the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Oct 2022
Historical Analysis of Reservoir Storage Trends and Resilience Across Contiguous US from 1980–2019
Abstract. All major river systems in the Contiguous United States (CONUS) are impacted by dams. Many regional and global studies have looked at reservoir resilience to extreme events and quantified static characteristics, yet analysis of historical reservoir operations has been limited by a lack of data. Here we use the first national dataset of historical reservoir operations in CONUS, ResOpsUS, to analyze reservoir storage trends and operations over the last 40 years. We characterized seasonal operating patterns and show clear regional trends. In the eastern US which is dominated by flood control storage we see that storage peaks in the winter months with sharper decreases in operational range in the summer. While in the more arid western US where storage is predominantly for irrigation, we find that storage peaks during the spring and summer with increases in the operational range during the summer months. The Lower Colorado region is an outlier because it is arid and dominate by irrigation, but its seasonal storage dynamics more closely mirrored that of flood control basins. Consistent with previous studies we show that reservoir storage has decreased over the past 40 years, although our national fraction filled decreases are 50 % less than those shown previously. We also find that declines are occurring faster in more arid regions. Operational ranges (i.e. the difference between monthly max and min storage) have been increasing over time in more arid regions and decreasing in more humid regions. We also quantified hydrologic drought using the standardized streamflow index (SSI) and compared time it took for reservoir storage (expressed as anomalies in fraction filled) and SSI to recover. As would be expected, we see longer drought periods and more prolonged negative reservoir storage anomalies in the more arid basins. That said, nearly all regions have we show that the reservoir storage takes longer to recover from drought that the streamflow.
- Preprint
(2857 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2022-1051', Anonymous Referee #1, 15 Nov 2022
This paper presents an analysis of reservoir storage trends across the contiguous United States (CONUS), based on a new dataset of historical storage that authors put together recently (ResOpsUS). I agree with authors that conclusions based on this dataset would be a useful complement and even an upgrade on existing studies of reservoir behaviour that are based on other data sources. I also broadly agree with the main findings. Yet, this study has several weaknesses, detailed below:
- It is not clear what the direct connection is with the special issue on the representation of reservoirs in hydrological models. In reality, conclusions are not related to this topic. Literature on the topic is introduced at times, and the methodological assumptions made when modelling reservoirs are discussed (e.g. lines 104-110) but with never a mention of how this study actually relates to any of these assumptions.
- More generally, the scope of the paper is unclear. Authors (rightly) realise that there are many ways an analysis of this data might contribute to the literature, but they seem to not choose one (or more) focal point(s). As a result, the literature review is the introduction is disjointed, and several themes are broached without a clear focus.
- A major selling point of the paper is the use of new data, but in fact, this is not exploited by the paper. They do not rely on a comprehensive review of existing knowledge on reservoir storage / operations in the CONUS to propose a systematic view of how this dataset completes or updates this existing knowledge (and this would be a low hanging fruit for a useful paper!).
- The analysis has two weaknesses in leveraging the dataset to derive insights. (i) Authors aggregate by regions without ever looking at individual reservoirs. (ii) A consequence of this aggregation is to drop key regions full of reservoir such as the northwestern U.S.. With these choices, it is not clear they take full advantage of the fine grain data offered by their new dataset.
- As a consequence of unclear scope and methodological choices, findings are all unsurprising and it is never clear how they compare in relation with the literature. Besides, many of the findings on drying at the national level seem to be related to drying in one particular region: the southwestern U.S and in particular the Colorado river basin. It would be great for authors to quantify this contribution from a particular region to national conclusions, or abstain to make these national conclusions.
- The resilience angle is interesting, but the concept is not clarified until it is clarified as a recovery metric, and no definition is proposed until the discussion (lines 602-605). If authors want to use their dataset to look at the drought recovery ability throughout the U.S., that is potentially a great idea (if they can show there is a gap in the literature there), but I’d suggest they leverage the dataset by looking at the data from individual reservoirs.
- Last but not least, the writing is unequal (see e.g. the abstract) and suggests a paper hastily put together. This impression is reinforced by the unclear scope and disjointed references to the literature (e.g., resilience definitions proposed in the discussion).
For these reasons, my take on this work is that it is probably not ready for publication at this stage. There is clear potential though, and I’d like to suggest than rather than trying to push preliminary work through, authors have much to gain by electing to do a deeper analysis before publishing their findings, probably over several excellent papers. I am also not sure they need to focus on this special issue until and unless they can use the data to address some of the assumptions made in representing reservoirs in hydrological models.
A few detailed comments (given the stage the paper is at I do not provide this for the whole text):
The text would benefit from a rigorous round of edits. Just in the abstract I found a series of imprecisions that would have been fixed by proofreading:
Line 7: as soon as they introduce the phrase, authors need to explain what “reservoir resilience” means in their context
Line 13: “dominated” instead of “dominate”
Line 15: “reservoir storage has decreased”, is it average annual storage?
“our national fraction filled decreases are 50% less than those shown previously”, please rephrase to make this clearer to the reader.
Lines 16-17: the definition of operational range is great to see but should occur the first time the term is introduced.
The idea that the lack of data has been a limiting factor (lines 8 and 27) deserves clarification: what do you see thanks to the new dataset that studies with less data could not?
Line 38: “there are large dams are spread out”
Line 39: provide reference for the standard.
Line 100: repetition of the idea expressed in lines 28-30.
Lines 104-110: is the aside on reservoirs in hydrological models part of the paper’s scope?
Methods: this is a collection of indicators. Many of them are interesting to look at, but it is not clear how they complement each other, and what question(s) they are meant to answer when taken together.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC1 -
AC1: 'Reply on RC1', Jen Steyaert, 11 Jan 2023
This paper presents an analysis of reservoir storage trends across the contiguous United States (CONUS), based on a new dataset of historical storage that authors put together recently (ResOpsUS). I agree with authors that conclusions based on this dataset would be a useful complement and even an upgrade on existing studies of reservoir behaviour that are based on other data sources. I also broadly agree with the main findings. Yet, this study has several weaknesses, detailed below:
Thank you for the thoughtful review and for acknowledging the usefulness of our work. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally must rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we can provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region has an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
- It is not clear what the direct connection is with the special issue on the representation of reservoirs in hydrological models. In reality, conclusions are not related to this topic. Literature on the topic is introduced at times, and the methodological assumptions made when modelling reservoirs are discussed (e.g. lines 104-110) but with never a mention of how this study actually relates to any of these assumptions.
We agree that we were not explicit enough in our original manuscript regarding its connection with the special issue. We have outlined our relevance in our overview above as well as our plan to make this clearer moving forward. As stated above, our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally must rely on large scale assumptions of how reservoirs operate. Specifically, our findings support assume minimum storage is approximately 10% of the maximum storage capacity for all but the 100 reservoirs that have storage capacity values in GRanD less than observed maximum storage in ResOpsUS. Additionally, our seasonal analysis and storage trends are both useful to model calibration. Finally, this regional based analysis may be useful for developing reservoir operations and dynamics in areas outside the US that have similar characteristics such as aridity, reservoir uses, and population that are data limited.
We also agree that updating the document to provide an example of why this study is important is necessary to rounding out the introduction. To make this connection clearer, we will edit lines 124 – 127 to include the dataset description. Next, we also will also add a paragraph at the end of the introduction to include why this analysis is important based on the reservoir assumptions on lines 104 – 110). Lastly, we plan to weave in the narrative regarding assumptions and modelled rule curves into our analysis.
- More generally, the scope of the paper is unclear. Authors (rightly) realise that there are many ways an analysis of this data might contribute to the literature, but they seem to not choose one (or more) focal point(s). As a result, the literature review is the introduction is disjointed, and several themes are broached without a clear focus.
Thank you for your comment. To complement our detailed response above, we plan to point out two main questions and focus our discussion and conclusion on analyzing the assumptions (stated above) made by modelling studies. The two main questions are as follows:
- What is really happening with reservoirs both seasonally and over time and how does this compare with the assumptions previously made in large scale modeling studies?
- How have reservoir operations evolved over time and do we see changes in the response to droughts?
To ensure cohesion with these questions, we will focus on 1) what really is happening with reservoirs both seasonally and over time during the 1980 – 2019 period, 2) what drought sensitivity looks like to hydrological drought during the period from 1980 – 2019 and 3) are the assumptions in reservoir models accurate. Additionally, our discussion and conclusion will be reorganized to discuss the ways in which our findings support or refute the common assumptions in reservoir models as stated above. Most importantly, this restructuring will focus on the following results mentioned in the table below.
- A major selling point of the paper is the use of new data, but in fact, this is not exploited by the paper. They do not rely on a comprehensive review of existing knowledge on reservoir storage / operations in the CONUS to propose a systematic view of how this dataset completes or updates this existing knowledge (and this would be a low hanging fruit for a useful paper!).
Thank you for this comment. As stated in our detailed response above, we will expand our results and discussion to directly point to the ways in which we are increasing understanding of reservoir operations through time. As stated above, previous models typically use optimized rule curves that are based on assumptions regarding how reservoirs work, but not necessarily on the historical data. Therefore, we will focus our results and discussion to show ways in which these assumptions may be accurate or inaccurate depending on the data. Additionally, we will note the select ways in which our results either compliments or correct analyses.
Some of the key points we will highlight include the fact that historical studies have shown that reservoir storage in CONUS has decreased slightly with most studies stating estimates between 0 – 15% regionally. In contrast, our analysis demonstrates that fraction filled has only decreased by 0.2% nationally and the Lower Colorado (with the sharpest slope) has a decrease of only 1.2%. Historical studies have shown that reservoirs increase drought severity which our findings align with in all cases except for the South Atlantic region. Our drought sensitivity analysis is also useful for determining if climate or changes in demand could be behind the storage changes, we observe across the US. Finally, the novelty of our operational range analysis is not to be underestimated as the addition of more reservoirs, operators and temporal scale allows us to evaluate changes not only regionally, but also monthly. This is important for calibrating hydrologic models, but also to show where reservoir operations are widening or tightening in response to demand and/or climatic shifts.
- The analysis has two weaknesses in leveraging the dataset to derive insights. (i) Authors aggregate by regions without ever looking at individual reservoirs. (ii) A consequence of this aggregation is to drop key regions full of reservoir such as the northwestern U.S.. With these choices, it is not clear they take full advantage of the fine grain data offered by their new dataset.
We agree that we can do more to leverage the uniqueness of ResOpsUS and in response to this comment we will add two figures focused on the individual dams. The first will be a map with point locations of all 678 reservoirs in ResOpsUS colored by if their storage trend is positive or negative over the 40-year period we are looking at. We will also cut Figure 4 and provide a two-panel map with the point location of each dam colored by the month of peak fraction filled (panel a) and the month of largest operational range (panel b). In addition to including the individual dam plots as stated above, we also increased our threshold for storage covered to be 40% which allows us to include HUC7 and HUC17 in our analysis. We now have 14 regions we are looking at out of the 18 total HUC2s.
- As a consequence of unclear scope and methodological choices, findings are all unsurprising and it is never clear how they compare in relation with the literature. Besides, many of the findings on drying at the national level seem to be related to drying in one particular region: the southwestern U.S and in particular the Colorado river basin. It would be great for authors to quantify this contribution from a particular region to national conclusions, or abstain to make these national conclusions.
Thank you for your comment. In response to this and other comments, we have provided a table denoting specific areas where our results align or refute previous work below. Additionally, we would like to note that while our findings align with current and past research, the novelty of ResOpsUS goes beyond previous work and allows us to provide greater insights into how reservoirs are operated. That said, we do note that greater emphasis on the impact of regional and local changes to the trends at the national level will enhance our argument. We plan to incorporate more discussion on this in our drying section.
Additionally, we also wanted to highlight some of the key results and the importance of the operational range, storage trends, and drought sensitivity analyses. First, these operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. That said, the maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). In fact, we observe tighter operations in Tennessee, Ohio and South Atlantic, while California and Sourris Red Rainy have wider operations.
To supplement the operational range analyses, we also want to highlight the fraction filled changes over time on the three spatial scales we looked at: CONUS wide, regional, and individual reservoirs. These three scales are important as the first two (CONUS wide and regional) allow us to directly compare to historical findings. Additionally, these findings tell us how reservoirs have historically been operated and about their drought sensitivity as reservoirs with increases in fraction filled have shifted their operations to increase fraction filled levels, while reservoirs with decreasing fraction filled trends are potentially facing unsustainable water demand that operations cannot keep up with. Specifically, we note that our regional analysis aligns with previous work in the western United States, however, it contradicts previous studies as storage in the eastern United States has also declined with the exception of the Tennessee, and Sourris Red Rainy basins
Lastly, we would like to highlight our regional drought sensitivity analysis. This compliments the fraction filled trends and allows us to provide potential reasons for the fraction filled trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high fraction filled ratios that denote fraction filled takes much longer to recovery than streamflow. This suggests that the decreasing fraction filled trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (fraction filled takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this recovery ratio.
- The resilience angle is interesting, but the concept is not clarified until it is clarified as a recovery metric, and no definition is proposed until the discussion (lines 602-605). If authors want to use their dataset to look at the drought recovery ability throughout the U.S., that is potentially a great idea (if they can show there is a gap in the literature there), but I’d suggest they leverage the dataset by looking at the data from individual reservoirs.
As stated above, we have opted to remove the resilience section and simply focus on the drought sensitivity. We believe that this will be a better focal point for the analysis as it complements the current work done by Turner et al., 202 regarding data driven reservoir operations. Most specifically, this analysis demonstrates regions where reservoir operations are sensitive to drought. This is important for modelling reservoir operations as Turner et al., 2020 showed that more drought sensitivity dams should undergo rigorous sensitivity and uncertainty analyses.
- Last but not least, the writing is unequal (see e.g. the abstract) and suggests a paper hastily put together. This impression is reinforced by the unclear scope and disjointed references to the literature (e.g., resilience definitions proposed in the discussion).
Without more details we are not sure exactly what the reviewer is pointing to with respect to this comment. However, in response to this comment, we have already started correcting the organization of our paper and will continue to polish the paper per the individual edits below. In addition, we are planning to do an extensive read through to make sure language is consistent, grammatical errors are corrected, and the paper is logically organized.
