Assessing the effectiveness of reservoir operation and identifying reallocation opportunities under climate change

Reservoirs are designed and operated to mitigate hydroclimatic variability and extremes to fulfill various beneficial purposes. Existing reservoir infrastructure capacity and operation policies derived from historical records are challenged by hydrologic regime change and storage reduction from sedimentation. Furthermore, climate change could amplify the water footprint of reservoir operation (i.e. non-beneficial evaporative loss), further influencing the complex interactions among hydrologic variability, reservoir characteristics, and operation decisions. Disentangling and quantifying these impacts is essential to assess the effectiveness of reservoir operation under future climate and identify the opportunities for adaptive reservoir management (e.g. storage reallocation). Using reservoirs in Texas as a testing case, this study develops data-driven models to represent the current reservoir operation policies and assesses the challenges and opportunities in flood control and water supply under dynamically downscaled climate projections from the Coupled Model Intercomparison Project Phase 6. We find that current policies are robust in reducing future flood risks by eliminating small floods, reducing peak magnitude, and extending the duration for large floods. Current operation strategies can effectively reduce the risk of storage shortage for many reservoirs investigated, but reservoir evaporation and sedimentation pose urgent needs for revisions in the current guidelines to enhance system resilience. We also identify the opportunities for reservoir storage reallocation through seasonal-varying conservation pool levels to improve water supply reliability with negligible flood risk increase. This study provides a framework for stakeholders to evaluate the effectiveness of the current reservoir operation policy under future climate through the interactions among hydroclimatology, reservoir infrastructure, and operation policy.


Introduction
Climate change is reshaping the hydroclimatic landscape, altering precipitation patterns (Muller and O'Gorman 2011), evapotranspiration characteristics (Milly and Dunne 2016), runoff regimes (Döll and Schmied 2012) and terrestrial water storage (Pokhrel et al 2021).These alterations can affect future water availability and demand (Schewe et al 2014, Konapala et al 2020, Caretta et al 2022), raising concerns about water management under a changing climate (Milly et al 2008, Cosgrove andLoucks 2015).Reservoirs, as one of the fundamental water infrastructures, are not immune to these changes.In the United States, more than 92 000 dams constitute the cornerstone of water resources management for various purposes, including flood mitigation, water supply, navigation, hydroelectricity generation, recreation, and environmental flow (National Inventory of Dams n.d.).However, the reservoir storage capacity and companion operation rules derived from historical hydrologic records (usually with limited record length) may fall short of accounting for the full spectrum of probable hydrologic variability and change (Ho et al 2017).For example, the 1922 Colorado River Compact was based on an anticipated average annual flow of 16.4 million acre-feet (MAF) measured from the relatively wet two decades at the beginning of the 20th century (Castle et al 2014).However, subsequent tree ring studies revealed lower long-term average annual flows for the Colorado River, ranging between 13.2 MAF (Hidalgo et al 2000) and 14.3 MAF (Woodhouse et al 2006).As a result, Lake Powell only reached its full capacity once in 1983 since its commission.Current reservoir systems on the Colorado River suffer from the overestimated water availability, which is further exacerbated by the ongoing Millennium Drought.Given inadequate historical records and non-stationary future climate, an inquiry emerges from reservoir operation stakeholders: Are de facto reservoir storage and operation rules, the backbone of modern water infrastructure, adequate to fulfill designed objectives under future climate?
