As attached.

Review of HESS-2021-72

This paper develops optimal water sharing policies for transboundary systems, focusing on the Grand Ethiopian Renaissance Dam. The technical approach is sound, and this system is an important one to study. The results are convincing because they show the tradeoff between minimizing storage variability (i.e. maximizing hydropower) and minimizing release variability for downstream water supply.

I believe that some moderate revisions to the structure and framing would make this paper a stronger contribution to HESS.

1. My biggest concern is confusion over the methodology, both what is being done, and the reason for the steps. The first step makes sense, optimizing a poicy structured as a radial basis function to allow flexible use of information. In my understanding, the steps are: optimize a RBF policy, use it to infer a linear drought constraint, and then use this constraint to re-optimize another RBF policy. Is that correct?

If so, why are these last steps needed? For example, the flow of logic in Figure 3 is very unclear. It seems to repeat optimization steps in places. How is the drought mitigation policy different from the RBF policy, and why is it needed? The RBF is already a function of storage, inflow, and the month. The drought mitigation policy sets minimum annual constraints for water releases. But it seems that this could have been part of the original RBF optimization, and/or enforced in the simulation model to ensure that the constraint is met at all times.

More explanation and structure on these points throughout Sections 1-3 would be very beneficial. While the results are convincing overall, I was not sure that the drought mitigation policy is needed in the end (Fig 14), as there is only a small advantage to these extra steps.

2. It is also not clear how the drought mitigation policy uses forecast information, or why the original RBF policy does not. Especially because around Line 345 the results suggest that the forecast information is not very useful in the optimized policies.

3. Another concern is the novelty of the approach. The study follows best practices for direct policy search (DPS) and arrives at a convincing result. However, I believe this approach has been used for transboundary systems before. It seems the new component of this study is deriving water-sharing policies (i.e. annual linear constraints) that are specific to drought periods. This may be a novel contribution, but as in point (1) it is not clear why this needs to be done here. The link could also be stronger between the linear constraint and the idea of a negotiated water-sharing policy between transboundary stakeholders.

Last, if the novelty relates to transboundary basins, is there any component of the methodology that is specifically designed for this case? The contribution may be more general, although the transboundary application is critically important.

4. The results are a bit too long, with 14 figures. These could be condensed to sharpen the contributions. In my opinion a few figures that might be removed are: Fig 7, Fig 9 (previous figures already show the change in standard deviation), and Fig 12. 

5. In the introduction and methods, there are several places where the references are grouped together in long lists. It would be stronger to highlight individual contributions from these previous studies where possible. 

A minor point about the introduction: clearly simulation-optimization is relevant, but why is policy fitting relevant to this study? 

6. There are several possible discussion points that could be included briefly. First, the tradeoffs found in this study are based on a period of historical data, but what about the future? Do we expect patterns of variability to be similar? Second, is there a way to better understand the characteristics of the optimized policies so that they could be reported to stakeholders, for example? These can be points for future work but deserve some discussion.

This paper presents a water-sharing policy framework that incorporates reservoir operating rules optimization based on conflicting uses and hydrologic variability, specifically tailer to drought conditions. The framework is illustrated using GERD as a case study. The results clearly show the trade-off between annual hydropower generation and the inter-annual variability of releases. The paper is well-written and the topic should be of interest to both researchers and practitioners. As explained below, my main concern is with the methodology, which seems to be overly “complex” given the relative simplicity of the case study.

Comments/questions:

1. The emphasis is put on average annual power output. However, power companies are also concerned by the firm energy, i.e. the energy output that you can guarantee 90% or 95% of the time. Can you show us what the trade-off would look like when the average energy output is replaced by the firm energy?
2. Does the term “water releases” include turbined outflows AND spillage losses?
3. When minimizing the variance of water releases, do you end up to a point where the energy output starts decreasing due to excessive spillages losses?
4. My main concern. Why didn’t you constrain the operating rule with a minimum “water release” (or minimum deviation from a target release) in the first step of the methodology and construct your Pareto front by varying that minimum like in the traditional constraint method in MOP? The system is small (just one reservoir) and it looks like to me that the Pareto front could be traced out using mathematical programming techniques in a MO framework. In my opinion the introduction must be revised to better explain why that framework was proposed instead of traditional MOP approaches.
5. Line 90. Please check the paper from Teasley and McKinney, JWRPM, 2011 on water and benefits sharing in the Aral Sea Basin.
6. I have a small gripe with the title. The methodology is actually applicable to any reservoir and it is not limited to transboundary river basins. Please remove “transboundary” from the title and revise the text accordingly.
7. Line 56. A wide variety of physiographic conditions is not limited to transboundary river basins!
8. Figure 12. Could you also include spillages losses and evaporation losses? Keeping the water level as high and as constant as possible will likely increases these two losses, up to a point where they can negatively impact the power output and the total outflows.

Review of Article hess-2021-72

Transboundary water sharing policies conditioned on hydrologic variability to inform reservoir operations

Author(s): Guang Yang and Paul Block

General comments

The article is well-written, scientifically sound and within the scope of  the journal.

Detailed comments

Following are the aspects that could be incorporated into the article.

The actual values of the constraints defined in equation 4 to 7 need to be specified. This is especially pertinent because the range of optimal power generation obtained in the study range from 1707 to 1788 MW while the estimated power generation of 15130GWH per year which implies an average power generation of 4202 MW. The reservoir storage trajectories on Figure 6d reveal an assumed maximum storage of 74 billion cubic metres which is the intended capacity of GERD. The article does not indicate whether the upper power generation limit (PU_t in equation 7) equals the intended installation capacity of 5000 MW. On Figure 6d, the reservoir trajectory for a power output of 1788MW is very high and close to full storage for most of the period suggesting that achieving 15130GWH (4202 MW) would require more of what has been simulated as spillage to run through the turbines to generate more electricity. A discussion of how the analysis here relates to the intended installed capacity and power generation would enhance the relevance of the article to the practical transboundary issues regarding the operation of the GERD.

Following are the aspects that could be incorporated as recommendations for further work.

The study applies the historic sequence (from 1965 to2017) just downstream of the GERD dam and reliability considerations are incorporated by the statistical treatment of the residuals of the linear function relating annual releases to annual inflows (equation 10) during low flow years. The resulting range of variability for the different exceedance levels as illustrated in Figure 4 is low and probably underestimates the impact of hydrologic variability. Since the historic sequence is not very long and seems to include only two severe drought periods (from 1978-1988 and 1991 - 1997 as seen on Figures 7 and also reflected on Figure 6d), the extension of the historic inflow records using (its correlation with) the longer-term records available in the Nile basin could enable a more realistic assessment of the effects of droughts on the system and how the GERD could be best operated during such severe droughts. It is during such periods of severe water shortage that tensions are likely to rise among the riparian countries. A more comprehensive probabilistic approach based on stochastically generated ensembles of the extended inflows could also be considered.