For these reasons, my take on this work is that it is probably not ready for publication at this stage. There is clear potential though, and I’d like to suggest than rather than trying to push preliminary work through, authors have much to gain by electing to do a deeper analysis before publishing their findings, probably over several excellent papers. I am also not sure they need to focus on this special issue until and unless they can use the data to address some of the assumptions made in representing reservoirs in hydrological models.
Thank you for this comment. We greatly appreciate the insights you have provided and feel that the paper will be greatly enhanced by these additions. We also strongly believe that these results will be beneficial to the larger reservoir community as stated in our general response above, reviewer comments 1 and 5, and the table below.
A few detailed comments (given the stage the paper is at I do not provide this for the whole text):
The text would benefit from a rigorous round of edits. Just in the abstract I found a series of imprecisions that would have been fixed by proofreading:
Line 7: as soon as they introduce the phrase, authors need to explain what “reservoir resilience” means in their context
Thank you for this comment. We will remove the reference to reservoir resilience as we do not believe the analysis at this point allows us to directly speak on resilience. Rather we will update this to be drought sensitivity.
Line 13: “dominated” instead of “dominate”
We have corrected this mistake.
Line 15: “reservoir storage has decreased”, is it average annual storage?
You are correct, it is average annual fraction filled. We have added that to the beginning of the comment.
“our national fraction filled decreases are 50% less than those shown previously”, please rephrase to make this clearer to the reader.
We have corrected this to say “a much smaller decrease than previous work” line 16.
Lines 16-17: the definition of operational range is great to see but should occur the first time the term is introduced.
Thank you for this comment. We have moved the description of operation range to first instance of the phrase “operational range” on line 12.
The idea that the lack of data has been a limiting factor (lines 8 and 27) deserves clarification: what do you see thanks to the new dataset that studies with less data could not?
To more clearly highlight the novelty and importance of this work, we have created Table 1 below to directly show how our results agree or disagree with previous work. Mainly, we see that our analysis supports that storage has declined across CONUS, similar to previous work, yet the magnitude of that decrease is different. Additionally, the regional differences are supported in the western US, but not in the southeastern US as Ohio and Lower Mississippi also have declines. Previous studies only looking at storage capacity have stated that the Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 – 2000, yet we see there are decreasing storage trends in reservoirs along this period. Lastly, when looking at operational targets in Patterson et al., 2018, California dams spend more time under target and eastern dams in wetter regions spend more time above operational targets. Our operational trend analysis suggests the opposite.
Line 38: “there are large dams are spread out”
We have rephrased this to be clearer and more direct.
Line 39: provide reference for the standard.
We have added the reference to iCOLD in the bibliography
Line 100: repetition of the idea expressed in lines 28-30.
We have rephrased this section.
Lines 104-110: is the aside on reservoirs in hydrological models part of the paper’s scope?
We plan to include one to two paragraphs discussing the assumptions made in hydrologic models and how our work compliments it. We added a more detailed response to this under comment 1.
Methods: this is a collection of indicators. Many of them are interesting to look at, but it is not clear how they complement each other, and what question(s) they are meant to answer when taken together.
Thank you for your note. We believe that the organization we propose in comment 2 combined with more direct connections to the assumptions in reservoir models will pull this paper together. We also believe this will enhance our story and provide valuable validation points for hydrologic modelers.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC1
-
RC2: 'Comment on egusphere-2022-1051', Anonymous Referee #2, 02 Dec 2022
The authors have assembled a new data set on reservoir operations that has great potential to generate new insights. They analyze this dataset to explore seasonal and regional patterns as well as long term trends in reservoir operations. Specifically, they focus on the fraction filled, operational range and variance, and drought recovery in their analysis. Overall the paper documents an exploratory analysis which demonstrates some of the capabilities of the new data set. Unfortunately, this analysis stays at the exploratory level, is not clearly motivated by a research question or problem statement, and does not generate new insights or hypotheses. There is great potential here for the authors to use apply this data set to uncover new insights at the intersection of climatology, hydrology and reservoir operations. To do this would require either major revisions or resubmitting later as a new manuscript. To clarify these general reflections, I have provided several specific comments below.
Comments
- The introduction motivates the introduction of a new data set. However, it does not motivate the subsequent analysis. Because of this the reasoning behind subsequent analytical choices such as inclusion/exclusion criteria, aggregation choices, and metrics selection is unclear, and the reader is left without a clear benchmark by which to judge these choices. The authors can address this by posing one or more research questions and rewriting the introduction to motivate these questions and the approach taken to address them. Note that this change might require not only a rewriting of the introduction but the selection of different analysis and methods if warranted by the selected question(s).
- There is a great deal of diversity in both the physical design reservoirs and the design of their operational rules across the U.S. This diversity is driven by reservoir function(s), local seasonal and interannual streamflow patterns, and daily and seasonal demand patterns, among other factors. Therefore, I am concerned that what is observed when aggregated at the regional level is really the average of different signals, from different functions and operating priorities, and that this average signal has lost much of the interesting information. Further, per my first comment, it is not clear what the motivation for aggregating streamflow and reservoir storage by hydrologic regions.
- Across each region the size of reservoirs both in absolute volume and in size relative to streamflow, streamflow variability and demand, vary substantially. Pooling all reservoirs in each region based on capacity and filled volume heavily weights the findings to the largest reservoirs in each region. This could be warranted, depending on the research question, but again the motivation for this choice is not clear.
- Figure 5 presents drought conditions over the whole study area and the authors the discuss correlations between drought occurrence and severity, and changing in storage. On line 426 the authors mention that decreases in the interannual fraction filled are correlated with climatic shifts. Are they correlated in a mathematical sense or just in an approximate visual sense? If it is the later, please rephrase to avoid misinterpretation. Further, drought conditions vary so broadly over CONUS that I am unsure how this is meaningfully related to reservoir operations.
- Assessing resilience requires defining what function is to be recovered or maintained. Following from this, metrics serving to measure resilience should capture the function or functions of interest. The authors apply a drought recovery metric to assess reliance across regions where reservoirs have different dominant functions, notably flood control. In flood control dominated regions, assessing the resilience via a drought recovery metric is not meaningful.
- The discussion and conclusions do not make a clear case for what this manuscript adds to the existing science. The discussion and conclusions should respond to the motivation and aims presented in the introduction, clearly articulate what was learned and why it matters and put the findings in the context of prior work. The authors note on a number of occasions that this analysis confirms prior findings but do not make the case for the value of their own analysis.
Minor Comments
- Line 264, Figure c should read Figure 1c
- In the paragraph beginning on line 378, the authors reference both Figures 4 and 5. However, the reference to Figure 5 do not make sense and I believe all references here should be to Figure 4.
- Please clarify how monthly variance is calculated for Figure 4. Does the daily fraction filled mentioned in the caption refer to the average of all reservoirs in the region?
- To determine the drought severity shown in Figure 5, what spatial extent was used?
- On line 499 the authors state that the “second set are regions that have predominately positive trends and greater than or equal to three statistically significant trends (Souris Red Rainy, California and Upper Colorado).” However, Figure 7 shows that the Upper Colorado has predominately negative trends. Is the figure wrong or is the text?
- On line 524 the authors state that recovery times were capped at 50 months because only five regions had recovery times greater than 50 months. Given that there are only 11 regions in the study, this is not a reasonable justification. Please revise or further justify this choice.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC2 -
AC2: 'Reply on RC2', Jen Steyaert, 11 Jan 2023
Reviewer 2:
The authors have assembled a new data set on reservoir operations that has great potential to generate new insights. They analyze this dataset to explore seasonal and regional patterns as well as long term trends in reservoir operations. Specifically, they focus on the fraction filled, operational range and variance, and drought recovery in their analysis. Overall the paper documents an exploratory analysis which demonstrates some of the capabilities of the new data set. Unfortunately, this analysis stays at the exploratory level, is not clearly motivated by a research question or problem statement, and does not generate new insights or hypotheses. There is great potential here for the authors to use apply this data set to uncover new insights at the intersection of climatology, hydrology and reservoir operations. To do this would require either major revisions or resubmitting later as a new manuscript. To clarify these general reflections, I have provided several specific comments below.
Thank you for your comprehensive review. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally have to rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all of our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we are able to provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region have an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
Comments
- The introduction motivates the introduction of a new data set. However, it does not motivate the subsequent analysis. Because of this the reasoning behind subsequent analytical choices such as inclusion/exclusion criteria, aggregation choices, and metrics selection is unclear, and the reader is left without a clear benchmark by which to judge these choices. The authors can address this by posing one or more research questions and rewriting the introduction to motivate these questions and the approach taken to address them. Note that this change might require not only a rewriting of the introduction but the selection of different analysis and methods if warranted by the selected question(s).
Thank you for your comment. We do agree that the paper could be better motivated in the introduction. We also agree that our questions could be better described to make reading the paper easier. We have therefore opted to focus our introduction around the following questions:
- What is really happening with reservoirs both seasonally and over time and how does this compare with the assumptions previously made in large scale modeling studies?
- How have reservoir operations evolved over time and do we see changes in the response to droughts?
- There is a great deal of diversity in both the physical design reservoirs and the design of their operational rules across the U.S. This diversity is driven by reservoir function(s), local seasonal and interannual streamflow patterns, and daily and seasonal demand patterns, among other factors. Therefore, I am concerned that what is observed when aggregated at the regional level is really the average of different signals, from different functions and operating priorities, and that this average signal has lost much of the interesting information. Further, per my first comment, it is not clear what the motivation for aggregating streamflow and reservoir storage by hydrologic regions.
We initially aggregated the data so we could make comments about regional dynamics. Upon this review, we recognize the importance of an individual dam analysis, therefore we have opted to include two extra figures focused solely on individual dynamics. The first will be a map with point locations of all 678 reservoirs in ResOpsUS colored by if their fraction filled trend is positive or negative over the 40-year period we are looking at. We have also opted to cut figure 4 and provide a two-panel map with the point location of each dam colored by the month of peak fraction filled (panel a) and the month of largest operational range (panel b). In addition to including the individual dam plots as stated above, we also increased our threshold for fraction filled covered to be 40% which allows us to include HUC7 and HUC17 in our analysis. We now have 14 regions we are looking at out of the 18 total HUC2s.
- Across each region the size of reservoirs both in absolute volume and in size relative to streamflow, streamflow variability and demand, vary substantially. Pooling all reservoirs in each region based on capacity and filled volume heavily weights the findings to the largest reservoirs in each region. This could be warranted, depending on the research question, but again the motivation for this choice is not clear.
Aggregating the data by region, does cause the region to be weighted by the largest dam. Since regional waterways will be impacted by all the dams in a given region and weighted accordingly, we opted to aggregate the analysis by region. Based on our above questions, aggregating this data by region also allows us to compare with previous studies and note areas where large scale modelling study assumptions may not align with the regional results. That said, we do recognize that some of the regional dynamics seen are more indicative of the largest dam in a region. We will include two more regions and add two more figures focusing on the individual dams.
Additionally, the two questions outlined above in comment two will allow us to focus our analysis. Specifically, we will be able to refute or support modelling assumptions such as minimum storage equals 10% of maximum storage capacity and provide important trends for model calibration. To support the second question, we will provide more discussion regarding how our analyses fits in with previous work as shown in the general response and the table below.
- Figure 5 presents drought conditions over the whole study area and the authors the discuss correlations between drought occurrence and severity, and changing in storage. On line 426 the authors mention that decreases in the interannual fraction filled are correlated with climatic shifts. Are they correlated in a mathematical sense or just in an approximate visual sense? If it is the later, please rephrase to avoid misinterpretation. Further, drought conditions vary so broadly over CONUS that I am unsure how this is meaningfully related to reservoir operations.
The correlations are in an approximate visual sense. Since drought conditions and the reservoir uses vary across the entirety of CONUS, we opted to aggregate both for a more approximate analysis. This is also why we analyzed drought impacts regionally in Figure 6 and Section 3.3 so we could have a more quantitative analysis of drought impacts and sensitivity on storage. This figure is building the story of how storage has changed in CONUS over time and gives drought severity as one potential impact.
- Assessing resilience requires defining what function is to be recovered or maintained. Following from this, metrics serving to measure resilience should capture the function or functions of interest. The authors apply a drought recovery metric to assess reliance across regions where reservoirs have different dominant functions, notably flood control. In flood control dominated regions, assessing the resilience via a drought recovery metric is not meaningful.
Thank you for this comment. We agree that using a hydrologic drought recovery metric may not necessarily be accurate for addressing resilience in flood control reservoirs. TO remedy this, we have opted to remove the word resilience and focus on drought sensitivity. We still think that hydrologic drought sensitivity is important to flood control reservoirs as variations in streamflow impact their operational patterns.
- The discussion and conclusions do not make a clear case for what this manuscript adds to the existing science. The discussion and conclusions should respond to the motivation and aims presented in the introduction, clearly articulate what was learned and why it matters and put the findings in the context of prior work. The authors note on a number of occasions that this analysis confirms prior findings but do not make the case for the value of their own analysis.
Thank you for noting our results confirm previous results. We have provided a general response to the lack of novelty in our paper in the general comment above and noted key points in the below table where our work refutes or supports previous work. To summarize these two responses, our study, using observed storage from ResOpsUS, finds that the magnitude of storage decrease in CONUS is much smaller than previous studies have shown. Additionally, our operational range analysis points to widening of operational ranges in California, potentially to increase water supply, and decreasing of operational ranges in Tennessee, Ohio, and the South Atlantic. These results are in direct opposition to the analysis by Patterson et al., 2018 and the difference is potentially due to the higher number of reservoirs in our study. Our regional drought sensitivity also shows that only the South Atlantic has a positive recovery ratio and therefore has storage return to normal before a drought is over. Lower Mississippi and California have large storage ratios which denote that storage takes much longer to recover than streamflow. This drought sensitivity analysis allows us to more deeply analyze the potential causes of the storage trends in CONUS and provides a unique perspective on how observed reservoir operations have responded to droughts in the past forty years.
Minor Comments
- Line 264, Figure c should read Figure 1c
We agree and will fix this in the text.
