to evaluate the effect of WYT and season/antecedent precipitation respectively? In Figure 7, you compared their results in one figure, yet I’m not sure whether the results of these two methods are comparable or not?
4. Table 2: Does symbol # represent the number of events? If so, please clarify.
5. Line 289: How you selected these two contrasting watersheds? The explanation of why you selected these two watersheds as example is needed. Is that possible to compare the results between Arroyo Seco and Valley Creek (this one has similar characteristics with Clear Creek)? Or Maybe Arroyo Seco and Shitike Creek (this one has similar contributing area with Arroyo Seco)? Will the results you observed from Arroyo Seco and Clear Creek also apply to Arroyo Seco and Valley Creek?
6. Figure 6: Can you explain what negative values on the x-axis for volume, peak flow and response time mean?
7. Line 329: How do you calculate this relative significance rates?
8. Line 350: Can you re-phrase this sentence? It is a bit confused by ‘for in no metric groups’.
9. The results section contains an large proportion of analysis on undisturbed rainfall-runoff events, while the analysis of wildfire impacts on streamflow is not sufficiently thorough. Only examples from two watersheds were presented. The focus of the paper should be on wildfire disturbed streamflow. Adjustment of results proportions and focus of analysis is needed.
10. Discussions with more recent large sample rainfall-runoff events controls analysis should be added, i.e. Jahanshahi and Booij (2024) https://doi.org/10.1080/02626667.2024.2302420 ,Zheng et al. (2023) https://doi.org/10.1029/2022WR033226.
11. In the discussion section, it should also have a separate subtitle and section focus more on the impact of wildfire to streamflow.

The writing and methods are generally ok, and this manuscript seems prima facie like it might be appropriate for publications assuming the below comments are sufficiently addressed. With that said, there are a number of issues that are making the methods used difficult to understand and by extension, difficult for readers/reviewers to gauge the appropriateness of the methods used. Most notably, there is a significant amount of subjective interpretation that could easily be replaced with statistical methods.

General comments:

Isn’t a watershed an imprecise concept? You can look at an individual subdrainage which is part of a drainage which is part of a watershed which might be part of a larger watershed complex. However, the “Study Area” section does not acknowledge this when describing how the watersheds were selected. It would be helpful to address this uncertainty and briefly justify why a certain selection approach was used.

Some of the data processing steps are unclear. Particularly regarding what conversions/algorithms were applied to the precipitation data and why they were used.

The fourth step of the RREDI algorithm is concerning. Why couldn’t these events be removed in an earlier step?

The methods make heavy use of visual interpretation to validate and set thresholds. A more quantitative/deterministic approach is highly recommended. Example: “Total annual streamflow and precipitation were plotted for the undisturbed period of record to visually identify the annual precipitation threshold above which streamflow increased linearly with precipitation.” Additionally, “Winter, melt, and summer hydrologic seasons were identified for each watershed based on inspection of the average annual hydrograph”. There are methods available for specifically handling these kind of problems. See change point analysis.

The problem of multiple comparisons should be addressed here. There are a large number of statistical tests being performed and there is a risk that some of the detected relationships are spurious.   

There is also a need to make sure that “significant” relationships are of “practical significance”. You seem to be working with large datasets that may easily yield detectable differences/relationships, but may only have negligible effect sizes.

It seems that each of the time-varying hydrologic controls (TVHC) effect on effect metrics were done individually (Table 3). I am curious how the results would look if a statistical model were created that included all of the TVHCs in it as covariates and used that to calculate effect sizes on the effect metrics.

Introduction

Line 33: I suggest using a reference from an academic journal.

Line 33: “… in a changing climate the occurrence and severity of wildfire is increasing…” is this generally true? It seems like an oversimplification. Are there regions where wildfire frequency is decreasing. See “Global trends in wildfire and its impacts: perceptions versus realities in a changing world” by Doerr and Santin 2016.

Line 33: It would be useful to include in the introduction a reference about how climate change might cause some watersheds to transition from being snow-melt fed to predominately precipitation.

Study Watersheds

Line 106: Unclear why this reference is here. Is this a dataset? A recommendation to use 15-minutes time series?

Line 110: Can you include a table that details how these filtering criterion reduced the number of watersheds at each step? Started with 10k watersheds -> 5000 when filter criteria 1 applied -> 200 when filter criteria 2 was applied -> criteria 3 -> …  -> criteria n.

Line 131: A more representative quantity (instead of the precipitation at the centroid) might be the sum/mean precipitation over the boundaries of the watershed [addressing this is of extremely high importance in my opinion].

Line 135: Unclear how linear interpolation is being used here. You have hourly data and are filling in at 15 minute time steps? The interpolation algorithm should be described in crystal clear detail here.

Line 138: What is the significance/relevance of 2011? And what quantity is being compared? Is this analysis justifying an assumption made from the data? It might be good to include this as supplementary material if so.

Methods

Line 153: Sorted how?

Line 154: Unnecessary to bring up the use of LOWESS in this subsection.

Line 157: The last sentence seems like it should be its own subsection and the selection of these two watersheds seems arbitrary.

Line 175: This seems sensible, except the time scale of precipitation is unclear (see comment for Line 135)

Line 182: “were also normalized” The specific methods used here are vague and are not reproducible.

Line 184: “diurnal cycling” this is never defined or discussed in the introduction.

 Line 195: I don’t think “systematic assessment” is an appropriate description of what is happening here.  You are seemingly visually inspecting the time series for three kinds of water years (wettest/driest/typical).

Line 199: “assessed with respect to the four event identification issues described above” (1) I eventually figured out what was meant by “four event identification issues” but (2) how they are “assessed” is vague. You could remove this and just keep the next sentence to avoid any confusion.

Line 241: Effect of WYT [on?].

Line 244: Good opportunity to remind readers of how many event metrics there are and to provide an in-text citation of a list of these metrics (or explicitly list them out if there are not too many).

Line 250: See multiple comparisons problem. The relevance of a “significance rate” is not clear either.

Line 260-265: inscrutable

Results

Line 267: Let the numbers and the readers decide if that’s true. Delete.

Line 270: Getting lost as to what “events” are storms? Runoff? Similarly, I am confused why accuracy has anything to do with this. I was under the impression the number of events were already known and were used to extract parameters from the precipitation-stream flow time series.

Line 284: From my understanding of the methods, this is not a reproducible threshold identification since it is based on visual inspection. Another analyst may look at the data and conclude that the break should be different.

Line 295: Clear how? Was a statistical test performed? Line 298 suggests that only descriptive statistics were used to reach this conclusion.

Discussion

Line 398: Again, not clear how classification rates applies to this analysis as this seems to be a supervised learning problem (I think you knew whether a time period was a rain-runoff event). If that isn’t the case, I am concerned that there is some circular logic going on here with the accuracy statistics (a rain-runoff event is defined as one that is identified by the RREDI algorithm, which will by definition produce an optimistic estimate of accuracy). At any rate, to provide a reliable estimate of accuracy it is important to not “double dip” and to give the algorithm “new data” to assess its performance. Common methods for this would be the hold-out validation, k-folds cross validation, monte carlo cross validation, bootstrap, etc.

Line 403: I really think using a spatial aggregate is appropriate here because of the limitations you identify here. Please do this in the next analysis.