This paper presents an interesting work on the diagnose of NWM model with the aid of isotope data. The isotope simulation identifies different bias characteristics in high and low elevation regions. The bias of isotope simulation in low regions is attributed to the contributions of irrigation return flows, and some statistical analysis and literature reviews are conducted to support this hypothesis. Overall, the logic is clear, the analysis is solid, and the written is good, making this paper worth publishing in HESS. However, I would like to point out some major concerns related to the isotope dataset and mass balance calculation.

1. The calculation of surface water isotope: From a hydrological viewpoint, the calculation surface water isotope ratio (equation 1) is confusing. The authors determine the isotope ratio through dividing the summed isotope fluxes by the summed runoff and groundwater fluxes. However, the term “runoff” usually refers to the sum of surface runoff and subsurface runoff. I don’t know whether the “runoff” provided by NWM model refers to surface runoff or the total runoff. According to the equation, it seems that the runoff is actually only surface runoff. If this is the case, I suggest the authors to make it clear in the main text. Otherwise, the isotope ratio of surface water should be [Rgw*Fgw+Rp*(Fro-Fgw)]/Fro.

2. Choice of isotope dataset:

The authors adopted the monthly long term average precipitation and groundwater isotope data as the input data. This is okay for groundwater because its isotope composition is rather stable. However, the precipitation isotope usually has very strong temporal variation, especially during wet season. Given that a high-resolution dataset of hydrological fluxes produced by NWM was adopted, it might be better to use a high-resolution precipitation isotope dataset, such as the output of isotope enabled general circulation models (iGCM, such as https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2008JD010074). The reliability of the precipitation and groundwater isotope dataset itself also need to be evaluated, at least citing some descriptions about their accuracies in published papers. Otherwise, it would be hard to determine whether the simulation biases come from the NWM data or the isotope input data.

Besides, I find that the authors evaluated the isotope simulation performance by comparing the long term simulated results with the measurement surface water isotope sampled at a specific time. Such kind of measurement data would be highly dependent on the specific precipitation events and the corresponding isotope composition before sampling. So it might be more reasonable to compare the measurement data with the corresponding simulated result at the sampling time. We can observe a smaller range of simulated isotope compared to measurement in Figure 3. I think the aggregation of precipitation input and simulated isotope could be the reason of this.

Nonetheless, I understand that replacing the input data and repeating the whole calculation process is challenging. If this is difficult to achieve, please consider addressing these issues in a discussion section.

Specific issues:

• L142: Provide the full name of HUC2 at the place it appears for the first time.
• L208~L215: How were the 10 random draws generated, and how were they used in specify to evaluate uncertainties? There seems to be results related to uncertainty in the result section.
• L400: Figure S8 should be S10?
• L442-443: This is a strong statement and is the basic of following analysis. Consider providing more explanation on it.
• L449: There is an additional letter “d”
• Please provide the r2 and p values in the scatter figures such as Figure 8 and 10
• Table 3: There is an additional symbol “+” in the last row

Putman et al. use long term mean summer hydrology, gridded precipitation and groundwater isotope ratios, and in-stream water isotope ratio observations to evaluate the accuracy of the National Water Model, which is known to perform poorly at low elevations and in highly managed basins in the western United States. The authors use a water isotope mass balance approach to estimate river reach isotope ratios using National Water Model derived fluxes and compare ‘modeled’ isotope ratios with over 4,000 in-stream isotope observations. The differences between observed and modeled ẟ¹⁸O and d-excess are used to evaluate statistical patterns in evapoconcentration of observations relative to modeled isotope values. Putman et al. conclude that offset between modeled and observed water isotope values are diagnostic of the lack of agricultural irrigation practices represented in the National Water Model and test this using 2015 Water Use Census data (annual water use by county) and developing an approach for estimating the amount of streamflow sourced from agricultural irrigation on a coarse catchment scale.

Overall, this manuscript is well written and provides a robust framework for evaluating model accuracy, considering different temporal and spatial patterns in the residuals, representing uncertainty in previously published datasets, and the utility of stable water isotope ratios in diagnosing misrepresentation of physical processes in hydrological models. This study is a solid contribution and worthy of publication in HESS; however, there are a few points I think need some clarification before the manuscript is finalized. Below is a discussion of my major comments, followed by more minor, line specific comments. Thank you for the opportunity to engage with this interesting and useful study.