- In the paragraph beginning on line 378, the authors reference both Figures 4 and 5. However, the reference to Figure 5 do not make sense and I believe all references here should be to Figure 4.
Thank you for your comment. Yes, all the references to figure 4 should be figure 5.
- Please clarify how monthly variance is calculated for Figure 4. Does the daily fraction filled mentioned in the caption refer to the average of all reservoirs in the region?
Monthly variance was calculated using the daily fraction filled values for each region. These daily fractions filled values are the average fraction filled values for all reservoirs in the region. We have opted to remove this figure from our analysis and will also remove the description of the results here.
- To determine the drought severity shown in Figure 5, what spatial extent was used?
To determine the drought severity, we aggregated the three-month SPI values from NCAR across the entirety of the CONUS domain. The NCAR SPI values do not have a stated spatial resolution in their metadata page. More information can be found on the NCAR page: https://climatedataguide.ucar.edu/climate-data/standardized-precipitation-index-spi
- On line 499 the authors state that the “second set are regions that have predominately positive trends and greater than or equal to three statistically significant trends (Souris Red Rainy, California and Upper Colorado).” However, Figure 7 shows that the Upper Colorado has predominately negative trends. Is the figure wrong or is the text?
We agree that this statement is does not match Figure 7. We will remove Upper Colorado from this grouping and add it to the same grouping as Arkansas White Red as those dynamics are much more similar.
- On line 524 the authors state that recovery times were capped at 50 months because only five regions had recovery times greater than 50 months. Given that there are only 11 regions in the study, this is not a reasonable justification. Please revise or further justify this choice.
Recovery times in the analysis were not capped at 50 months, rather we capped the figure legend at 50 months so that the variations in the smaller numbers could be seen. We have elected to change this wording to make this clearer.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC2
-
RC3: 'Comment on egusphere-2022-1051', Anonymous Referee #3, 08 Dec 2022
This paper presents a study of historical trends in reservoir storage on a national scale using a new dataset that provides historical operational data for reservoirs across the US. The seasonal and annual behavior of storage reservoirs across different regions were considered throughout the paper. While the new dataset, accompanied by the analysis of this research offers valuable insight into the subject, I still think that there are major issues that must be addressed before this paper is ready to be published. These issues are as follows:
- There exist many grammatical and dictation mistakes which suggest that the paper is rushed, and not enough time has gone into the revision and polishing process. In addition, there are many general terms used in the text that should be explained in more details. Some of these mistakes and general terms are mentioned in the minor comments.
- Considering that a major part of the analysis is statistical, I think that preprocessing of data should be explained in more details. For instance, how outliers were chosen, and how plentiful they were. What is the percentage of missing data that is calculated by linear interpolation? Or explaining Sen’s slope in more details and explaining why 0.1 value was chosen to signify a significant trend? In addition, you may consider using a larger number of distributions rather than GEV. Some of the points are also mentioned by line in minor comments.
- The results and discussion are mostly limited to identifying trends in reservoir storage using different metrics. There should be a much deeper focus on what is the implication of these trends, seasonality, and uncertainty in storage for water resources engineers and planners, and elaborate on how the insight from this research helps them to better understand the risks involved with a reservoir storage management and better prepare for the uncertain future.
- Does using the drought index for reservoirs that are designed for flood control provide any valuable insight? I think that it would be better to consider how trends in reservoir storage in humid regions affect the ability of reservoirs to control flooding, especially for extreme events.
Minor comments:
Line 28-30: Grammatically incorrect: based ‘on’ various remote sensing …
Line 32: You are mentioning that you have also used ‘release’ data to study the historical behavior of reservoirs, while in practice, you did not use the release data in your analysis.
Line 38: Grammatically incorrect: There are 2,000 large dams ‘that’ are spread out across the US
Line 25-38-132: In line 25, large dams are defined as having a capacity larger than 10 km3, while in line 38, the value is equal to 0.1 km3. Also in line 132, the values that defines large dams in 0.01 km3. Which one is the better definition of large dams?
Line 49:50: It is better to not use ‘in fact’ and ‘finally’ in the same sentence. You also mentioned ‘water supply’ in the previous sentence and are repeating it again here.
Line 54: Please explain how dams can decrease the severity of extreme events in one or two sentences. In line 74-75, it is also mentioned that reservoirs can increase the drought duration.
Line 63: How have dams changed the run-off regimes? Please explain in more details.
Line 65: What type of system? Please be a little clearer.
Line 57: Grammatically incorrect. I think that the word ‘promote’ can be removed.
Line 68: Grammatically incorrect. Use ‘have’ instead of ‘has.
Line 71: Disrupt how?
Line 72-74: The sentence is not clear. Two ‘increase’ in one sentence.
Line 93-94: The sentence is grammatically incorrect.
Line 112: There are two ‘is’ in a sentence.
Line 114-115: What do you mean by ‘reservoir dynamic’? I think that you can be more precise about it.
Line 119: What trend?
Line 123-126: Again, you are using general terms such as ‘reservoir patterns’, and ‘regional dynamics’. I think that you should be more specific in defining these terms.
Line 150: What do you mean by ‘hydrologic boundaries’?
Line 168-170: How plentiful were the outliers which suggested values higher than the storage capacity? And how much higher were they compared to the capacity?
Line 170: What was the average period of gap in the data? For instance, did you use the interpolation for a large gap of data, to the extent that interpolation may become irrational?
Line 177: I suggest using ‘received little impact’ instead of ‘had little impact’.
Line 187: Please explain the ‘Fraction Filled’ in a few sentences before introducing the formula.
Line 208: Please briefly explain ‘Sen’s slope’.
Line 224: Considering that you are using daily streamflow values, and not annual maxima, you should consider more than one distribution to find the best fit to the data. Or at least, you must use tests such as ‘Kolmogorov–Smirnov test’ to ensure that the probability distribution is decently representing the data.
Line 226: SSI values ‘can’ be calculated …
Line 232: I think it is better to write the formulation for both conditions of P being larger or smaller than 0.5.
Line 262: What do you mean by ‘reservoir setting’?
Line 266: ‘dominate’
Line 274: Hydroelectricity reservoirs are most common ‘in’ the Tennessee Basin and South Atlantic.
Line 328: ‘ion’?
Line 334: The statement does not seem true. Based on the Figure 2, the lower Colorado river has a very small month-to-month variation.
Line 337: ‘is’ shown in Figure 2.
Line 341:342: This statement does not seem correct based on the Figure 2.
350: ‘Note’ that …
For the figure 2, I think that in addition to the black line that shows the median Fraction Filled and purple shadow that shows maximum minus minimum, you can also add 95% confidence intervals (maybe as dashed lines) to make sure that an extreme outlier (minimum or maximum) does not greatly impact the shadowed purple part.
Line 378: It is mentioned that operational variance provides a more holistic measure compared to the operating range. However, I am not convinced why you had to use both metrics for the analysis. If variance is better, why using operating range? Does each one of them provide a separate vision that justifies using both? Otherwise, it might be better to use only one of them to avoid extra confusion.
Line 201 and 347: In line 201, it says that region with more than 40% of storage coverage are considered, while line 437 suggest the value as 50%. Which one is correct?
Line 438: Why October? Please explain the reason for choosing this month.
Line 514: On both regional and ….’a’ is extra.
Line 656: 83% ‘of’ regions …
Line 658-659: The sentence is grammatically incorrect.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC3 -
AC3: 'Reply on RC3', Jen Steyaert, 11 Jan 2023
Reviewer 3:
This paper presents a study of historical trends in reservoir storage on a national scale using a new dataset that provides historical operational data for reservoirs across the US. The seasonal and annual behavior of storage reservoirs across different regions were considered throughout the paper. While the new dataset, accompanied by the analysis of this research offers valuable insight into the subject, I still think that there are major issues that must be addressed before this paper is ready to be published. These issues are as follows:
Thank you for your comprehensive and thoughtful review. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally have to rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all of our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we are able to provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region have an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
- There exist many grammatical and dictation mistakes which suggest that the paper is rushed, and not enough time has gone into the revision and polishing process. In addition, there are many general terms used in the text that should be explained in more details. Some of these mistakes and general terms are mentioned in the minor comments.
Thank you for your comment. We will do an extensive edit for grammatical errors and ensure that all terms used in the paper are clarified appropriately. We will also ensure that all general terms are explained in detail to ensure clarity and congruency.
- Considering that a major part of the analysis is statistical, I think that preprocessing of data should be explained in more details. For instance, how outliers were chosen, and how plentiful they were. What is the percentage of missing data that is calculated by linear interpolation? Or explaining Sen’s slope in more details and explaining why 0.1 value was chosen to signify a significant trend? In addition, you may consider using a larger number of distributions rather than GEV. Some of the points are also mentioned by line in minor comments.
We agree that more description of the methods used are appropriate for the statistical analysis. Outliers in reservoir storage were chosen as points that existed above the storage capacity values in GRanD. In some cases, there were dams (approximately 100) where the storage consistently sat above the storage capacity in GRanD. These outliers could be a result of Lehner et al., 2011 replacing maximum storage capacity with normal storage capacity due to limited data. In these cases, we opted to use the maximum observed storage instead of the maximum storage capacity in GRanD as our storage capacity to ensure storage changes were deducible.
At this point, we went through and calculated the percentage missing from the 1980 –2019 period for each region. When aggregating by region, we initially took the average storage value for each region and that yields no missing data. At this point, we have not yet done the individual dam analysis, but we will include the percentage of data that is gap-filled in the next round of edits.
Sen’s slopes are less biased towards end points than linear regressions as the slope is calculated between consecutive points and the final slope value is the median of all these slopes. This is preferred as we are able to see what the median rate of change across all the points are. Secondly, this method allows us to have a confidence interval to determine how likely it is that these slopes are due to random chance. We opted for 10% or 0.1 as our confidence bounds as this is the typical value for most statistical analysis.
We did look at a variety of different of distributions (log-logistic, log-normal, Pearson type III, generalized pareto, and Weibull) when calculating the Standardized Streamflow Indices. Upon analysis and through reading the Vicente-Serrano et al., 2012, we opted to only use the GEV distribution. This was due first to the robustness of GEV explained in the Vicente-Serrano et al., 2012 for multiple hydrologic conditions (both high and low flows) as well as our desire to ensure all the indices had the same distribution fit for congruency.
- The results and discussion are mostly limited to identifying trends in reservoir storage using different metrics. There should be a much deeper focus on what is the implication of these trends, seasonality, and uncertainty in storage for water resources engineers and planners, and elaborate on how the insight from this research helps them to better understand the risks involved with a reservoir storage management and better prepare for the uncertain future.
Thank you for your comment. We agree that more analysis related to the meaning of these results is needed in the discussion section and our generalized response does note some of the updates we will include. To summarize the generalized response, we will focus part of the discussion on either refuting or supporting common assumptions made in reservoir models (i.e., minimum storage capacity is 10% of maximum storage capacity) and provide cases where this might not be entirely accurate as dams due to discrepancies between observed storage values and reported maximum storage capacity values in datasets such as GRanD. In the 100 instances where storage capacity values routinely fall below observed storage values, modelers and water managers may be overestimating the amount of storage in a reservoir at a given time. Additionally, the underestimation of seasonal dynamics in reservoir models also contributes to this underestimation and therefore our seasonal analysis is important.
When looking at implications of reservoir storage trends, drought sensitivity and seasonal trends, the effect of climate and demand cannot be underestimated. When decreasing storage trends are combined with drought sensitivity, we see how shifts in demand or climate have affected these trends. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio. Finally, the operational range trends demonstrate that flood control dominated basins in Tennessee, Ohio and Lower Mississippi have decreased, while California has increased. These trends are potentially due to the impact of climatic shifts which resulted in changes to operational policies.
- Does using the drought index for reservoirs that are designed for flood control provide any valuable insight? I think that it would be better to consider how trends in reservoir storage in humid regions affect the ability of reservoirs to control flooding, especially for extreme events.
We agree that floods are also interesting, however, this would require an event-based analysis which is outside our scope. Additionally, analyzing hydrologic drought can be useful for flood control reservoirs as we do see flood control regions with storage that takes longer to recover than the streamflow index (Ohio, Tennessee, Lower Mississippi and Sourris Red Rainy). Our updated analysis will focus on providing linkages between the drought sensitivity and storage trend analysis to inform potential causes of the declining trends seen in CONUS.
Minor comments:
Thank you for all these minor comments. In responding to this review, we went through and made the necessary grammatical edits stated and directly answer relevant questions. We also plan to dedicate time to polishing the resubmitted article.
Line 28-30: Grammatically incorrect: based ‘on’ various remote sensing …
We have updated this.
Line 32: You are mentioning that you have also used ‘release’ data to study the historical behavior of reservoirs, while in practice, you did not use the release data in your analysis.
Thank you for noting this. We will update this to state storage data and we in fact did not use release data.
Line 38: Grammatically incorrect: There are 2,000 large dams ‘that’ are spread out across the US
We have updated this.
Line 25-38-132: In line 25, large dams are defined as having a capacity larger than 10 km3, while in line 38, the value is equal to 0.1 km3. Also in line 132, the values that defines large dams in 0.01 km3. Which one is the better definition of large dams?
Thank you for pointing this out. The better denotation of large dams is 0.01km^3 based on the standard in GRanD and we have updated this to be the case.
Line 49:50: It is better to not use ‘in fact’ and ‘finally’ in the same sentence. You also mentioned ‘water supply’ in the previous sentence and are repeating it again here.
We will note this and make sure to vary the language within sentences to avoid repetitions.
Line 54: Please explain how dams can decrease the severity of extreme events in one or two sentences. In line 74-75, it is also mentioned that reservoirs can increase the drought duration.
This is a great place to improve our detailed analysis of the drought sensitivity. Preliminary analysis into this suggests that the larger the dams are in a region, the longer it takes for a region to recover. Most specifically, this is seen in the western US where the occurrence of large dams causes hydrologic drought to perpetuate for longer, while on the eastern US hydrologic droughts do not last as long.
Line 63: How have dams changed the run-off regimes? Please explain in more details.
Dams change run off regimes in two ways. First, large dams increase the storage along river systems and decrease the amount of water that flows downstream naturally. Secondly, the occurrence of dams allows water to be released during non-natural periods, such as during low flows, or at different seasons. We will add lines to this section to clarify this point.
