The study “Interannual Variations of Terrestrial Water Storage in the East African Rift Region” addresses an interesting topic and provides some valuable insights. However, several issues need to be addressed before the manuscript can be considered for publication. I recommend major revisions based on the following main points and some other minor comments presented below.



Main points:



1.    The authors state that “human intervention in the form of dam management at Lake Victoria substantially contributes to the TWS variability” (lines 15-16); however, they didn’t provide a clear estimation of the magnitude of this contribution. It would be interesting to see the relative contribution of natural variability and human interventions to the observed TWS fluctuations.

2.    The study proposes a clustering approach to identify the East African Rift region as having similar interannual TWS dynamics. However, the justification for focusing on this specific region could be further improved by providing a stronger rationale for selecting this study area. The manuscript could highlight the East African Rift region's unique hydrological characteristics, ecological significance, or socio-economic importance.

3.    Although the study compares TWS variations with precipitation, evapotranspiration, and surface water storage in the major lakes, the analysis of the underlying drivers remains somewhat superficial. The study could provide more information about the potential mechanisms that link these factors to TWS variability in the region (e.g., land use/land cover changes, soil moisture dynamics, groundwater recharge, and human water abstractions). A more comprehensive discussion of these drivers would beef up the interpretations and conclusions of the study.

4.    The study cites some relevant literature; however, it could improve by discussing how the proposed study’s findings compare to or advance previous research on TWS variability in the East African Rift region. The paper would benefit from a more thorough synthesis of the existing knowledge and a clearer articulation of this study’s novel contributions.

5.    The study lacks a thorough assessment of the uncertainties associated with the GRACE/GRACE-FO data, the precipitation and evapotranspiration datasets, and the surface water storage estimates. It would be interesting to see a more detailed description of the potential sources of error and their implications for the results. Also, the authors could elaborate more on the limitations, such as the coarse spatial resolution of GRACE data and the lack of ground-based validation data. These limitations could be explicitly acknowledged and discussed.

6.    The current conclusion section is somewhat vague and does not fully address the broader implications of the findings for water resources management, ecosystem conservation, or climate change adaptation in the region (conditioned to the rationale for selecting the study area as per comment 2). The authors could elaborate on the potential applications of the study’s findings.



Minor comments:



7.    Between lines 35-40, where it is “Niger Basin in West Africa,” it should be “Volta Basin in West Africa” in the context of the sentence.

8.    Lines 134-137: The description of the water occurrence map processing is unclear. Please provide more details on how the 95% occurrence threshold was determined and how it affects the estimation of lake surface areas.

9.    Lines 139-143: Please discuss the limitations of the surface water storage analysis based on a simplified relationship between lake level and area changes based on empirical cumulative distribution functions (ECDF). What could be the potential uncertainties it introduces in the storage estimates? For example, the monotonic and continuous relationship between lake level and area might not always be the case in reality. Lakes with complex bathymetry or irregular shorelines may exhibit non-monotonic or discontinuous relationships between level and area. However, the ECDF approach can handle outliers or anomalies in the input data more robustly than a linear regression used by Ferreira et al. (2018).

10.    Lines 240-247: The discussion of the differences between the two SPEI datasets seems speculative. Please provide more evidence to support the claim that the divergence after 2008 is caused by differences in precipitation data rather than PET estimation methods.

11.    Lines 290-295: The description of the Nalubaale Dam and its impact on Lake Victoria's water levels is incomplete. Please provide more information on the characteristics of the dam (e.g., operating rules) and downstream effects on the Victoria Nile and other water bodies. A study area section presenting the East African Rift Region would be useful.

12.    Lines 314-315: Please provide a more rigorous assessment of the data quality and its impact on the correlation analysis.

13.    Lines 367-368: That concluding statement seems too broad and not fully supported by the analysis. Please refine this conclusion and provide a more nuanced interpretation of the relative contributions of natural and anthropogenic factors to TWS variability.

