General Comments

The authors tackle an important topic, namely how to best combine gauge and satellite precipitation estimates for applications.  The hydrological validation they pursue is a reasonable way of testing how the various input datasets perform, and this is clearly the strong point of the manuscript.  As such, the existing examples and conclusions related to the hydrology seem solid.

The other big point of the manuscript, as the title makes clear, is the combination scheme that employs CA interpolation, but here the manuscript falls short.  I would expect a step-by-step demonstration that all the extra mathematical complication produces precipitation fields that are physically meaningful and more consistent with the input fields than some simpler scheme.  I would consider it mandatory to address this issue.

The English is not ready for publication; I have commented on a few word choices that I found confusing or that might escape a general technical editor, but not otherwise.  Other issues needing attention are listed below.

Specific Comments

1. CA and regularly gridded data: It is not clear to me how CA handles the regularly gridded data.  Does it assign each satellite gridbox value to the finer gridbox closest to that satellite gridbox’s center?  Conventionally, gridded data give you an average value across the entire box, so if you just assign that value to the box’s center, doing an interpolation and then averaging that field back to the original resolution will not give you the right average, in general.  Specifically, locally convex-up areas will be underestimated (particularly sharp peaks), concave-up areas will be overestimated, and the boundary of the precipitating region will spread somewhat into the non-raining area.  I consider discussion of these issues to be mandatory.
2. What is the model grid spacing?: Section 4.3 seems closest to stating this, but I missed seeing a declarative statement giving the specific value of the grid spacing.  If you’re really down at “hundreds of meters”, it needs to be made clear that the effective resolution of the precipitation data is back at 5, 10, or more km.  Finer than that is just more and more precisely defining smooth variations between the available value locations.
3. Example interpolated precipitation fields: Given the emphasis on the innovative combination of data sources, I would expect to see a sequence of maps illustrating the process and improvement versus simpler “traditional” schemes.  This should happen first, before pointing to the accumulated precipitation in Fig. 5 and the aggregate hydrological results in Fig. 6 (for example).  This aspect of the manuscript is where the issues in item 1 need to be addressed.
4. Boundary bias: This phrase comes up several times, but it was never quite defined as to what boundary was being discussed.  It is stated that data outside the basin is used, which presumably should solve a problem at the basin boundary.  Perhaps the problem is along the southwest side of the basin, where no additional gauges are shown in Fig. 1.  The statement in L.417-418 should appear a lot sooner in the text and be more explicit about how this works.
5. Fig. 1: I question whether all those gauges outside the basin are really useful and therefore worth depicting.  I would suggest that as you move outside the boundary you can stop after you pass about 3 gauges (which of course varies with coverage).  Was it really not possible to obtain gauge data to the southwest of the basin?  This introduces the boundary effects the manuscript discusses (right?).
6. Fig. 7 caption: The statements after the first sentence are interpretation and belong in the text.

Technical Corrections

1. L.180, 720: “O” is actually that author’s last name; “Sungmin” is her first name.
2. Nine occurrences: “IMERG” is mis-stated as “IMERGE”.
3. L.52: Not sure “captative” is the right word.
4. L.88-89: Awkward phrasing, including that “peculiarity” is probably something like “availability”.
5. L.168 and 3 other locations: I’m not sure what “rain bandwidth” means.
6. L.186: Should this be “…that are usually not instrumented.”?
7. L.193: Fig. 2 refers to a “workflow”, not “rationale”, which seems better.
8. L.193: Fig. 2 says three tasks, not four.
9. L.202: Unclear phrasing; are you saying something like “the model is not calibrated specifically for this study’s cases”?
10. L.428: I think “intuitive” could better be “subjective”.
11. Fig. 1: a) What are the blue lines? b) The hydrometers are nearly invisible; maybe they should be plotted in white?
12. Fig. 2 caption: The phrasing is awkward, perhaps something like “… workflow, consisting of three main tasks:”
13. Fig. 3: The mostly-dark colors make it really hard to distinguish the basin and gauge coverages. I’d say the Google Earth background and the basin’s blue need to be much lighter.