Major Comments

1. I’d like some additional discussion or clarity on why the interpretive framework in Figure 2 is applicable to the results of the d-excess(diff) and δ¹⁸O(diff) calculations presented. Typically, d-excess is calculated from the isotope ratios of a single/discrete water sample, but here d(diff) is calculated from isotope values that are a mix of spatially and temporally averaged modeled isotope ratios and point observations of in-stream isotope values. The framework in Figure 2 is based on dual isotope fractionation processes as units of water moves through the hydrological system. Given that the d-excess(diff) calculation here is based on spatially and temporally averaged modeled values, then d-excess(diff) and δ¹⁸O(diff), used as indices of evapoconcentration, are likely masking or missing a lot of variability in climate conditions over time and space. Some additional reasoning would be useful for the reader. The discussion section describes how interannual variability is considered by looking at the regressions between the modeled values and the entire observational dataset versus the averaged observations, but I didn’t clearly understand the lines being drawn between that interannual variability (due to variable climates?) and the consideration of how much variance is missing in the NWM that can be attributed to agricultural return flows.

2. The diagnosis of the National Water Model inaccurately representing agricultural return flows is well reasoned in the study and a conclusion that makes sense given the difficulty of many hydrological models in representing irrigation practices given that water use data is difficult to obtain (water users are often reluctant to share this information freely in highly managed areas). One concern I have is the practicality of reasonably incorporating agricultural return fluxes into the model and the approach for estimating this contribution taken in the manuscript. The simplified way of calculating ratios of water use contributions to stream flow in this study seems like a reasonable first pass, but there are many unknowns. For one thing, water use can vary widely between water year types – so including some level of uncertainty or variability in the 2015 Water Use Census data would be helpful. For example, 2015 was a critically dry year in California, so water use data is likely reduced during that year compared to a wetter year on record and the ratio of groundwater to surface water use in the Central Valley is likely inflated for that year. Applying water use data that is from a specific year as a point of understanding contributions to streamflow should likely consider the isotope data from that specific year to match, since it’s not representative of long-term mean conditions. I’d like to see some explicit discussion of what the water use data represents in the main text and whether it's representative of long-term conditions. One thing that could be considered (it may or may not be appropriate here) is the EPA’s EnviroAtlas’ dataset of different types of water use. They have estimated longer term datasets for agricultural water use, industrial, domestic, etc. https://www.epa.gov/enviroatlas

Line Specific Comments

Line 89-92: d-excess is typically calculated for corresponding δ²H and δ¹⁸O values for a specific sample/observation. In this study, d(diff) is calculated from estimated average isotope values. Is d-excess still a reliable metric for evaporation when you are calculating from long term averages/values calculated using mass balance? It seems like the mass balance calculation step would not include all the non-equilibrium fractionation processes that could impact the d-excess value. Some additional reasoning somewhere in the text would be helpful for the reader.

Line 108-110: remove “associated with irrigation”

Figure 1 Caption: in text citation format typo for (Bowen 2022b)

Line 191: “Where available, we filled these data gaps using method outlined in Text S2.” I would briefly explain that the authors used the gridded DJF precipitation isotope products to help fill the gaps, since they are listed in Figure 1, but not mentioned anywhere in the main text.

Line 220: typo, should be “This decision was made…”

Line 245: “We evaluated the results with all unaveraged observations and mean isotope ratio at river reaches with multiple observations.” I’m not clear on what this means. The correlation/regression analyses would need to be done between monthly average isotope ratios for an apples-to-apples comparison, rather than mixing discrete observations with monthly average model values.

Line 250: “( Text S3)” has an extra space after first parenthesis.

Figures 6: This is a really nice figure illustrating the different temporal evolution of δ¹⁸O(diff) and d(diff) throughout the different major basins in the western US!  Please list the distribution statistics in the caption (i.e., box represents what percentiles, what are the smaller, shaded boxes in the Great Basin boxes, diamonds are outliers?).

Line 454: typo, remove “a” after due to

Line 458: p<0.001 is listed for significance, whereas it has been p<0.1 or p<0.01 in other parts of the manuscript. I suggest staying consistent with listing the p-value in the text.

Figure 8: y-axis label I believe should be listed as ‰ instead of %.