14.    Please revise the English since there are several issues (e.g., Line 5 shows “region region”, Line 92 shows “We analyses…”)

General Comments

The study presents an analysis of long-term variations in GRACE total water storage variations (TWS) over the last 22 years for Africa and compares the data to surface water storage (SWS) variations in major lakes derived from satellite altimetry in central Africa. The authors compare the datasets also with meteorological/drought data via time series analysis and statistical methods. They discuss the influence of human and climate on the variability in TWS and surface SWS in Central Africa. They also provide some novel insight into the TWS dataset through a cluster analysis for the continent Africa. The authors have conducted a good work. It provides detailed information on data and methods used and provides very interesting insight into the water storage variations in the study area around Lake Victoria in Africa. I do think, however, that a more structured organization of the manuscript; a quantification of uncertainty of the surface water storage estimates; and, following from that, a more comprehensive discussion and concise conclusion of the results would make the work clearer and more significant.

On organization of the work: In Section 3-7, the authors cover certain topics. For Section 3-5 they combine respective methods, results and discussions into one section. For Section 6, some of the relevant methods are explained in Section 2, then some more method description is added after results and discussion in this section. The authors often jump between results presentations along various figures and corresponding piece-wise discussion and conclusions. It makes it harder for the reader to discriminate objective facts from opinion or suggestions by the authors. Some of this becomes especially a problem in Section 5 and even more so Section 7, which are presented the least clear. A clearer structure should be introduced to the entire manuscript, for example, for example, either the methods, or the discussions should be split off in some way. Also, some of the figure organization need some improvements, for example some legends are incomplete. Introducing panel letters might help to address results in figures with more than two panels more clearly. Some figures might be better suited for a supplement. Further suggestions are given below.

On uncertainty of the results: estimates of TWS from GRACE as well as SWS from altimetry and subsequent modeling includes several sources of uncertainty, e.g. measurement errors, parameter uncertainty. These uncertainties should be discussed. But since the authors are quantifying percentage of explained signal variance, a quantification is also suggested, especially for SWS. The conclusions need to be put into perspective of the uncertainties (see also next section).

On the conclusions: The study's conclusion on the nature of the driver of TWS variations, i.e. whether it is either human or climate during certain temporal periods, is not fully supported by the results and analysis provided. First, this statement is mainly directly addressed in Section 7, where water levels of various lakes and river level are compared, and the impact of dam management is highlighted. There is no direct comparison with SWS and TWS variations provided. Second, a correlation of TWS to drought indicators is not an explanation or proof of climate dominance, as stated in the conclusion (L357-358), because human water use (e.g. of surface or groundwater) itself is typically also heavily influenced by drought conditions and might therefore similarly impact TWS. In addition, in the rest of the manuscript, the authors only analyze the SWS portion of TWS variation but no soil moisture or groundwater, hence, a large portion of TWS variation remains unexplained, and therefore a conclusion on human or climate dominance in TWS remains very speculative. I am also wondering if such a conclusion is even relevant to emphasize on the importance of the work, but rather may take away from the actual interesting quantitative and qualitative findings of the work on the importance of SWS in the region. This could be more highlighted by slightly altering the discussion of the findings.

In addition, the authors do not comprehensively quantify and discuss why TWS may be rising overall in the Central Africa/Lake Victoria region over the last two decades (they did so only for specific sections of the TWS time series or in relation to P and SWS). It was shown that precipitation plays an important role. However, the P increase (or change in ET) does not indicate if and where the water is stored. (Here, the authors could also make the role of the hydrological processes - flux versus storage - more clear in the work.) Then, is the overall TWS increase mostly due to the accumulation of water in the lakes/reservoirs, or may other storages also play a role? The results the authors show, do suggest that quite some of the increase sources from the lakes. However, since up to 50% of annual variations occur only during very specific times, e.g., dry years (further comments below), and the size of the linear trend is quite different (Figure 10, additional numeric quantification of this overall increase might be helpful to compare SWS and TWS) a large part of the interannual increase is still unexplained by SWS. However, the correlation between the time series (TWS and SWS in Figure 10, bottom) is striking and the overall rise over the last decades very congruent, just the amplitudes are not matching. So, the question is, does the uncertainty of the SWS amplitudes (from sensors and model parameter) (or from TWS) play a role here? Or are maybe other storage components besides SWS equally important for explaining TWS rise in the region? Just as an example (no need to cite), Werth et al. (2017) have suggested groundwater storage increase may play a role for the storage increase in the Niger basin, and the argument was supported by reports of increasing groundwater levels in the region. Since the cluster for Niger and Lake Victoria have some similarity, maybe groundwater might be relevant in your study area as well. Such or similar thoughts could be included in the discussion and conclusions of the work.

