Steyaert et al. present an approach to deriving reservoir operations for global scale hydrologic models. The study is of value to the water resources modeling community, but requires major revisions before being accepted for publication in HESS. Some elements of the method are not well justified, including the categorization of dams into “irrigation-like” and “hydro-like”, as well as use of a release decision approach based on downstream demand aggregated across an arbitrary command area. Use of a random forest model to extrapolate curves is a nice idea but is not evaluated fully (i.e., using a cross-validation scheme) and appears quite ineffective based on the results shown. Although the level and depth of analysis conducted is impressive, the quality of results/figures is quite poor, and often confusing. The study can be simplified and reworked to deliver more clear and compelling results (with more impactful figures) on improvements offered by a data-derived storage scheme. The paper would also benefit from a significant reduction in number of words. The introduction is 13 paragraphs long and contains a lot of general detail. I encourage the authors to rewrite the introduction in a way that brings immediate focus to the problem area, most recent literature addressing that problem, and aims of the study. Three or four paragraphs will suffice. The abstract, currently almost 400 words, can be halved without loss of essential information.

Title: Awkward repetition of "reservoir operations". Did you mean "A data derived workflow for simulating reservoir operations in a global hydrologic model" ? Also, this wording suggests that it is the *workflow* that is data derived, rather than the reservoir operation. So, did you actually mean something like "Data derived reservoir operations in a global hydrologic model" ?

Abstract L2. "most of the data was not openly accessible" . I would suggest that this remains true. Specify the type of data.

L27. water supply reservoirs, flood control reservoirs, and hydropower dams are found in all climate types.

L187. Do you mean: “…to determine reservoir rule curves that specify seasonal flood and conservation pools…” ?

L205. Not clear what is meant by “yearly maps of static reservoir characteristics”. Also, since L180 I have been reading and wondering the motivation and reasoning behind these two categories (“hydropower-like” and “irrigation-like”). Please try to clarify the role of this categorization early in the study.

L250. Please add further detail here on whether any efforts were made to ensure reservoirs were placed on correct streams. From what I read, it seems the lat/lon of the reservoirs are snapped to the PCR-GLOBWB grid then assigned that grid cell.

L270. Ok—here I am now realizing that irrigation-like and hydropower-like categories are used to inform releases, with the starfit approach solely defining storage curves. Doesn’t this mean the operations are not full data-driven but rather half data driven (storage curves) and half “generic” (release policy based on command area demand and reservoir purpose)?

L313 – missing reference to equation 5.

L325-330. I would be very unsure about labels of water supply / irrigation vs hydro etc within GranD leading to a neat splitting of dams respectively operated for downstream demand versus maintaining high storage levels. Apart from the issue of inaccurate reservoir purposes in the available global datasets, one rarely finds such simple distinctions in reality. Are you able to show that two categories of operations actually exist, e.g., by comparing the starfit curves for irrigation-like versus hydropower-like dams in the set of 1752 observed dams? I would be surprised if you find a clear distinction. If this is the case, I don’t see strong justification for the splittling—which in a way complicates the study.

L335. It’s unclear to me what the command area offers. The storage curves can guide the release without a downstream demand. Were any tests performed to evaluate whether this downstream demand actually improves on accuracy?

L342. How are surface water abstractions considered? Is this based on demand within the same grid cell as the reservoir?

Equation 6. Maybe I missed this, but how is Env defined? Also, how is the flood release defined? Is this just spill required to draw the reservoir back to the active zone?

L350. Unclear what is being done here. Are you creating an active zone per dam type and country? Why? I thought the random forest provides full parameterization for each dam.

L381. After validating the model and demonstrating effectiveness with the 25% out validation, why not re-train with all 1,752 structures before extrapolating? Also, given the importance of the random forest to the overall framework, I strongly suggest the authors pursue a k-fold cross validation scheme rather than single training and test samples.

L385. How many reservoirs end up being constrained to these bounds? Also, it’s not clear what is meant by flood peak here. Do you mean upper bound of active storage?

Table 2. Here would be very interesting to see a version that drops the command area and demand parameters (as well as hydro/irrigation split) entirely. I can’t see a strong justification for the demand-based release or the command area (or the hydro / irrigation split for that matter). A simple way to test this would be to take the mid-point of the active zone (i.e. assume just one curve to target) and operate toward that at all times (giving you a very simple release function).

L503. Above you state that Clinton dam has a hydropower main purpose.

Figure 4. Is this average monthly discharge over a number of years, or are you showing a single year’s output?

L588 – this is an inadequate way to evaluate storage dynamics improvement. You have observation and results. Compute NSE / RMSE / KGE or similar for each dam (sim vs obs) and show the difference across a distribution (perhaps splitting by continent or large basin).

Figure 7. It’s not clear why the data-derived storage curves result in a different seasonal storage pattern than GloLAKES for North America. Aren’t the curves based on GloLAKES data?

I have attached a file with my comments.