Line 468: The meaning of the first sentence is unclear. The δ¹⁸O(diff) and d(diff) are statistically significant relative to what?

Figure 9: Please explain the statistics of what the boxplots represent in the caption. Also, is not all data shown? Every land cover type looks like it has groundwater d less than -20 but that’s where the plot stops.

Line 556: typo, should be “stream”

This is a well-written, comprehensive study that applies a novel evaluation framework using isotopes to diagnose potential error sources in hydrologic models. The study offers useful analyses and conclusions for hydrologic model evaluation, with a focus on processes, which are supported by multiple lines of evidence. I believe the study merits publication, but I have several main comments that I offer for the authors to consider:

• Isotope tracer expertise is assumed. I suspect that many HESS readers are not experts in isotope tracers, and there is a lot of assumed expertise and jargon. The manuscript would benefit from more explanations up front and throughout so that readers unfamiliar can benefit from this novel approach. I point out examples in my specific comments.
• The main conceptual method is hard to pick out from all the details. The authors do an excellent job of explaining their methods in great detail, but I found myself missing the forest for the trees in my first read through. Having never read a tracer study, the main approach – comparing the observed isotopes with the NWM derived estimates (which comes from both NWM fluxes AND gridded datasets) was hard to follow. I think this should be clarified earlier, and I offer a few suggestions to improve the Method section organization in my minor comments. For example, Figure 1 is very detailed, but hard to follow all that’s going on at the outset of the Methods, so I suggest adding a general conceptual overview and/or simple flow chart to guide the reader (including what’s in Equation 1, otherwise it appears much too late for the reader to follow what’s happening). I recognize that this is a subjective suggestion, but I think it would help to increase the reach of your paper.
• The key results are hard to pick out from the supporting results. I appreciate the comprehensive results and multiple lines of evidence presented, but will caution the authors that it can make it difficult for readers to focus on the key results on which the main conclusions are based. The authors may want to review all the results and see if there are any that they might like to include in the Supplemental (I make a few suggestions). I recognize that this is a subjective suggestion, and that there is a tradeoff to including too few versus too many results, but I think a slightly more curated results section would help to increase the reach of your paper. (Also, I will note that the authors do a nice job of summarizing the key results in the Abstract and Conclusions, so this is just a suggestion for the Results themselves).

Specific comments:

Abstract:

Line 3: Is “fidelity” the right work here?

Line 4: In parenthesis, I am not familiar with the delta 18 Oxygen and delta 2 hydrogen notation.  Is there a way to define them here for non-experts? Maybe just remove the parenthetical all together from this sentence?  You may need to define them later (line 8, etc). 

Line 4-7: I had a hard time understanding generally what you did from these sentences until I read the manuscript and thoroughly studied Figure 1. Here’s a suggestion of what might help a reader like me (at least in terms of laying out the general conceptual framework of what was done – please fix details if I have them wrong):

“In this study, we compare observational river isotope data with estimates of river isotopes derived from the NWM. The evaluation is done in 5 basins in the western US in summer from 2000 and 2019. In terms of observations, we use 4503 in-stream water isotope observations in 877 reaches. In terms of the corresponding estimates of river isotopes, these are calculate using a mass balance equation based on NWM-fluxes and estimates of isotope ratios from long term mean gridded precipitation and groundwater datasets.” 

Introduction

Line 18-26. This paragraph is quite general in scope and I don’t think is needed. I think you could delete and start with paragraph 2 (i.e., line 27), using something like the first line of your abstract, i.e., : “Hydrologic models, such as the National Oceanic and Atmospheric Administration’s National Water Model (NWM), provides critical analyses and predictions of streamflow that support water management decisions. The NWM is an application of the WRF-Hydro model (Gochis et al., 2018), and is fully routed with high spatial and temporal resolution, providing short and medium term streamflow … “ etc.

Line 27: The operational NWM is based on the WRF-Hydro model (not its data); the NWM is an application of the WRF-Hydro model. Perhaps you are confusing WRF-Hydro (a hydro model) with WRF (a meteorological model). Suggest saying, “…like the National Oceanic and Atmospheric Administration’s National Water Model (NWM) which is an application of the WRF-Hydro model (Gochis et al., 2018)” or you could say “is based on the WRF-Hydro model (Gochis et al., 2018).”

