The authors present a QPE data for a large number of watersheds throughout the Appalachians developed using  inverse method for correcting radar/gage QPE based on observed streamflow and a hydrologic model. I still start by disclosing that I also reviewed 2024WR038446 (Water Resources Research), which is among the as-yet unpublished studies by the same authors. As with that manuscript, I found this study to be intriguing and somewhat challenging, and ultimately have the same general concern as I expressed (not as clearly as here) in my review for that prior work.

Basically, my interpretation is that the authors have developed an approach for adjusting QPE based on back-trajectories of simulated streamflow such that the water budget closes at the event scale. This makes sense; I toyed with similar ideas myself years ago (though never did any real work on it). It does raise a potential concern, that I don’t feel the authors did a great job addressing in either study. Specifically, if you use this approach, it seems to me that the appropriate way of judging success is whether the adjusted rainfall looks “better,” i.e. closer, in terms of amount and spatiotemporal pattern, to some reference precipitation. Obviously, that’s hard to do, since we don’t have good reference precipitation in many locations. Instead, the authors show that the simulated hydrographs using the corrected precipitation have improved. This doesn’t seem convincing—of course they have improved. You’ve adjusted the rainfall specifically to make sure that the hydrographs improve; then used the improved hydrographs as evidence that you have fixed the rainfall problems. But does that mean that the rainfall is more accurate? If the hydrologic model is good (and I trust that the authors’ model is good) then the answer is “probably.” If the model is not good, the answer is “probably not.” The authors don’t really answer the question.

I would appreciate the authors’ response to that criticism. In addition, the authors need to state more clearly the differences between this study and others, particularly Liao and Barros (2024a), which I have reviewed, and Liao and Barros (2024b), which I have not. I guess this study is essentially a “scaling up” of the methods from those papers to more watersheds over a larger region? Fine, but please state it clearly.

I will add that, similar to the earlier manuscript, the writing and overall presentation quality of figures should be improved. There are a large number of minor grammatical problems, especially with run-on sentences (three in the abstract alone) and missing articles (mainly “the”) and some verb tense problems. These didn’t make it impossible to understand the paper but do distract from the study’s strengths. The figures should include legends, readable font sizes,

The statement “uncertainty from the model and model parameters is assumed to be negligible” is not really reasonable, and is inconsistent with the following sentence that states that these have “secondary importance.” “Secondary importance” is ok (at least for flood events, since as you note, forcing uncertainty will be large), “negligible” is not. Indeed, you aren’t neglecting them in your method. Instead, you transfer the calibration effort from model parameters to rainfall. That is perhaps a reasonable thing to do for flood simulations with a high-quality distributed hydrologic model,as in this study. But it is an issue that should be more clearly acknowledged in your studies. And returning to my first concern, it is problematic if you are unable to quantify whether that produces improved rainfall—if your rainfall is practically incorporating model structure and parameter error, there is the risk that it produces unrealistic rainfall outcomes. But in your study, we are left to wonder.

I realize that I might be misunderstanding the issue entirely. If so, please clarify.

Comments on “StageIV-IRC – A High-resolution Dataset of Extreme Orographic Quantitative Precipitation Estimates (QPE) Constrained to Water Budget Closure for Historical Floods in the Appalachian Mountains” by Liao and Barros, submitted to Earth System Science Data, for possible publication.

The authors developed a framework based on their previous efforts on inverse rainfall-runoff modeling, and bias correction methods. Based on the framework, they developed a bias-corrected rainfall product that consists of 200+ storm events spreading across 20 basins in the Appalachian Mountains. The new rainfall product maintains a better basin-scale water budget and shows promising results for flash flood modeling in this region. As a manuscript submitted to ESSD, the uniqueness, usefulness, and completeness of the dataset should the most important criteria. While I have no concerns about the uniqueness or completeness, the usefulness of the bias-corrected dataset is questionable. In addition, the dataset covers only a very limited region, only head watersheds in the Appalachians. Whether the proposed framework to other watersheds or storm events is another concern of mine. This is because there are many subjective choices in the framework. Justification of these routines is needed. I agree that the golden rule is to test the performance of the dataset through hydrological modeling, but this does not mean that we should “force” it to happen. The core of the proposed bias-correction framework has been developed in the authors’ previous studies. I would thus suggest the authors to pick up either one route (framework or dataset) and resubmit to a more suitable journal. I have some other concerns, which are listed below (not necessarily in the order of importance).

1. Model uncertainty. The authors emphasize that they are use a non-calibrated model in this region and model uncertainty is minimal. However, they attribute some of the poor performance in the new product due to model uncertainty (not capable of groundwater modeling). This is problematic, especially when transferring the framework to other regions with diverse land surface properties of runoff-generation mechanisms.
2. Presentation quality. I would suggest the authors to substantially improve the presentation quality if resubmitted to other journals. This includes the structure of sentences which are too complicated to be understood, clear structure of the Introduction, more details for the IRC framework (right now it is only described in the figure), and less details in the equation and metrics of model evaluation. In addition, the manuscript lacks a map that clearly show the study region, including the watersheds, rain gauges, and different regions. This makes readers outside of US a miserable experience.
3. Introduction. The Introduction session should be reorganized in a more concise and logical way. The currently broad theme matter distracts from a focus on flash floods in mountainous areas. The second paragraph offers irrelevant context to the overall topic. It is vital to emphasize the significance of high-resolution precipitation estimates specifically in these regions. Instead of laboring extensively over the methods and details of this study, try to present a comprehensive overview of possible solutions to the challenges in QPE. The importance of Appalachian region should be highlighted as well. Why do the authors believe the new product should be able to contribute to earth system science?
4. More specific comments. I will not elaborate them all. They can wait till later rounds of reviews if applicable.

• Line 534-543: This paragraph is not appropriate in the Results session.
• Line 603-605: A lack of understanding in Karst-terrain physics does not justify the omission of model parameter calibration.
• Line 632-644: Discuss whether all flood-generating storms in mountainous regions align with terrain gradients and whether the greater consistency between precipitation spatial pattern and terrain gradients indicates better bias correction.
• Figure 5-7: Explain why results for specific seasons are shown.
• Figure 11: Show the terrain gradients, which makes it comparable to the spatial pattern of precipitation.