Line 65: What type of system? Please be a little clearer.
We have removed the reference to system and instead replaced it with “water supply”.
Line 57: Grammatically incorrect. I think that the word ‘promote’ can be removed.
Thank you. It can and we have updated this locally.
Line 68: Grammatically incorrect. Use ‘have’ instead of ‘has.
We have updated this.
Line 71: Disrupt how?
We will switch the wording to focus on the fact that flashier systems have changed reservoir operations. Most specifically, the eastern US has seen an increase in flooding events which has caused reservoir operators to release more water to keep flood storage low. We will include a brief explanation of this in the resubmitted version.
Line 72-74: The sentence is not clear. Two ‘increase’ in one sentence.
We have updated this locally.
Line 93-94: The sentence is grammatically incorrect.
We have updated this locally.
Line 112: There are two ‘is’ in a sentence.
This is updated locally
Line 114-115: What do you mean by ‘reservoir dynamic’? I think that you can be more precise about it.
Thank you. We agree that we can be more precise. We will edit this language to specifically point out the seasonality of reservoir operations and storage fluctuations in our resubmission.
Line 119: What trend?
The trends referred to here are changes in fraction filled and regional comparisons. We plan to rephrase this sentence in the resubmission to ensure more clear and concise wording.
Line 123-126: Again, you are using general terms such as ‘reservoir patterns’, and ‘regional dynamics’. I think that you should be more specific in defining these terms.
Thank you, we agree that we could be more specific. In addition to updating these instances, we also will go through the paper and remove broad or vague terms to enhance clarity.
Line 150: What do you mean by ‘hydrologic boundaries’?
The term “hydrologic boundaries” in this sentence refers to the watershed boundaries from the Watershed Boundary Dataset. We have updated hydrologic boundaries to read “watershed boundaries”
Line 168-170: How plentiful were the outliers which suggested values higher than the storage capacity? And how much higher were they compared to the capacity?
In our analysis there are approximately 100 dams with instances where storage values were higher than the maximum storage capacity. In most cases, the maximum storage value was off by less than 10 MCM. In some cases (less than 10), the maximum storage value was off by less than 100 MCM. It is very possible that some of these maximum storage values could denote spills, however, in all cases they were not simply point discrepancies and therefore we opted to change the maximum storage value.
Line 170: What was the average period of gap in the data? For instance, did you use the interpolation for a large gap of data, to the extent that interpolation may become irrational?
As we averaged the data for all the dams across the HUC2 units and then linearly interpolated, we did not have large gaps to fill (only one or two days of data). We also opted for the period from 1980 – 2019 as the majority of the dams have a full period of record between these dates. For the proposed individual dam analysis, we may need to evaluate the percentage of data that is gap filled to see if the results from these dams can still be used.
Line 177: I suggest using ‘received little impact’ instead of ‘had little impact’.
We have updated this locally.
Line 187: Please explain the ‘Fraction Filled’ in a few sentences before introducing the formula.
Thank you for mentioning this. We plan to include a brief two sentence introduction to what fraction filled is. We have updated this to read: “The FF timeseries uses the total average storage for a given day in each region in ResOpsUS and divides that storage by the total storage capacity of all the dams in that region on that same day.”
Line 208: Please briefly explain ‘Sen’s slope’.
We agree that a brief explanation of Sen’s slopes is necessary as well as our rational for using them instead of a linear interpolation. We will include a definition and rational of Sen’s slope similar to this:
“Sen’s slopes are the median of all the slopes calculated between consecutive data points and the final slope value is the median of all these slopes. This is preferred as we are able to see what the median rate of change across all the points are and are less biased towards end points.”
Line 224: Considering that you are using daily streamflow values, and not annual maxima, you should consider more than one distribution to find the best fit to the data. Or at least, you must use tests such as ‘Kolmogorov–Smirnov test’ to ensure that the probability distribution is decently representing the data.
As stated previously, we looked at multiple distributions to fit the streamflow data. To ensure that the distribution was the same across all gages and regions, we chose to use the most robust according to Vicente-Serrano et al., 2012 which was GEV. In our revision, we will test our assumption with a Kolmogorov-Smirnov test.
Line 226: SSI values ‘can’ be calculated …
This is updated locally.
Line 232: I think it is better to write the formulation for both conditions of P being larger or smaller than 0.5.
Thank you for your comment. We opted to use the same format as Vicente-Serrano et al., 2012 as these were not equations that we derived and would prefer not to change them.
Line 262: What do you mean by ‘reservoir setting’?
We have updated this to say climate.
Line 266: ‘dominate’
We have updated this locally.
Line 274: Hydroelectricity reservoirs are most common ‘in’ the Tennessee Basin and South Atlantic.
We have updated this locally.
Line 328: ‘ion’?
We have updated this to be “in.”
Line 334: The statement does not seem true. Based on the Figure 2, the lower Colorado river has a very small month-to-month variation.
Thank you for noting this. We should have added Upper Colorado to this sentence and not simply stated Colorado. We have updated this locally.
Line 337: ‘is’ shown in Figure 2.
We have updated this locally.
Line 341:342: This statement does not seem correct based on the Figure 2.
We agree that this sentence is not supported by the figure. We have removed it locally.
350: ‘Note’ that …
We have updated this.
For the figure 2, I think that in addition to the black line that shows the median Fraction Filled and purple shadow that shows maximum minus minimum, you can also add 95% confidence intervals (maybe as dashed lines) to make sure that an extreme outlier (minimum or maximum) does not greatly impact the shadowed purple part.
Thank you for this suggestion. We will look at adding the 95% confidence interval. At the time of submission, we were hesitant to add additional lines to ensure the panel plots were neat.
Line 378: It is mentioned that operational variance provides a more holistic measure compared to the operating range. However, I am not convinced why you had to use both metrics for the analysis. If variance is better, why using operating range? Does each one of them provide a separate vision that justifies using both? Otherwise, it might be better to use only one of them to avoid extra confusion.
In response to this and a previous reviewer comment, we have opted to remove the operational variance plots (Figure 4) and keep the emphasis on operational range. This is done to complement the operational range trends in Figure 7 and to ensure there is room for additional figures we have proposed.
Line 201 and 347: In line 201, it says that region with more than 40% of storage coverage are considered, while line 437 suggest the value as 50%. Which one is correct?
At the time of submission, we were looking at regions that had greater than or equal to 50% storage covered. Upon reviewer comments on including more regions, we have decided to decrease this threshold to 40% storage covered. We have updated both instances to read 40% storage covered locally.
Line 438: Why October? Please explain the reason for choosing this month.
By choosing one month to evaluate the trends over, we could ensure that, and storage trends were due to changes in storage levels and not seasonal storage fluctuations. We chose October as that is the start of the water year and is a key data point for reservoir operators to plan for storage in the coming years. The storage at the start of the operational year (typically in October) is also important for reservoir modelers as this value is used to derive reservoir storage and releases.
Line 514: On both regional and ….’a’ is extra.
We have corrected this locally.
Line 656: 83% ‘of’ regions …
We have corrected this locally.
Line 658-659: The sentence is grammatically incorrect.
We have corrected this locally.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC3
Status: closed
-
RC1: 'Comment on egusphere-2022-1051', Anonymous Referee #1, 15 Nov 2022
This paper presents an analysis of reservoir storage trends across the contiguous United States (CONUS), based on a new dataset of historical storage that authors put together recently (ResOpsUS). I agree with authors that conclusions based on this dataset would be a useful complement and even an upgrade on existing studies of reservoir behaviour that are based on other data sources. I also broadly agree with the main findings. Yet, this study has several weaknesses, detailed below:
- It is not clear what the direct connection is with the special issue on the representation of reservoirs in hydrological models. In reality, conclusions are not related to this topic. Literature on the topic is introduced at times, and the methodological assumptions made when modelling reservoirs are discussed (e.g. lines 104-110) but with never a mention of how this study actually relates to any of these assumptions.
- More generally, the scope of the paper is unclear. Authors (rightly) realise that there are many ways an analysis of this data might contribute to the literature, but they seem to not choose one (or more) focal point(s). As a result, the literature review is the introduction is disjointed, and several themes are broached without a clear focus.
- A major selling point of the paper is the use of new data, but in fact, this is not exploited by the paper. They do not rely on a comprehensive review of existing knowledge on reservoir storage / operations in the CONUS to propose a systematic view of how this dataset completes or updates this existing knowledge (and this would be a low hanging fruit for a useful paper!).
- The analysis has two weaknesses in leveraging the dataset to derive insights. (i) Authors aggregate by regions without ever looking at individual reservoirs. (ii) A consequence of this aggregation is to drop key regions full of reservoir such as the northwestern U.S.. With these choices, it is not clear they take full advantage of the fine grain data offered by their new dataset.
- As a consequence of unclear scope and methodological choices, findings are all unsurprising and it is never clear how they compare in relation with the literature. Besides, many of the findings on drying at the national level seem to be related to drying in one particular region: the southwestern U.S and in particular the Colorado river basin. It would be great for authors to quantify this contribution from a particular region to national conclusions, or abstain to make these national conclusions.
- The resilience angle is interesting, but the concept is not clarified until it is clarified as a recovery metric, and no definition is proposed until the discussion (lines 602-605). If authors want to use their dataset to look at the drought recovery ability throughout the U.S., that is potentially a great idea (if they can show there is a gap in the literature there), but I’d suggest they leverage the dataset by looking at the data from individual reservoirs.
- Last but not least, the writing is unequal (see e.g. the abstract) and suggests a paper hastily put together. This impression is reinforced by the unclear scope and disjointed references to the literature (e.g., resilience definitions proposed in the discussion).
For these reasons, my take on this work is that it is probably not ready for publication at this stage. There is clear potential though, and I’d like to suggest than rather than trying to push preliminary work through, authors have much to gain by electing to do a deeper analysis before publishing their findings, probably over several excellent papers. I am also not sure they need to focus on this special issue until and unless they can use the data to address some of the assumptions made in representing reservoirs in hydrological models.
A few detailed comments (given the stage the paper is at I do not provide this for the whole text):
The text would benefit from a rigorous round of edits. Just in the abstract I found a series of imprecisions that would have been fixed by proofreading:
Line 7: as soon as they introduce the phrase, authors need to explain what “reservoir resilience” means in their context
Line 13: “dominated” instead of “dominate”
Line 15: “reservoir storage has decreased”, is it average annual storage?
“our national fraction filled decreases are 50% less than those shown previously”, please rephrase to make this clearer to the reader.
Lines 16-17: the definition of operational range is great to see but should occur the first time the term is introduced.
The idea that the lack of data has been a limiting factor (lines 8 and 27) deserves clarification: what do you see thanks to the new dataset that studies with less data could not?
Line 38: “there are large dams are spread out”
Line 39: provide reference for the standard.
Line 100: repetition of the idea expressed in lines 28-30.
Lines 104-110: is the aside on reservoirs in hydrological models part of the paper’s scope?
Methods: this is a collection of indicators. Many of them are interesting to look at, but it is not clear how they complement each other, and what question(s) they are meant to answer when taken together.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC1 -
AC1: 'Reply on RC1', Jen Steyaert, 11 Jan 2023
This paper presents an analysis of reservoir storage trends across the contiguous United States (CONUS), based on a new dataset of historical storage that authors put together recently (ResOpsUS). I agree with authors that conclusions based on this dataset would be a useful complement and even an upgrade on existing studies of reservoir behaviour that are based on other data sources. I also broadly agree with the main findings. Yet, this study has several weaknesses, detailed below:
Thank you for the thoughtful review and for acknowledging the usefulness of our work. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally must rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we can provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region has an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
- It is not clear what the direct connection is with the special issue on the representation of reservoirs in hydrological models. In reality, conclusions are not related to this topic. Literature on the topic is introduced at times, and the methodological assumptions made when modelling reservoirs are discussed (e.g. lines 104-110) but with never a mention of how this study actually relates to any of these assumptions.
We agree that we were not explicit enough in our original manuscript regarding its connection with the special issue. We have outlined our relevance in our overview above as well as our plan to make this clearer moving forward. As stated above, our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally must rely on large scale assumptions of how reservoirs operate. Specifically, our findings support assume minimum storage is approximately 10% of the maximum storage capacity for all but the 100 reservoirs that have storage capacity values in GRanD less than observed maximum storage in ResOpsUS. Additionally, our seasonal analysis and storage trends are both useful to model calibration. Finally, this regional based analysis may be useful for developing reservoir operations and dynamics in areas outside the US that have similar characteristics such as aridity, reservoir uses, and population that are data limited.
We also agree that updating the document to provide an example of why this study is important is necessary to rounding out the introduction. To make this connection clearer, we will edit lines 124 – 127 to include the dataset description. Next, we also will also add a paragraph at the end of the introduction to include why this analysis is important based on the reservoir assumptions on lines 104 – 110). Lastly, we plan to weave in the narrative regarding assumptions and modelled rule curves into our analysis.
- More generally, the scope of the paper is unclear. Authors (rightly) realise that there are many ways an analysis of this data might contribute to the literature, but they seem to not choose one (or more) focal point(s). As a result, the literature review is the introduction is disjointed, and several themes are broached without a clear focus.
Thank you for your comment. To complement our detailed response above, we plan to point out two main questions and focus our discussion and conclusion on analyzing the assumptions (stated above) made by modelling studies. The two main questions are as follows:
- What is really happening with reservoirs both seasonally and over time and how does this compare with the assumptions previously made in large scale modeling studies?
- How have reservoir operations evolved over time and do we see changes in the response to droughts?
To ensure cohesion with these questions, we will focus on 1) what really is happening with reservoirs both seasonally and over time during the 1980 – 2019 period, 2) what drought sensitivity looks like to hydrological drought during the period from 1980 – 2019 and 3) are the assumptions in reservoir models accurate. Additionally, our discussion and conclusion will be reorganized to discuss the ways in which our findings support or refute the common assumptions in reservoir models as stated above. Most importantly, this restructuring will focus on the following results mentioned in the table below.