In addition, a few clarifications on the methods and discussions are requested in specific comments further below.

Specific Comments

Abstract: The authors state that the study’s main objective “determine whether natural variability or human interventions caused these changes” in TWS variations. However, based on the presented results, the authors can only discuss this for SWS, not for TWS, since they do not analyze other storage components (see comment above).

Introduction: Clarify why were specifically the interannual variations analyzed and not (also) the seasonal variations?

L91: SPEI is typically labeled a drought index. On the data website they define it as follows: “The SPEI is a multiscalar drought index based on climatic data.”

L130: Approach to estimate water area bases on optical data. How would the uncertainty of the water occurrence probability due to weather conditions affect the final SWS estimate of the study? Also, this drawback of visual light imagery has been solved by other studies that rely on radar data to detect surface water occurrence, with the advantage that they are not weather-dependent. The authors could include in the discussion, why they have not referred to such data instead, or how application of radar instead of visible light remote sensing images might enhance the accuracy of the method.

L137: cululative > culmulative

L145: Add a statement to further spell out what your assumption on the lake profile shape for the volume estimation is, e.g. how steep is the pyramid wall inclined?

L30ff/L147: Please clarify, if all lakes in the region were included? Or to what percentage are smaller lakes neglected?

Equation 2) How representative is such a profile for the lakes? This approach probably has some uncertainty because the lake wall angle is likely heterogeneity inclined, for example, shallower near the shore. Can this introduce a significant error to the total surface water storage estimate? And how large is the uncertainty? It would help to provide a reasonable range of uncertainty for this.

L151: I appreciate that the authors spatially filter the surface water data to mimic the sensitivity of the GRACE observations to water mass changes. The author’s did not, however, clearly state if the applied gaussian filter width of 350 km is comparable to that applied during the GRACE data processing as conducted for the COST-G dataset. A different filter width can significantly alter the amplitude in storage variations. Since the GRACE dataset used is a unified from various datasets, this might be a bit more complex to evaluate. However, a discussion of it is missing. Optionally, this could be included as another source of uncertainty in the surface water storage time series.

151: Please indicate how the filtering was conducted, e.g. in the spatial or frequency (spherical harmonic) domain.

L154: the term “simple” is vague here. I assume you are referring an assumption for stationarity of the temporal components in the time series, as stated further below in L159? Different approaches available (e.g. fourier based, or others) are not more or less simple, but instead they are potentially better applicable to climate processes. Also, the non-stationarity of climate signals is not only present in seasonal components but also in the inter-annual/trend components, hence, why STL is better applicable for both. Please rephrase to make this clearer.

L160ff: how does the smoothing parameter affect the signal decomposition? What was the criteria for choosing them. I understand this is a trial and error approach, and requires some empirical decision making. However, it would be good to try to write down what you were aiming for, when choosing the parameter.

L156/Section 3: Please indicate if the STL is loss-free or not.

L171/Figure 4: If I understand this correctly, the black time series (original in a) is corresponding to the blue long-term signal in b (no-data gap)? I wonder if it makes sense to match the color (same in c and d)?

Figure 5&6: The clusters are coded two ways, once by colors and once by numbers. It would be easier to if this is limited to either one. Or also add the colors in the titles, behind numbers in figure 5, e.g. cluster 5 (red) and add numbers to colored dendogram in Figure 6.

Figure 6: I was wondering, if it would be sufficient to have this in a supplement. The additional information is minor, as the time series in Figure 5 already show degree of similarity.