The paper of Lombardi et al. deals with an interesting topic concerning the reliability and usability of satellite rainfall products for hydrological applications. To this aim, they consider a case study in northern Italy (Tanaro watershed) and test different precipitation fields achieved using only rain gauges, satellites, or several merging options. Precipitation fields are used as inputs for a hydrological model. According to the behaviour of the resulting hydrograph, the precipitation fields are judged in terms of reliability, supporting the discussion with several (very many, indeed) skill scores.

Despite the interest in the topic, I believe several changes in the paper are needed before it can be considered ready for publication in HESS.

First, I have concerns about the methodology followed. As a first step, since the main point is the reliability of the satellite rainfall products in providing quantitative precipitation estimates, I would have set up a validation exercise with observed (rain gauge) precipitation using, e.g., a leave-one-out or cross-validation approach.

Then, an indirect validation through simulated hydrographs should be justified more strongly, mainly if performed with a not-calibrated hydrological model (L202). This approach could be tricky and misleading because of the inherent limitations of the non-calibrated hydrological model so that errors can counterbalance and smooth each other. I suggest a preliminary calibration of the hydrological model with a reference precipitation field (e.g., only rain gauges) and only afterwards assess the changes caused to the modelled hydrograph with other methods.

Another concern is the description of the CA-based technique for interpolating and merging precipitation, which is unclear to me. An example should be given. Among other things, it is unclear why the authors also consider the time evolution of the precipitation field.

Last but not least, the paper could be structured much better. The different options for achieving the precipitation fields are not clearly presented (e.g., some acronyms are provided in Fig. 2, then they are explained much later -L400- and not in the Methods, but in the Results section). It is unclear why the authors decided to rely on 17 scores. This choice is somewhat confusing, in my opinion. Furthermore, many of the methods are presented in the results section or even in the conclusions (please refer to specific comments below). The discussion should refer to similar analyses performed by other authors to contextualise the results better. The conclusions section should be more than a summary of the paper.  

In summary, I saw the possible added value that this paper can bring to the scientific community. Still, a thorough review is needed regarding the methodological approach and the structuring of the article. Please find below some other minor to moderate comments. I hope my review helps improve the quality of the paper.

LL46-49: please revise. Dealing with predicted rainfall, it's impossible to remove uncertainty (correctly, in fact, the second sentence of the paragraph refers only to observed rainfall data). I suggest focusing on why accurate spatial distribution of rainfall observation is important.

L104: "The work": do the authors refer to Shi et al. (2020)?

LL111-112: I can't entirely agree with this statement. Indeed, there are a lot of studies dealing with this topic.

LL118-119 is a repetition of the second main objective declared at LL115-116. If the authors agree with my comment, please consider if the previous sentence (LL117-118) is well-placed and contextualised.

L120: are all these 352 stations really useful? I guess the authors only need those lying into or close to the analysed watershed. From this point of view, it's unclear why the authors consider a much broader spatial domain than the investigated watershed (which, moreover, is not at the centre of the domain itself). I guess many stations, for example, lying in the north and northwest, are useless for this case study.

L146: 1700 m3/s is a peak flow, average daily flow or what else? Some lines below the authors refer to a peak of 4350 m3/s (if "Autorità di Bacino del Fiume Po" is a reference, please add the year; if not, please explain/translate it in English).

L 165: Eq. (1) is quite ambiguous. It is well known that the reference area is hardly a radius, especially in orographically complex regions such as the Alps.

L176: Earth

Figure 2 and related caption: please revise. There are several errors: e.g. Guage uncal, "each case studies" [study], "eight […] setting" [settings]. Furthermore, the terms uncal, cal, uncal1, uncal5, etc. are explained much later (L400). The explanation of the different inputs for the eight simulations should be highlighted much better (maybe with a devoted Table?).

L325: finds

LL360-364: I guess this is methods, not results

Section 5.1 also is not results, but data and methods

L406: it's 5.2

LL418-419: however, while some of the stations considered in the study are located much further north of the watershed (as I claimed before, I believe they are useless to this study), after the French border, there are no stations surrounding it. This drawback should be discussed.

LL519-528: that's methodology.