- A major selling point of the paper is the use of new data, but in fact, this is not exploited by the paper. They do not rely on a comprehensive review of existing knowledge on reservoir storage / operations in the CONUS to propose a systematic view of how this dataset completes or updates this existing knowledge (and this would be a low hanging fruit for a useful paper!).
Thank you for this comment. As stated in our detailed response above, we will expand our results and discussion to directly point to the ways in which we are increasing understanding of reservoir operations through time. As stated above, previous models typically use optimized rule curves that are based on assumptions regarding how reservoirs work, but not necessarily on the historical data. Therefore, we will focus our results and discussion to show ways in which these assumptions may be accurate or inaccurate depending on the data. Additionally, we will note the select ways in which our results either compliments or correct analyses.
Some of the key points we will highlight include the fact that historical studies have shown that reservoir storage in CONUS has decreased slightly with most studies stating estimates between 0 – 15% regionally. In contrast, our analysis demonstrates that fraction filled has only decreased by 0.2% nationally and the Lower Colorado (with the sharpest slope) has a decrease of only 1.2%. Historical studies have shown that reservoirs increase drought severity which our findings align with in all cases except for the South Atlantic region. Our drought sensitivity analysis is also useful for determining if climate or changes in demand could be behind the storage changes, we observe across the US. Finally, the novelty of our operational range analysis is not to be underestimated as the addition of more reservoirs, operators and temporal scale allows us to evaluate changes not only regionally, but also monthly. This is important for calibrating hydrologic models, but also to show where reservoir operations are widening or tightening in response to demand and/or climatic shifts.
- The analysis has two weaknesses in leveraging the dataset to derive insights. (i) Authors aggregate by regions without ever looking at individual reservoirs. (ii) A consequence of this aggregation is to drop key regions full of reservoir such as the northwestern U.S.. With these choices, it is not clear they take full advantage of the fine grain data offered by their new dataset.
We agree that we can do more to leverage the uniqueness of ResOpsUS and in response to this comment we will add two figures focused on the individual dams. The first will be a map with point locations of all 678 reservoirs in ResOpsUS colored by if their storage trend is positive or negative over the 40-year period we are looking at. We will also cut Figure 4 and provide a two-panel map with the point location of each dam colored by the month of peak fraction filled (panel a) and the month of largest operational range (panel b). In addition to including the individual dam plots as stated above, we also increased our threshold for storage covered to be 40% which allows us to include HUC7 and HUC17 in our analysis. We now have 14 regions we are looking at out of the 18 total HUC2s.
- As a consequence of unclear scope and methodological choices, findings are all unsurprising and it is never clear how they compare in relation with the literature. Besides, many of the findings on drying at the national level seem to be related to drying in one particular region: the southwestern U.S and in particular the Colorado river basin. It would be great for authors to quantify this contribution from a particular region to national conclusions, or abstain to make these national conclusions.
Thank you for your comment. In response to this and other comments, we have provided a table denoting specific areas where our results align or refute previous work below. Additionally, we would like to note that while our findings align with current and past research, the novelty of ResOpsUS goes beyond previous work and allows us to provide greater insights into how reservoirs are operated. That said, we do note that greater emphasis on the impact of regional and local changes to the trends at the national level will enhance our argument. We plan to incorporate more discussion on this in our drying section.
Additionally, we also wanted to highlight some of the key results and the importance of the operational range, storage trends, and drought sensitivity analyses. First, these operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. That said, the maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). In fact, we observe tighter operations in Tennessee, Ohio and South Atlantic, while California and Sourris Red Rainy have wider operations.
To supplement the operational range analyses, we also want to highlight the fraction filled changes over time on the three spatial scales we looked at: CONUS wide, regional, and individual reservoirs. These three scales are important as the first two (CONUS wide and regional) allow us to directly compare to historical findings. Additionally, these findings tell us how reservoirs have historically been operated and about their drought sensitivity as reservoirs with increases in fraction filled have shifted their operations to increase fraction filled levels, while reservoirs with decreasing fraction filled trends are potentially facing unsustainable water demand that operations cannot keep up with. Specifically, we note that our regional analysis aligns with previous work in the western United States, however, it contradicts previous studies as storage in the eastern United States has also declined with the exception of the Tennessee, and Sourris Red Rainy basins
Lastly, we would like to highlight our regional drought sensitivity analysis. This compliments the fraction filled trends and allows us to provide potential reasons for the fraction filled trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high fraction filled ratios that denote fraction filled takes much longer to recovery than streamflow. This suggests that the decreasing fraction filled trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (fraction filled takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this recovery ratio.
- The resilience angle is interesting, but the concept is not clarified until it is clarified as a recovery metric, and no definition is proposed until the discussion (lines 602-605). If authors want to use their dataset to look at the drought recovery ability throughout the U.S., that is potentially a great idea (if they can show there is a gap in the literature there), but I’d suggest they leverage the dataset by looking at the data from individual reservoirs.
As stated above, we have opted to remove the resilience section and simply focus on the drought sensitivity. We believe that this will be a better focal point for the analysis as it complements the current work done by Turner et al., 202 regarding data driven reservoir operations. Most specifically, this analysis demonstrates regions where reservoir operations are sensitive to drought. This is important for modelling reservoir operations as Turner et al., 2020 showed that more drought sensitivity dams should undergo rigorous sensitivity and uncertainty analyses.
- Last but not least, the writing is unequal (see e.g. the abstract) and suggests a paper hastily put together. This impression is reinforced by the unclear scope and disjointed references to the literature (e.g., resilience definitions proposed in the discussion).
Without more details we are not sure exactly what the reviewer is pointing to with respect to this comment. However, in response to this comment, we have already started correcting the organization of our paper and will continue to polish the paper per the individual edits below. In addition, we are planning to do an extensive read through to make sure language is consistent, grammatical errors are corrected, and the paper is logically organized.
For these reasons, my take on this work is that it is probably not ready for publication at this stage. There is clear potential though, and I’d like to suggest than rather than trying to push preliminary work through, authors have much to gain by electing to do a deeper analysis before publishing their findings, probably over several excellent papers. I am also not sure they need to focus on this special issue until and unless they can use the data to address some of the assumptions made in representing reservoirs in hydrological models.
Thank you for this comment. We greatly appreciate the insights you have provided and feel that the paper will be greatly enhanced by these additions. We also strongly believe that these results will be beneficial to the larger reservoir community as stated in our general response above, reviewer comments 1 and 5, and the table below.
A few detailed comments (given the stage the paper is at I do not provide this for the whole text):
The text would benefit from a rigorous round of edits. Just in the abstract I found a series of imprecisions that would have been fixed by proofreading:
Line 7: as soon as they introduce the phrase, authors need to explain what “reservoir resilience” means in their context
Thank you for this comment. We will remove the reference to reservoir resilience as we do not believe the analysis at this point allows us to directly speak on resilience. Rather we will update this to be drought sensitivity.
Line 13: “dominated” instead of “dominate”
We have corrected this mistake.
Line 15: “reservoir storage has decreased”, is it average annual storage?
You are correct, it is average annual fraction filled. We have added that to the beginning of the comment.
“our national fraction filled decreases are 50% less than those shown previously”, please rephrase to make this clearer to the reader.
We have corrected this to say “a much smaller decrease than previous work” line 16.
Lines 16-17: the definition of operational range is great to see but should occur the first time the term is introduced.
Thank you for this comment. We have moved the description of operation range to first instance of the phrase “operational range” on line 12.
The idea that the lack of data has been a limiting factor (lines 8 and 27) deserves clarification: what do you see thanks to the new dataset that studies with less data could not?
To more clearly highlight the novelty and importance of this work, we have created Table 1 below to directly show how our results agree or disagree with previous work. Mainly, we see that our analysis supports that storage has declined across CONUS, similar to previous work, yet the magnitude of that decrease is different. Additionally, the regional differences are supported in the western US, but not in the southeastern US as Ohio and Lower Mississippi also have declines. Previous studies only looking at storage capacity have stated that the Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 – 2000, yet we see there are decreasing storage trends in reservoirs along this period. Lastly, when looking at operational targets in Patterson et al., 2018, California dams spend more time under target and eastern dams in wetter regions spend more time above operational targets. Our operational trend analysis suggests the opposite.
Line 38: “there are large dams are spread out”
We have rephrased this to be clearer and more direct.
Line 39: provide reference for the standard.
We have added the reference to iCOLD in the bibliography
Line 100: repetition of the idea expressed in lines 28-30.
We have rephrased this section.
Lines 104-110: is the aside on reservoirs in hydrological models part of the paper’s scope?
We plan to include one to two paragraphs discussing the assumptions made in hydrologic models and how our work compliments it. We added a more detailed response to this under comment 1.
Methods: this is a collection of indicators. Many of them are interesting to look at, but it is not clear how they complement each other, and what question(s) they are meant to answer when taken together.
Thank you for your note. We believe that the organization we propose in comment 2 combined with more direct connections to the assumptions in reservoir models will pull this paper together. We also believe this will enhance our story and provide valuable validation points for hydrologic modelers.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC1
-
RC2: 'Comment on egusphere-2022-1051', Anonymous Referee #2, 02 Dec 2022
The authors have assembled a new data set on reservoir operations that has great potential to generate new insights. They analyze this dataset to explore seasonal and regional patterns as well as long term trends in reservoir operations. Specifically, they focus on the fraction filled, operational range and variance, and drought recovery in their analysis. Overall the paper documents an exploratory analysis which demonstrates some of the capabilities of the new data set. Unfortunately, this analysis stays at the exploratory level, is not clearly motivated by a research question or problem statement, and does not generate new insights or hypotheses. There is great potential here for the authors to use apply this data set to uncover new insights at the intersection of climatology, hydrology and reservoir operations. To do this would require either major revisions or resubmitting later as a new manuscript. To clarify these general reflections, I have provided several specific comments below.
Comments
- The introduction motivates the introduction of a new data set. However, it does not motivate the subsequent analysis. Because of this the reasoning behind subsequent analytical choices such as inclusion/exclusion criteria, aggregation choices, and metrics selection is unclear, and the reader is left without a clear benchmark by which to judge these choices. The authors can address this by posing one or more research questions and rewriting the introduction to motivate these questions and the approach taken to address them. Note that this change might require not only a rewriting of the introduction but the selection of different analysis and methods if warranted by the selected question(s).
- There is a great deal of diversity in both the physical design reservoirs and the design of their operational rules across the U.S. This diversity is driven by reservoir function(s), local seasonal and interannual streamflow patterns, and daily and seasonal demand patterns, among other factors. Therefore, I am concerned that what is observed when aggregated at the regional level is really the average of different signals, from different functions and operating priorities, and that this average signal has lost much of the interesting information. Further, per my first comment, it is not clear what the motivation for aggregating streamflow and reservoir storage by hydrologic regions.
- Across each region the size of reservoirs both in absolute volume and in size relative to streamflow, streamflow variability and demand, vary substantially. Pooling all reservoirs in each region based on capacity and filled volume heavily weights the findings to the largest reservoirs in each region. This could be warranted, depending on the research question, but again the motivation for this choice is not clear.
- Figure 5 presents drought conditions over the whole study area and the authors the discuss correlations between drought occurrence and severity, and changing in storage. On line 426 the authors mention that decreases in the interannual fraction filled are correlated with climatic shifts. Are they correlated in a mathematical sense or just in an approximate visual sense? If it is the later, please rephrase to avoid misinterpretation. Further, drought conditions vary so broadly over CONUS that I am unsure how this is meaningfully related to reservoir operations.
- Assessing resilience requires defining what function is to be recovered or maintained. Following from this, metrics serving to measure resilience should capture the function or functions of interest. The authors apply a drought recovery metric to assess reliance across regions where reservoirs have different dominant functions, notably flood control. In flood control dominated regions, assessing the resilience via a drought recovery metric is not meaningful.
- The discussion and conclusions do not make a clear case for what this manuscript adds to the existing science. The discussion and conclusions should respond to the motivation and aims presented in the introduction, clearly articulate what was learned and why it matters and put the findings in the context of prior work. The authors note on a number of occasions that this analysis confirms prior findings but do not make the case for the value of their own analysis.
Minor Comments
- Line 264, Figure c should read Figure 1c
- In the paragraph beginning on line 378, the authors reference both Figures 4 and 5. However, the reference to Figure 5 do not make sense and I believe all references here should be to Figure 4.
- Please clarify how monthly variance is calculated for Figure 4. Does the daily fraction filled mentioned in the caption refer to the average of all reservoirs in the region?
- To determine the drought severity shown in Figure 5, what spatial extent was used?
- On line 499 the authors state that the “second set are regions that have predominately positive trends and greater than or equal to three statistically significant trends (Souris Red Rainy, California and Upper Colorado).” However, Figure 7 shows that the Upper Colorado has predominately negative trends. Is the figure wrong or is the text?
- On line 524 the authors state that recovery times were capped at 50 months because only five regions had recovery times greater than 50 months. Given that there are only 11 regions in the study, this is not a reasonable justification. Please revise or further justify this choice.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC2 -
AC2: 'Reply on RC2', Jen Steyaert, 11 Jan 2023
Reviewer 2:
The authors have assembled a new data set on reservoir operations that has great potential to generate new insights. They analyze this dataset to explore seasonal and regional patterns as well as long term trends in reservoir operations. Specifically, they focus on the fraction filled, operational range and variance, and drought recovery in their analysis. Overall the paper documents an exploratory analysis which demonstrates some of the capabilities of the new data set. Unfortunately, this analysis stays at the exploratory level, is not clearly motivated by a research question or problem statement, and does not generate new insights or hypotheses. There is great potential here for the authors to use apply this data set to uncover new insights at the intersection of climatology, hydrology and reservoir operations. To do this would require either major revisions or resubmitting later as a new manuscript. To clarify these general reflections, I have provided several specific comments below.
Thank you for your comprehensive review. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally have to rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all of our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we are able to provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region have an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
Comments
- The introduction motivates the introduction of a new data set. However, it does not motivate the subsequent analysis. Because of this the reasoning behind subsequent analytical choices such as inclusion/exclusion criteria, aggregation choices, and metrics selection is unclear, and the reader is left without a clear benchmark by which to judge these choices. The authors can address this by posing one or more research questions and rewriting the introduction to motivate these questions and the approach taken to address them. Note that this change might require not only a rewriting of the introduction but the selection of different analysis and methods if warranted by the selected question(s).