Line 34: “fidelity” does not seem like the right word here.

Line 84: Sentence that starts with “tracers” – can you add a general sentence on tracers and isotopes (before getting into the 16O, etc), for the non-expert? When you get to the parenthetical 16 oxygen, etc, please define these. I start catching your drift a bit later when you define 2H/18O as “heavy” and 1H and 16O as “light”, though as a non-expert I’m not sure to what it is compared.

Line 89: The expression: “The secondary parameter, deuterium excess..” is not clear to a non-expert.

This paragraph (lines 84-92) is very dense, especially for a non-expert (of isotopes), try rereading as someone who doesn’t know about isotopes and generalizing a bit as possible, to help the reader understand this powerful evaluation tool.

Line 93-94. What is isotopic fractionation? Can you say this another way for the non-expert?

Line 110: What is evopoconcentrated and evaporative enrichment?

Line 115-116: Same as in the Abstract, I wasn’t really sure what you did until I read through and studied Figure 1, although this is easier for me to understand than the Abstract. Could you start most generally, saying: “In this study, we compared stream water isotope observations with estimates of water isotopes derived from an isotope mass balance. The isotope mass balance is from xyz NWM and lmnop gridded long term, etc”. 

Methods

127-128. Same as previous comments, I initially had trouble understanding what was done here. Even though I like Figure 1, it is very dense. I’m wondering if you could have a very simple conceptual figure first to ground the reader before Figure 1, where you just show the 3 main pieces: (1) Direct obs, (2a) NWM, (2b) gridded ratios, as well as equation 1 and where the pieces fit in on the model side. This could be in a section called “2.1 Conceptual Framework”, and could include the general figure, which would introduce sections 2.4 and 2.5, and it would absorb the current specs listed in “2.1 Temporal domain”, “2.2 Spatial…” and “2.3 Data assim”. Then Figure 1, with all the details, could come later.

Figure 1: This is a very nice figure, but dense… See previous comments. One note: For NWM data feeding into Equation 1: maybe have the notation (Fgw) and (Fro) and for obs data have (Rgw) and (Rro) there to link with Equation 1.  This might be better suited to a more general conceptual figure though (see previous comment).

2.3. Data assimilation: This seems minor to be a full section, and I wasn’t sure what this was related to. Can this be part of 2.2 Spatial domain or just Supplemental? Or if you decide to have a Conceptual Framework section, it could be absorbed in that.

Line 160. Correction, the NWM is based on the WRF-Hydro model – which is an open source, community hydrologic model, it is not based on inputs from it (I think you might be confusing WRF-Hydro with WRF, where WRF would provide inputs): “The operational hydrologic model is based on the open-source, community hydrologic model, WRF-Hydro (Gochis et al., 2020b, a)…”

Line 170: Do you mean Figure 1 here?

Equation 1: Seeing this equation helped me to see how the pieces fit together. If you decide to include a conceptual model, I suggest having this equation in it to see how each piece fits in (you could do it generally, for just one reach, as a demonstration, so it was a simpler equation without the subscripts).

Figure 2 – I like this figure and how it showed the way to interpret. I often had to look back at this figure to interpret later results.

Line 371: What is a meteoric water line? What is a surface water line?

Line 380: Is Table 1 needed for the main text or could it go into the supplement?

Line 386. What is an isotopologue?

Line 405: “The strongest signal in our data is that of evaporation, evidenced by combinations of positive δ18Odiff and negative ddiff in arid regions.” <- This is an important conclusion, but there are so many results it’s hard to quickly see what evidence this is from – I had to really go back and study all the figures and tables to realize I needed to imagine all the points from Figure 4 as if they were on Figure 2. Can you add something to that effect to guide the reader? Or add the colored quadrants to remind the reader? Or maybe just say in the caption of Figure 4, “see Figure 2 for what the different locations on the x- y- axis mean”?

Table 2: Is this need for the main text or could it go into the supplement?

Section 3.3. This section and Figure 5 did not seem particularly important to the results/conclusions, and one suggestion would be to put it in the Supplemental (so that the other key results are less buried). If so, you could just have one sentence at the end of the previous section saying something like “ There was little interannual variability, which we interpret to mean there was pervasive presence of eval… etc.. see Supplemental xxx”.