L206: I suggest to add brief explanation: regions with overall positive trend are those located in Central Africa (including blue, yellow, dark green, pink).

L207ff: Here, the authors shift from a 7-cluster analysis to an 8-cluster analysis without a more detailed explanation. This should either be a new paragraph, to make that shift more clear.  Alternatively, I am wondering if Figures 5-7 could be combined. For example, why is cluster 8 not also shown in Figure 5?

L207: if I understand it correctly, the sub-clusters in Figure 7 are also appearing in the cluster tree in Figure 6, as the authors emphasis on that here. However, in Figure 6 they are colored all light blue. I was wondering, if it makes sense to mark the purple cluster 8 also in Figure 6, to be more clear.

L209: … has even larger TWS amplitudes than … > … has a larger TWS amplitude than ...

L210-211: change the word “marked” to ”significant”,“distinct”, or “fast”

L214/Figure 9: The graph in Figure 9 does not look like the values are accumulated, but rather filtered with some kind of moving-window filter of certain width (or accumulated within a moving window). In case of only accumulating, you would have only values every n months, with n being the accumulation period. Please clarify.

L214-215: You compare accumulated precipitation with SST filtered TWS. The two time series are treated with different methods. Are they really comparable this way? Why do the authors not also apply an SST filter (using the same parameter as for TWS) to the precipitation data instead? This would also save them from estimating the correct filter-width for P.

Figure 8 might also be ok for a supplement, instead of the main manuscript?

L220-2029: I am wondering if this can be shortened, as P becomes less relevant given their concluding that E is missing to better compared to TWS. However, this conclusion is rather trivial from a hydrological perspective.

L218: Maybe add a sentence explaining the purpose of the violin plot. Does the change in width of the blue areas (violins) have any meaning?

L232-233: unclear formulations, please rephrase a bit simpler.

L233-234: unclear formulation, rephrase. “…longterm observation of ?”; also you do not put P-E in relation to TWS, but SPEI

Figure 9: add precipitation to the legend.

L243: do > does

L253: for the names > for their names

L256: I cannot see the 50% in Figure 10, the color bar is kind of vague. The top left Figure 10 colors seem saturated given the color bar. What are the maximum value in Figure 10 top row? It looks to me more like 30%, given the time series in Figure 10 bottom.

Figure 10: The red polygon shown in the upper three panels is neither labeled the legend, nor in the caption. I assume it is outline for cluster 7? Please add.

L261: space missing

Figure 11 caption: correct spelling of de-sesonalized

Figure 11: compares PEV and correlation for de-sesonalized SWS and TWS. It would be useful to show the deseasonalized time series somewhere, e.g. add to Figure 11 or Figure 10 bottom?

L285: the 50% occur only for years with very low TWS, but not for wetter years. Hence, this feels like an overstatement (also in the abstract). Maybe it would be more representative to also estimate the median or mean of the explained percentage over the years? Or it would be more transparent to discriminate between dry and wet years (see also comment for absract above)?

L289: Victoria Nile > Nictoria Nile River

L291-295: this information might be better suited already in Section 2.3 to provide more detail on the surface water bodies in the region and how they are managed. It would already help for understanding previous sections.

L311: Can you provide a reference to support this statement?

L235: govern > governed

L235-235: sentence unclear, reformulate

Figure 13: This is not compiled well to support the discussion in Section 7. Maybe presenting the time series in a single or stacked panels and/or in comparison to TWS and/or SWS time series would help the purpose more?

L363: reformulate sentence, a lake cannot lead, rather results for the lake.

L360-362: I disagree, SWS does not fully explain the steady increase of TWS, as shown in Figure 10, only partially. The value of this multi-year TWS/SWS rise was also not quantified in the manuscript, maybe it would help to add this?

L366: The connection between dam discharge and TWS is not clearly shown in the manuscript.

References

Werth, S., White, D., & Bliss, D. W. (2017). GRACE Detected Rise of Groundwater in the Sahelian Niger River Basin. Journal of Geophysical Research: Solid Earth, 122(12), 10,459-10,477. https://doi.org/10.1002/2017JB014845