Thank you for your comment. We do agree that the paper could be better motivated in the introduction. We also agree that our questions could be better described to make reading the paper easier. We have therefore opted to focus our introduction around the following questions:
- What is really happening with reservoirs both seasonally and over time and how does this compare with the assumptions previously made in large scale modeling studies?
- How have reservoir operations evolved over time and do we see changes in the response to droughts?
- There is a great deal of diversity in both the physical design reservoirs and the design of their operational rules across the U.S. This diversity is driven by reservoir function(s), local seasonal and interannual streamflow patterns, and daily and seasonal demand patterns, among other factors. Therefore, I am concerned that what is observed when aggregated at the regional level is really the average of different signals, from different functions and operating priorities, and that this average signal has lost much of the interesting information. Further, per my first comment, it is not clear what the motivation for aggregating streamflow and reservoir storage by hydrologic regions.
We initially aggregated the data so we could make comments about regional dynamics. Upon this review, we recognize the importance of an individual dam analysis, therefore we have opted to include two extra figures focused solely on individual dynamics. The first will be a map with point locations of all 678 reservoirs in ResOpsUS colored by if their fraction filled trend is positive or negative over the 40-year period we are looking at. We have also opted to cut figure 4 and provide a two-panel map with the point location of each dam colored by the month of peak fraction filled (panel a) and the month of largest operational range (panel b). In addition to including the individual dam plots as stated above, we also increased our threshold for fraction filled covered to be 40% which allows us to include HUC7 and HUC17 in our analysis. We now have 14 regions we are looking at out of the 18 total HUC2s.
- Across each region the size of reservoirs both in absolute volume and in size relative to streamflow, streamflow variability and demand, vary substantially. Pooling all reservoirs in each region based on capacity and filled volume heavily weights the findings to the largest reservoirs in each region. This could be warranted, depending on the research question, but again the motivation for this choice is not clear.
Aggregating the data by region, does cause the region to be weighted by the largest dam. Since regional waterways will be impacted by all the dams in a given region and weighted accordingly, we opted to aggregate the analysis by region. Based on our above questions, aggregating this data by region also allows us to compare with previous studies and note areas where large scale modelling study assumptions may not align with the regional results. That said, we do recognize that some of the regional dynamics seen are more indicative of the largest dam in a region. We will include two more regions and add two more figures focusing on the individual dams.
Additionally, the two questions outlined above in comment two will allow us to focus our analysis. Specifically, we will be able to refute or support modelling assumptions such as minimum storage equals 10% of maximum storage capacity and provide important trends for model calibration. To support the second question, we will provide more discussion regarding how our analyses fits in with previous work as shown in the general response and the table below.
- Figure 5 presents drought conditions over the whole study area and the authors the discuss correlations between drought occurrence and severity, and changing in storage. On line 426 the authors mention that decreases in the interannual fraction filled are correlated with climatic shifts. Are they correlated in a mathematical sense or just in an approximate visual sense? If it is the later, please rephrase to avoid misinterpretation. Further, drought conditions vary so broadly over CONUS that I am unsure how this is meaningfully related to reservoir operations.
The correlations are in an approximate visual sense. Since drought conditions and the reservoir uses vary across the entirety of CONUS, we opted to aggregate both for a more approximate analysis. This is also why we analyzed drought impacts regionally in Figure 6 and Section 3.3 so we could have a more quantitative analysis of drought impacts and sensitivity on storage. This figure is building the story of how storage has changed in CONUS over time and gives drought severity as one potential impact.
- Assessing resilience requires defining what function is to be recovered or maintained. Following from this, metrics serving to measure resilience should capture the function or functions of interest. The authors apply a drought recovery metric to assess reliance across regions where reservoirs have different dominant functions, notably flood control. In flood control dominated regions, assessing the resilience via a drought recovery metric is not meaningful.
Thank you for this comment. We agree that using a hydrologic drought recovery metric may not necessarily be accurate for addressing resilience in flood control reservoirs. TO remedy this, we have opted to remove the word resilience and focus on drought sensitivity. We still think that hydrologic drought sensitivity is important to flood control reservoirs as variations in streamflow impact their operational patterns.
- The discussion and conclusions do not make a clear case for what this manuscript adds to the existing science. The discussion and conclusions should respond to the motivation and aims presented in the introduction, clearly articulate what was learned and why it matters and put the findings in the context of prior work. The authors note on a number of occasions that this analysis confirms prior findings but do not make the case for the value of their own analysis.
Thank you for noting our results confirm previous results. We have provided a general response to the lack of novelty in our paper in the general comment above and noted key points in the below table where our work refutes or supports previous work. To summarize these two responses, our study, using observed storage from ResOpsUS, finds that the magnitude of storage decrease in CONUS is much smaller than previous studies have shown. Additionally, our operational range analysis points to widening of operational ranges in California, potentially to increase water supply, and decreasing of operational ranges in Tennessee, Ohio, and the South Atlantic. These results are in direct opposition to the analysis by Patterson et al., 2018 and the difference is potentially due to the higher number of reservoirs in our study. Our regional drought sensitivity also shows that only the South Atlantic has a positive recovery ratio and therefore has storage return to normal before a drought is over. Lower Mississippi and California have large storage ratios which denote that storage takes much longer to recover than streamflow. This drought sensitivity analysis allows us to more deeply analyze the potential causes of the storage trends in CONUS and provides a unique perspective on how observed reservoir operations have responded to droughts in the past forty years.
Minor Comments
- Line 264, Figure c should read Figure 1c
We agree and will fix this in the text.
- In the paragraph beginning on line 378, the authors reference both Figures 4 and 5. However, the reference to Figure 5 do not make sense and I believe all references here should be to Figure 4.
Thank you for your comment. Yes, all the references to figure 4 should be figure 5.
- Please clarify how monthly variance is calculated for Figure 4. Does the daily fraction filled mentioned in the caption refer to the average of all reservoirs in the region?
Monthly variance was calculated using the daily fraction filled values for each region. These daily fractions filled values are the average fraction filled values for all reservoirs in the region. We have opted to remove this figure from our analysis and will also remove the description of the results here.
- To determine the drought severity shown in Figure 5, what spatial extent was used?
To determine the drought severity, we aggregated the three-month SPI values from NCAR across the entirety of the CONUS domain. The NCAR SPI values do not have a stated spatial resolution in their metadata page. More information can be found on the NCAR page: https://climatedataguide.ucar.edu/climate-data/standardized-precipitation-index-spi
- On line 499 the authors state that the “second set are regions that have predominately positive trends and greater than or equal to three statistically significant trends (Souris Red Rainy, California and Upper Colorado).” However, Figure 7 shows that the Upper Colorado has predominately negative trends. Is the figure wrong or is the text?
We agree that this statement is does not match Figure 7. We will remove Upper Colorado from this grouping and add it to the same grouping as Arkansas White Red as those dynamics are much more similar.
- On line 524 the authors state that recovery times were capped at 50 months because only five regions had recovery times greater than 50 months. Given that there are only 11 regions in the study, this is not a reasonable justification. Please revise or further justify this choice.
Recovery times in the analysis were not capped at 50 months, rather we capped the figure legend at 50 months so that the variations in the smaller numbers could be seen. We have elected to change this wording to make this clearer.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC2
-
RC3: 'Comment on egusphere-2022-1051', Anonymous Referee #3, 08 Dec 2022
This paper presents a study of historical trends in reservoir storage on a national scale using a new dataset that provides historical operational data for reservoirs across the US. The seasonal and annual behavior of storage reservoirs across different regions were considered throughout the paper. While the new dataset, accompanied by the analysis of this research offers valuable insight into the subject, I still think that there are major issues that must be addressed before this paper is ready to be published. These issues are as follows:
- There exist many grammatical and dictation mistakes which suggest that the paper is rushed, and not enough time has gone into the revision and polishing process. In addition, there are many general terms used in the text that should be explained in more details. Some of these mistakes and general terms are mentioned in the minor comments.
- Considering that a major part of the analysis is statistical, I think that preprocessing of data should be explained in more details. For instance, how outliers were chosen, and how plentiful they were. What is the percentage of missing data that is calculated by linear interpolation? Or explaining Sen’s slope in more details and explaining why 0.1 value was chosen to signify a significant trend? In addition, you may consider using a larger number of distributions rather than GEV. Some of the points are also mentioned by line in minor comments.
- The results and discussion are mostly limited to identifying trends in reservoir storage using different metrics. There should be a much deeper focus on what is the implication of these trends, seasonality, and uncertainty in storage for water resources engineers and planners, and elaborate on how the insight from this research helps them to better understand the risks involved with a reservoir storage management and better prepare for the uncertain future.
- Does using the drought index for reservoirs that are designed for flood control provide any valuable insight? I think that it would be better to consider how trends in reservoir storage in humid regions affect the ability of reservoirs to control flooding, especially for extreme events.
Minor comments:
Line 28-30: Grammatically incorrect: based ‘on’ various remote sensing …
Line 32: You are mentioning that you have also used ‘release’ data to study the historical behavior of reservoirs, while in practice, you did not use the release data in your analysis.
Line 38: Grammatically incorrect: There are 2,000 large dams ‘that’ are spread out across the US
Line 25-38-132: In line 25, large dams are defined as having a capacity larger than 10 km3, while in line 38, the value is equal to 0.1 km3. Also in line 132, the values that defines large dams in 0.01 km3. Which one is the better definition of large dams?
Line 49:50: It is better to not use ‘in fact’ and ‘finally’ in the same sentence. You also mentioned ‘water supply’ in the previous sentence and are repeating it again here.
Line 54: Please explain how dams can decrease the severity of extreme events in one or two sentences. In line 74-75, it is also mentioned that reservoirs can increase the drought duration.
Line 63: How have dams changed the run-off regimes? Please explain in more details.
Line 65: What type of system? Please be a little clearer.
Line 57: Grammatically incorrect. I think that the word ‘promote’ can be removed.
Line 68: Grammatically incorrect. Use ‘have’ instead of ‘has.
Line 71: Disrupt how?
Line 72-74: The sentence is not clear. Two ‘increase’ in one sentence.
Line 93-94: The sentence is grammatically incorrect.
Line 112: There are two ‘is’ in a sentence.
Line 114-115: What do you mean by ‘reservoir dynamic’? I think that you can be more precise about it.
Line 119: What trend?
Line 123-126: Again, you are using general terms such as ‘reservoir patterns’, and ‘regional dynamics’. I think that you should be more specific in defining these terms.
Line 150: What do you mean by ‘hydrologic boundaries’?
Line 168-170: How plentiful were the outliers which suggested values higher than the storage capacity? And how much higher were they compared to the capacity?
Line 170: What was the average period of gap in the data? For instance, did you use the interpolation for a large gap of data, to the extent that interpolation may become irrational?
Line 177: I suggest using ‘received little impact’ instead of ‘had little impact’.
Line 187: Please explain the ‘Fraction Filled’ in a few sentences before introducing the formula.
Line 208: Please briefly explain ‘Sen’s slope’.
Line 224: Considering that you are using daily streamflow values, and not annual maxima, you should consider more than one distribution to find the best fit to the data. Or at least, you must use tests such as ‘Kolmogorov–Smirnov test’ to ensure that the probability distribution is decently representing the data.
Line 226: SSI values ‘can’ be calculated …
Line 232: I think it is better to write the formulation for both conditions of P being larger or smaller than 0.5.
Line 262: What do you mean by ‘reservoir setting’?
Line 266: ‘dominate’
Line 274: Hydroelectricity reservoirs are most common ‘in’ the Tennessee Basin and South Atlantic.
Line 328: ‘ion’?
Line 334: The statement does not seem true. Based on the Figure 2, the lower Colorado river has a very small month-to-month variation.
Line 337: ‘is’ shown in Figure 2.
Line 341:342: This statement does not seem correct based on the Figure 2.
350: ‘Note’ that …
For the figure 2, I think that in addition to the black line that shows the median Fraction Filled and purple shadow that shows maximum minus minimum, you can also add 95% confidence intervals (maybe as dashed lines) to make sure that an extreme outlier (minimum or maximum) does not greatly impact the shadowed purple part.
Line 378: It is mentioned that operational variance provides a more holistic measure compared to the operating range. However, I am not convinced why you had to use both metrics for the analysis. If variance is better, why using operating range? Does each one of them provide a separate vision that justifies using both? Otherwise, it might be better to use only one of them to avoid extra confusion.
Line 201 and 347: In line 201, it says that region with more than 40% of storage coverage are considered, while line 437 suggest the value as 50%. Which one is correct?
Line 438: Why October? Please explain the reason for choosing this month.
Line 514: On both regional and ….’a’ is extra.
Line 656: 83% ‘of’ regions …
Line 658-659: The sentence is grammatically incorrect.
Citation: https://doi.org/10.5194/egusphere-2022-1051-RC3 -
AC3: 'Reply on RC3', Jen Steyaert, 11 Jan 2023
Reviewer 3:
This paper presents a study of historical trends in reservoir storage on a national scale using a new dataset that provides historical operational data for reservoirs across the US. The seasonal and annual behavior of storage reservoirs across different regions were considered throughout the paper. While the new dataset, accompanied by the analysis of this research offers valuable insight into the subject, I still think that there are major issues that must be addressed before this paper is ready to be published. These issues are as follows:
Thank you for your comprehensive and thoughtful review. We have provided a detailed response to all of your comments below. In addition, we noted several high-level concerns that were brought up by all reviewers and would like to first provide a high-level summary of the main concerns that were brought up and our response here:
High level concerns noted by all reviewers:
- Not relevant to the special issues
While our paper is not directly focused on modelling reservoirs, we still believe that we are relevant to the special issue: Representation of water infrastructure in large scale hydrologic and Earth system models requires numerical approaches for simulation and data to support these approaches. Our analysis provides insights into actual reservoir behavior using the most expansive set of direct observations ever assembled for the US. This work can and should be used to evaluate and improve the representation of reservoirs in large scale hydrologic models which generally have to rely on large scale assumptions of how reservoirs operate.
For example, many reservoir models assume that reservoir storage cannot fall below 10% of the total storage capacity values. Our analysis support this as most regions or reservoirs have storage that does not fall below this 10% threshold. Additionally, our findings show that the 10% threshold would underestimate the lowest storage point in the 100 dams which had maximum observed storage greater than the value in GRanD. The Hanasaki et al., 2006 model (which is widely used in global hydrologic models) assumes that the total release per year depends on the starting storage of that give year. Due to the use of maximum storage capacity values, modelled rule curves typically overestimate the amount of storage in reservoirs, while also underestimating seasonal dynamics when focusing on release-based dynamics. Therefore, dynamic zoning of reservoirs may be necessary for multipurpose reservoirs or those with large interannual storage.
Similarly, understanding seasonal trends (something that has not been explored previously with direct observations at this scale) is important for modelers to ensure that regional dynamics are changing at the same or at similar rates across the 40-year period in our study. This can be especially important for model calibration as our results show significant regional variations in seasonal operating ranges. In basins where reservoir storage is more cyclical, like the Missouri basin, the Hanasaki et al., 2006 assumption may hold. In regions where storage is either increasing (Tennessee) or decreasing (Lower Mississippi, Upper Colorado, Lower Colorado, and California) this assumption may not necessarily hold as demand is also changing over time. Our seasonal scale analysis (Section 2) can be used to calibrate large scale reservoir models to make sure that the seasonal dynamics are aligning. This seasonal analysis is also quite useful for determining which reservoirs and regions have interannual storage and may therefore require more complex zoning or rule curves.
We would also like to highlight that in addition to comparing to the dynamics we present in this paper directly, we have already published the dataset and are publishing all of our analysis code with this work to facilitate custom analysis by modelers seeking to use this analysis to understand their model performance.
- Lacked novel findings
While the general reservoir behavior we demonstrate here does align with previous research. The scope of what we are able to provide, and the number of reservoirs included goes beyond previous work enabling us to provide much greater insights into actual reservoir operations than were previously possible at the national scale. Specifically, we would like to highlight the following key results which are not available in the existing literature:
- We quantify the month each reservoir and region have an observed storage maximum (typically winter for eastern dams and summer for western) and minimum (typically in the early summer for eastern and winter for western). This metric provides a key point of evaluation for modelling studies as it ensures that reservoir storage peaks and minimums are historically accurate to the monthly medians.
- We also determined the operational range (Figure 3 and Figure 7) for each region and over 600 dams spread throughout CONUS. Most modelled rule curves use operational range to determine operations, yet only Patterson et al., 2018 has looked at deviations from operational targets for 233 dams mainly along the Mississippi River. These operational range analyses are useful not only to deriving generic regional reservoir operations but can also be used as a key calibration tool to ensure that modelled operations align with shifts in operational range. Maximum and minimum operational ranges are important to understanding when reservoirs are storing water (increases in operational range) and releasing water (decreases in operational range). The trends are important for determining how those operational ranges have changed over time, specifically if operations have gotten tighter or wider in select regions.
- In fact, Patterson et al., 2018, showed that the Mississippi Valley Army Corps districts had a large portion of reservoirs that spent a portion of the year above operational targets since 1970. Conversely our operational trend analysis shows that eastern reservoirs have more decreasing operational range trends and therefore these reservoirs should be sitting under or closer to operational targets.
- Interestingly, our findings with respect to operational range show decreases across all almost all months in Tennessee, Ohio and South Atlantic regions which are mainly flood control dominated. These results refute the findings in Patterson et al., 2018 as their findings showed that flood control reservoirs along the Mississippi spent a large portion of the year above operational targets. Our findings also differ in the South Pacific region, where Patterson et al., 2018 observed reservoirs sitting below the operational target for most of the year. Our regional analysis, however, showed that only Lower Colorado has three significant decreasing operational range trends in June, July and October, while Upper Colorado had no significant trends and California had mainly increasing operational range trends.
- Trends in reservoir storage show that there are differences in behavior locally and regionally that may not be captured in modelled assumptions. Specifically, we find that storage is decreasing across the majority of CONUS (the sharpest declines are in the Lower Colorado basin) with the exception of the Sourris Red Rainy and Tennessee basins. This contradicts previous research that showed increase water yields in wetter basins and sharpest declines in Central Texas. We also find that while storage capacity has doubled in the United States since the 1980s, the available water in these reservoirs has decreased over the past 40 years.
- Previous work looked at storage capacity to evaluate changes in irrigation water supply and noted that the Colorado and Colombia basins had increased irrigation by 50% from 1981 – 2000. While we have not yet looked at the Colombia basin, but we plan to in the final draft, our current analysis shows decreasing storage trends in the Upper and Lower Colorado in this period.
- Our analysis also supports previous work that states storage has declined across CONUS, yet our storage changes are significant (compared to Zhao & Gao 2018) and are less than the 0 - -15% Wisser et al., 2013 depicted. In fact, their highest decreasing storage trend due to sedimentation in the Colorado basin (between -5% to -15%) is still a magnitude higher than the storage trend we observe when looking at the same regions (approximately -1%). This is due to Wisser et al., 2013 using storage capacity to analyze changes in a smaller subset of reservoirs.
- Our regional drought sensitivity analysis compliments the storage trends and allows us to provide potential reasons for the storage trends observed. Most notably on average, the more arid regions experience longer droughts, while more humid regions experience shorter hydrologic droughts. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio.
- Paper Unfocused
We agree that our paper was a bit too broad. In our revisions we will address this by clearly focusing all our analysis around how the directly observed behaviors we quantify compare to common assumptions made in larger scale reservoir parameterizations. This will also improve our relevance to the special issue and more directly provide value added for large scale models. Our analysis will be grouped into two parts (1) general behaviors and how they compare to assumptions, (2) the degree to which behaviors are changing over time and our models need to adapt. We have illustrated a start to how we will accomplish this in the table provided at the end of our response where we have summarized how our analysis supports or refutes findings of previous studies.
What we will do moving forward
We acknowledge that the reviewers had significant concerns regarding both the novel contributions of our work and our relevance to the special issue. As discussed above we plan to refocus our paper around how our results agree and disagree with common assumptions and to highlight the novel findings that might change assumptions made when modeling reservoir operations.
We will modify our introduction, discussion, and conclusion to center around our main points stated above and to clearly state our purpose. We will also expand on the table provided at the end of our response and incorporate this into our manuscript to clearly highlight how the key findings support or refute previous work. After reorganizing we will also spend time polishing the article for grammatical errors and clarity. It is the hope that these edits will ensure the article is neat, clear, and direct.
In addition to the writing plan outlined here, we also plan to update our analysis to include more individual dam analysis. To increase the number of dams in our study, we will decrease the threshold for regional inclusion to 40% instead of 50%. This allows us to include HUC 7 (Upper Mississippi) and HUC 17 (Pacific Northwest) in our analysis. Secondly, we will remove Figure 4 as the information is redundant with Figure 3. This is done to create space for the two additional figures, while also keeping the necessary information on operational ranges for the trend analysis in Section 3. The first new figure will have two panels where panel a is a point map of all the dams in our analysis colored by the month of their peak fraction filled and panel b is a point map of the month of largest operational range. The second figure will also be a point map of all the dams colored by if the dam has a positive or negative trend over the 40-year period we are looking at.
- There exist many grammatical and dictation mistakes which suggest that the paper is rushed, and not enough time has gone into the revision and polishing process. In addition, there are many general terms used in the text that should be explained in more details. Some of these mistakes and general terms are mentioned in the minor comments.
Thank you for your comment. We will do an extensive edit for grammatical errors and ensure that all terms used in the paper are clarified appropriately. We will also ensure that all general terms are explained in detail to ensure clarity and congruency.
- Considering that a major part of the analysis is statistical, I think that preprocessing of data should be explained in more details. For instance, how outliers were chosen, and how plentiful they were. What is the percentage of missing data that is calculated by linear interpolation? Or explaining Sen’s slope in more details and explaining why 0.1 value was chosen to signify a significant trend? In addition, you may consider using a larger number of distributions rather than GEV. Some of the points are also mentioned by line in minor comments.
We agree that more description of the methods used are appropriate for the statistical analysis. Outliers in reservoir storage were chosen as points that existed above the storage capacity values in GRanD. In some cases, there were dams (approximately 100) where the storage consistently sat above the storage capacity in GRanD. These outliers could be a result of Lehner et al., 2011 replacing maximum storage capacity with normal storage capacity due to limited data. In these cases, we opted to use the maximum observed storage instead of the maximum storage capacity in GRanD as our storage capacity to ensure storage changes were deducible.
At this point, we went through and calculated the percentage missing from the 1980 –2019 period for each region. When aggregating by region, we initially took the average storage value for each region and that yields no missing data. At this point, we have not yet done the individual dam analysis, but we will include the percentage of data that is gap-filled in the next round of edits.
Sen’s slopes are less biased towards end points than linear regressions as the slope is calculated between consecutive points and the final slope value is the median of all these slopes. This is preferred as we are able to see what the median rate of change across all the points are. Secondly, this method allows us to have a confidence interval to determine how likely it is that these slopes are due to random chance. We opted for 10% or 0.1 as our confidence bounds as this is the typical value for most statistical analysis.
We did look at a variety of different of distributions (log-logistic, log-normal, Pearson type III, generalized pareto, and Weibull) when calculating the Standardized Streamflow Indices. Upon analysis and through reading the Vicente-Serrano et al., 2012, we opted to only use the GEV distribution. This was due first to the robustness of GEV explained in the Vicente-Serrano et al., 2012 for multiple hydrologic conditions (both high and low flows) as well as our desire to ensure all the indices had the same distribution fit for congruency.
- The results and discussion are mostly limited to identifying trends in reservoir storage using different metrics. There should be a much deeper focus on what is the implication of these trends, seasonality, and uncertainty in storage for water resources engineers and planners, and elaborate on how the insight from this research helps them to better understand the risks involved with a reservoir storage management and better prepare for the uncertain future.
Thank you for your comment. We agree that more analysis related to the meaning of these results is needed in the discussion section and our generalized response does note some of the updates we will include. To summarize the generalized response, we will focus part of the discussion on either refuting or supporting common assumptions made in reservoir models (i.e., minimum storage capacity is 10% of maximum storage capacity) and provide cases where this might not be entirely accurate as dams due to discrepancies between observed storage values and reported maximum storage capacity values in datasets such as GRanD. In the 100 instances where storage capacity values routinely fall below observed storage values, modelers and water managers may be overestimating the amount of storage in a reservoir at a given time. Additionally, the underestimation of seasonal dynamics in reservoir models also contributes to this underestimation and therefore our seasonal analysis is important.
When looking at implications of reservoir storage trends, drought sensitivity and seasonal trends, the effect of climate and demand cannot be underestimated. When decreasing storage trends are combined with drought sensitivity, we see how shifts in demand or climate have affected these trends. For more humid regions, such as Lower Mississippi and California, we see high storage ratios that denote storage takes much longer to recovery than streamflow. This suggests that the decreasing storage trend in Lower Mississippi might be more related to demand and infrastructure shifts than directly related to climate. Conversely, the medium recovery ratio (storage takes slightly longer to recover than SSI) in Upper Colorado suggests that demand may not be the sole issue and climate is playing a large role in this storage ratio. Finally, the operational range trends demonstrate that flood control dominated basins in Tennessee, Ohio and Lower Mississippi have decreased, while California has increased. These trends are potentially due to the impact of climatic shifts which resulted in changes to operational policies.
- Does using the drought index for reservoirs that are designed for flood control provide any valuable insight? I think that it would be better to consider how trends in reservoir storage in humid regions affect the ability of reservoirs to control flooding, especially for extreme events.
We agree that floods are also interesting, however, this would require an event-based analysis which is outside our scope. Additionally, analyzing hydrologic drought can be useful for flood control reservoirs as we do see flood control regions with storage that takes longer to recover than the streamflow index (Ohio, Tennessee, Lower Mississippi and Sourris Red Rainy). Our updated analysis will focus on providing linkages between the drought sensitivity and storage trend analysis to inform potential causes of the declining trends seen in CONUS.
Minor comments:
Thank you for all these minor comments. In responding to this review, we went through and made the necessary grammatical edits stated and directly answer relevant questions. We also plan to dedicate time to polishing the resubmitted article.
Line 28-30: Grammatically incorrect: based ‘on’ various remote sensing …
We have updated this.
Line 32: You are mentioning that you have also used ‘release’ data to study the historical behavior of reservoirs, while in practice, you did not use the release data in your analysis.
Thank you for noting this. We will update this to state storage data and we in fact did not use release data.
Line 38: Grammatically incorrect: There are 2,000 large dams ‘that’ are spread out across the US
We have updated this.
Line 25-38-132: In line 25, large dams are defined as having a capacity larger than 10 km3, while in line 38, the value is equal to 0.1 km3. Also in line 132, the values that defines large dams in 0.01 km3. Which one is the better definition of large dams?
Thank you for pointing this out. The better denotation of large dams is 0.01km^3 based on the standard in GRanD and we have updated this to be the case.
Line 49:50: It is better to not use ‘in fact’ and ‘finally’ in the same sentence. You also mentioned ‘water supply’ in the previous sentence and are repeating it again here.
We will note this and make sure to vary the language within sentences to avoid repetitions.
Line 54: Please explain how dams can decrease the severity of extreme events in one or two sentences. In line 74-75, it is also mentioned that reservoirs can increase the drought duration.
This is a great place to improve our detailed analysis of the drought sensitivity. Preliminary analysis into this suggests that the larger the dams are in a region, the longer it takes for a region to recover. Most specifically, this is seen in the western US where the occurrence of large dams causes hydrologic drought to perpetuate for longer, while on the eastern US hydrologic droughts do not last as long.
Line 63: How have dams changed the run-off regimes? Please explain in more details.
Dams change run off regimes in two ways. First, large dams increase the storage along river systems and decrease the amount of water that flows downstream naturally. Secondly, the occurrence of dams allows water to be released during non-natural periods, such as during low flows, or at different seasons. We will add lines to this section to clarify this point.
Line 65: What type of system? Please be a little clearer.
We have removed the reference to system and instead replaced it with “water supply”.
Line 57: Grammatically incorrect. I think that the word ‘promote’ can be removed.
Thank you. It can and we have updated this locally.
Line 68: Grammatically incorrect. Use ‘have’ instead of ‘has.
We have updated this.
Line 71: Disrupt how?
We will switch the wording to focus on the fact that flashier systems have changed reservoir operations. Most specifically, the eastern US has seen an increase in flooding events which has caused reservoir operators to release more water to keep flood storage low. We will include a brief explanation of this in the resubmitted version.
Line 72-74: The sentence is not clear. Two ‘increase’ in one sentence.
We have updated this locally.
Line 93-94: The sentence is grammatically incorrect.
We have updated this locally.
Line 112: There are two ‘is’ in a sentence.
This is updated locally
Line 114-115: What do you mean by ‘reservoir dynamic’? I think that you can be more precise about it.
Thank you. We agree that we can be more precise. We will edit this language to specifically point out the seasonality of reservoir operations and storage fluctuations in our resubmission.
Line 119: What trend?
The trends referred to here are changes in fraction filled and regional comparisons. We plan to rephrase this sentence in the resubmission to ensure more clear and concise wording.
Line 123-126: Again, you are using general terms such as ‘reservoir patterns’, and ‘regional dynamics’. I think that you should be more specific in defining these terms.
Thank you, we agree that we could be more specific. In addition to updating these instances, we also will go through the paper and remove broad or vague terms to enhance clarity.
Line 150: What do you mean by ‘hydrologic boundaries’?
The term “hydrologic boundaries” in this sentence refers to the watershed boundaries from the Watershed Boundary Dataset. We have updated hydrologic boundaries to read “watershed boundaries”
Line 168-170: How plentiful were the outliers which suggested values higher than the storage capacity? And how much higher were they compared to the capacity?
In our analysis there are approximately 100 dams with instances where storage values were higher than the maximum storage capacity. In most cases, the maximum storage value was off by less than 10 MCM. In some cases (less than 10), the maximum storage value was off by less than 100 MCM. It is very possible that some of these maximum storage values could denote spills, however, in all cases they were not simply point discrepancies and therefore we opted to change the maximum storage value.
Line 170: What was the average period of gap in the data? For instance, did you use the interpolation for a large gap of data, to the extent that interpolation may become irrational?
As we averaged the data for all the dams across the HUC2 units and then linearly interpolated, we did not have large gaps to fill (only one or two days of data). We also opted for the period from 1980 – 2019 as the majority of the dams have a full period of record between these dates. For the proposed individual dam analysis, we may need to evaluate the percentage of data that is gap filled to see if the results from these dams can still be used.
Line 177: I suggest using ‘received little impact’ instead of ‘had little impact’.
We have updated this locally.
Line 187: Please explain the ‘Fraction Filled’ in a few sentences before introducing the formula.
Thank you for mentioning this. We plan to include a brief two sentence introduction to what fraction filled is. We have updated this to read: “The FF timeseries uses the total average storage for a given day in each region in ResOpsUS and divides that storage by the total storage capacity of all the dams in that region on that same day.”
Line 208: Please briefly explain ‘Sen’s slope’.
We agree that a brief explanation of Sen’s slopes is necessary as well as our rational for using them instead of a linear interpolation. We will include a definition and rational of Sen’s slope similar to this:
“Sen’s slopes are the median of all the slopes calculated between consecutive data points and the final slope value is the median of all these slopes. This is preferred as we are able to see what the median rate of change across all the points are and are less biased towards end points.”
Line 224: Considering that you are using daily streamflow values, and not annual maxima, you should consider more than one distribution to find the best fit to the data. Or at least, you must use tests such as ‘Kolmogorov–Smirnov test’ to ensure that the probability distribution is decently representing the data.
As stated previously, we looked at multiple distributions to fit the streamflow data. To ensure that the distribution was the same across all gages and regions, we chose to use the most robust according to Vicente-Serrano et al., 2012 which was GEV. In our revision, we will test our assumption with a Kolmogorov-Smirnov test.
Line 226: SSI values ‘can’ be calculated …
This is updated locally.
Line 232: I think it is better to write the formulation for both conditions of P being larger or smaller than 0.5.
Thank you for your comment. We opted to use the same format as Vicente-Serrano et al., 2012 as these were not equations that we derived and would prefer not to change them.
Line 262: What do you mean by ‘reservoir setting’?
We have updated this to say climate.
Line 266: ‘dominate’
We have updated this locally.
Line 274: Hydroelectricity reservoirs are most common ‘in’ the Tennessee Basin and South Atlantic.
We have updated this locally.
Line 328: ‘ion’?
We have updated this to be “in.”
Line 334: The statement does not seem true. Based on the Figure 2, the lower Colorado river has a very small month-to-month variation.
Thank you for noting this. We should have added Upper Colorado to this sentence and not simply stated Colorado. We have updated this locally.
Line 337: ‘is’ shown in Figure 2.
We have updated this locally.
Line 341:342: This statement does not seem correct based on the Figure 2.
We agree that this sentence is not supported by the figure. We have removed it locally.
350: ‘Note’ that …
We have updated this.
For the figure 2, I think that in addition to the black line that shows the median Fraction Filled and purple shadow that shows maximum minus minimum, you can also add 95% confidence intervals (maybe as dashed lines) to make sure that an extreme outlier (minimum or maximum) does not greatly impact the shadowed purple part.
Thank you for this suggestion. We will look at adding the 95% confidence interval. At the time of submission, we were hesitant to add additional lines to ensure the panel plots were neat.
Line 378: It is mentioned that operational variance provides a more holistic measure compared to the operating range. However, I am not convinced why you had to use both metrics for the analysis. If variance is better, why using operating range? Does each one of them provide a separate vision that justifies using both? Otherwise, it might be better to use only one of them to avoid extra confusion.
In response to this and a previous reviewer comment, we have opted to remove the operational variance plots (Figure 4) and keep the emphasis on operational range. This is done to complement the operational range trends in Figure 7 and to ensure there is room for additional figures we have proposed.
Line 201 and 347: In line 201, it says that region with more than 40% of storage coverage are considered, while line 437 suggest the value as 50%. Which one is correct?
At the time of submission, we were looking at regions that had greater than or equal to 50% storage covered. Upon reviewer comments on including more regions, we have decided to decrease this threshold to 40% storage covered. We have updated both instances to read 40% storage covered locally.
Line 438: Why October? Please explain the reason for choosing this month.
By choosing one month to evaluate the trends over, we could ensure that, and storage trends were due to changes in storage levels and not seasonal storage fluctuations. We chose October as that is the start of the water year and is a key data point for reservoir operators to plan for storage in the coming years. The storage at the start of the operational year (typically in October) is also important for reservoir modelers as this value is used to derive reservoir storage and releases.
Line 514: On both regional and ….’a’ is extra.
We have corrected this locally.
Line 656: 83% ‘of’ regions …
We have corrected this locally.
Line 658-659: The sentence is grammatically incorrect.
We have corrected this locally.
Finding from Literature
Related Paper
Our Finding
Support or refute
Reason for refuting (Difference)?
Reservoir storage has declined by 10% over 30-year period
Adsumilli et al., 2019
Decrease by 0.2 % across CONUS over 40-year periods
Support decrease but refute the exact number across CONUS
Different period and different dams/number
Southeast region and Arid western region have sharpest declines
Zhao & Gao 2018
The regional trends in Figure 6 show sharp decreases in Lower Colorado, but much of the western US, Louisiana and Ohio all have sharp decreases
Supports claim about the Western US; refutes sharp decreases in southeast as only Lower Mississippi has sharp decrease
Different dams in the study, and looking at surface area change not volume
Reservoir levels in western US at historic lows
Cayan et al., 2010, Williams et al., 202
Western US has declines (per Figure 6), and all are at “historic” lows with Upper and Lower Colorado the steepest
Supports
West Coast relies most on storage
Biemens et al., 2011
Figure 2 shows that Western dams have much higher median fraction filled
Supports that there is more storage in the western US, but we do not directly speak to reliance
Mississippi, Lower CO, and St. Lawrence basins use storage to mitigate water scarcity
Gaupp et al., 2015
Figure 2 shows that these basins (aside from Mississippi) have large amounts of storage
Supports
West coast US water scarcity (esp. the Bravo, Colorado, and Brazos rivers) where water withdrawals exceed storage
Gaupp et al., 2015
As these basins (Texas Gulf, Upper and Lower Colorado) all have decreasing storage trends, we can expect this to be partially true
Support
Water demand in these regions is just part of the whole story.
Water yields from 1985 – 2010 are greatest in wetter regions (i.e., east of the Mississippi and Pacific Northwest)
Brown et al., 2018
Aside from Sorris Red Rainy and Tennessee, Figure 6 refutes this claim
Refute
Over our period (1980- 2019) it appears that water yields could be higher, but there appears to be no real difference across the west and east
Water Demand highest in California and midwestern US
Brown et al., 2010
Figure 2 shows large amounts of storage in the western US probably due to the need for large demands for water
Support
The US has doubled the surface water availability with reservoirs.
Additionally, Colorado and Colombia basins have increased irrigation water supply by over 50% from 1981 - 2000
Biemen et al., 2009
As we have not yet looked at the pacific northwest, we cannot state whether reservoir water supply has increased. We can note that Figure 6 shows decreasing fraction filled trends for Upper and Lower Colorado
Refutes
Upper and Lower Colorado have decreasing storage trends which would suggest that that component of irrigation water supply has decreased from 1981 - 2019
Army Corps of Engineers analysis:
North Atlantic and Mississippi valley had 61 reservoirs that spent portion of the year above operational targets (mainly flood control dams with wetter conditions since 1970 (McCabe and Wolock 2002; Lins and Slack 2005; Hayhoe et al. 2007).
Patterson et al., 2018
Figure 7 shows that eastern reservoirs have more decreasing trends with respect to operational range, therefore their reservoirs should be closer to or under operational targets
Refutes
Could be the regions/dams we are looking at and the difference between operational range and operational targets
Remote sensing of evaporation changes from 1984 – 2015
Decreasing surface area for 134 out of 724 (central and western US) and increasing trend for 158 (mostly in the eastern US).
Zhao & Gao
Figure 6 agrees that there are decreasing trends in western US and agrees with increasing trends in part of the eastern US
Supports and refutes part
Only two basins in the eastern US have increasing trends, the rest are decreasing.
Regional decreases (average = -1%) are numerically larger than increases (average = 0.4%) which generally correlates to precipitation trends
Zhao & Gao 2019
Slopes from Figures 6 agree, yet magnitude is different (6.8*10^-4% for western region, and 0% for eastern regions)
Refutes numbers
Due to inclusion of more regions in Zhao & Gao and surface area vs storage
Central US (TX, OK, LA) has the largest decrease and AZ, NM, and CO all have decreasing trends.
Zhao & Gao
Figure 6 slopes disagree on central US as largest decrease
Refute
Lower Colorado has the largest decrease, Upper Colorado and Louisiana are second highest
Increasing precipitation trend in eastern US leads to slightly more surface area
Zhao & Gao 2019
We see more storage in western US (Figure 2) and more monthly variability in eastern US (which could lead to increased surface area)
Support
Surface area trend -0.011 *10^-9 m^2/year is not significant based on t test
Zhao & Gao 2019
Storage decreases of -0.2% across CONUS
Support
Both show a decrease of water in reservoirs, yet the two cannot be directly compared due to the different units
For instance, the surface areas of the reservoirs in the southwestern region of the US have shown signficant decreasing trends, which can be explained by the reduced precipitation during the last three decades (reported in Prein et al., 2016 and Barnett and Pierce, 2008)
Zhao & Gao 2019
The combination of Figure 6 and Figure 8 support this
Support
South Atlantic, South Pacific (i.e., California), and Southwestern Divisions had 85 reservoirs that spent a portion of the year under operational targets (possible due to multiple drought events)
Patterson et al., 2018
Figure 7 supports the statement about South Atlantic and Southwestern US, yet South Pacific is different
Support and refute
South Pacific (California) sees a large amount of positive operational range trends suggesting operators are staying above operational targets
71% of large and frequent departures from operational targets occurred in Upper Mississippi, Great Basin, Texas
Patterson 2018
Supports but none have a lot that are statistically significant
Support
Reservoir regulation reduces flood and drought severity, yet increases the duration locally and increases spatial drought connectedness
Brunner, 2021
Figure 8 supports the increase in drought duration in western reservoirs and Lower Mississippi, yet refutes this claim in the South Atlantic
Support and refute
The South Atlantic is the only region with storage recovering much quicker than SSI
1984 – 2015 for global reservoirs and found that the Southwestern US greatest storage declines in arid/subhumid basins in US southwestern US (-10%)
As well as reduced streamflow and precipitation trends
Hou et al 2022
Figure 6 supports this
Supports
Reservoir resilience in southwestern US and Mississippi basins has decreased drastically with increased vulnerability 30%
Notes: They saw positive relationship between reservoir storage and resilience with negative relationships between resilience and vulnerabilities (i.e., decreasing storage makes reservoirs more vulnerable to changes
Hou et al., 2022
Figure 8 supports this
Supports
Southwestern US has negative trends (between -0.1 - -0.5 m/year) while eastern US has slightly positive trends (0- 0.25)
Kraemer et al., 2020
Figure 6 supports this
Support
Difference in reservoirs looked at as this paper also used lakes
North American reservoir storage losses are 3.2% when looking at sedimentation
Wisser et al., 2013
Figure 5 shows that storage declines in US are 0.2%
Supports but refutes exact number
Wisser et al., 2013 is using a model to estimate storage changes from storage capacity
Citation: https://doi.org/10.5194/egusphere-2022-1051-AC3
Data sets
ResOpsUS Jennie C. Steyaert, Laura E., Condon, Sean W. D. Turner, Nathalie Voisin https://doi.org/10.5281/zenodo.5367383
Model code and software
ResOpsUS_Analysis Jennie C. Steyaert, Laura E. Condon https://github.com/jsteyaert/ResOpsUS_Analysis
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 617 | 352 | 39 | 1,008 | 21 | 27 |
- HTML: 617
- PDF: 352
- XML: 39
- Total: 1,008
- BibTeX: 21
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1