Man-made dams and reservoirs function as coupled nature-human systems (Liu et al 2007), where both system capacity and operation policy are the outcomes of natural (e.g.hydrologic variations) and anthropogenic drivers (e.g.operation targets, construction costs).Reservoir operation policies are mostly derived from optimality under a set of objective functions and constraints to tradeoff between the hydrologic variability and operation targets (Giuliani et al 2021).For example, the objective of maximizing water storage for supply often conflicts with the need for flood control.During flood seasons, reservoirs must reserve certain storage space to accommodate potential floodwaters, reducing the hydroelectricity generation and water supply reliability for later uses (Krzysztofowicz andDuckstein 1979, Jain et al 1992).Conversely, in dry periods, maximizing storage for water supply can compromise the reservoir's ability to handle flash floods (Croley et al 1979, Ding et al 2015).Moreover, one of the key factors usually missing in the analysis is the water footprint from reservoirs (i.e.represented by evaporative loss), especially in arid-and semi-arid regions.Mekonnen and Hoekstra (2012) estimated that the blue water footprint of a mere 8% of global hydroelectric capacity is equivalent to 10% of the water footprint for global crop production in 2000.Zhao et al (2022) demonstrated that the quantity of reservoir evaporative loss is equivalent to 20% of the global annual consumption of water use.The non-beneficial evaporative water loss is also an outcome of this coupled nature-human system, arising from interplays between hydrologic variability (e.g.seasonal inflow patterns and potential evapotranspiration), reservoir characteristics (e.g.reservoir head-area-volume relationship), and human management decisions (e.g. release schedules, conservation strategies).Despite the significant amount of reservoir evaporative loss, it remains unclear whether hydroclimatology and operation policy jointly affect reservoir evaporative loss and further contribute to the re-operation of reservoirs.Knowing the dominant factors and their temporal patterns in controlling reservoir evaporative loss will enable quantifying the tradeoffs among reservoir performances and the operation costs.
The challenges from the changing climate and potential benefits of reservoir re-operation pose the urgent need for examining options for adaptative reservoir management.Expanding reservoir infrastructure capacity would be one solution, but it is expensive under various physical and financial constraints (Iglesias and Garrote 2015).Another more viable avenue is to reallocate existing reservoir storage capacity among various project purposes (Carriere and Wurbs 1988, Johnson et al 1990, Wurbs et al 1990).Reallocation is the redistribution of a reservoir's water storage across its pools to address significant, long-term operational changes or specific requirements.The United States Army Corps of Engineers (USACE) has implemented 161 reservoir pool reallocations at 56 reservoirs for water supply by raising the top of the designated conservation pool, thus prioritizing water conservation purposes over flood control during dry seasons (Doyle and Patterson 2019).Reservoirs with a distinct seasonal operation pattern would most likely benefit from seasonal pool reallocation, such as those in Texas.For example, a seasonal-varying operation rule curve was planned and implemented to replace the original constant conservation pool since 1990 in the Wright Patman Reservoir, TX, allowing the conservation pool elevation to increase from an original 220.6 ft to a higher level of 227.5 ft between 1 April and 1 June.Summertime water storage capacity surged from 122 639 to 310 428 acre-feet, providing a much larger room for water reservation to mitigate drought conditions.Therefore, reservoir reallocation provides a promising approach for adaptative management under future climate with existing reservoir infrastructure.However, it remains unclear how to identify the strategies (e.g.timing and amount) of reservoir reallocation to enhance the overall system performance.
With the pressing needs discussed above, this study aims to investigate the challenges and opportunities of reservoir infrastructure and operation policy under a changing climate.Specifically, we will address the following research questions.First, are current reservoir infrastructure and operation policy sufficient to mitigate hydrologic changes under future climate to achieve designed operation goals?We will build data-driven models to represent the reservoir manager's operation decisions.The data-driven reservoir operation models will be driven by projected hydroclimatic forcings to evaluate the system performances (e.g.flood risk mitigation and water supply reliability) under future climate scenarios.Second, how do hydroclimatic variation, reservoir characteristics, and operation policy jointly determine the reservoir evaporative loss?A better understanding of reservoir water loss under interactions between hydroclimatic and anthropogenic drivers can further indicate the opportunities to evaluate tradeoffs and synergies among operation targets.We will incorporate reservoir evaporation processes into the data-driven reservoir model to represent the interactions between natural and operational components of reservoir dynamics.Third, can we identify the opportunities for reservoir re-operation strategies (e.g.reallocation of storage pools) to enhance the system performance with the de facto infrastructure?We will focus on the strategies of seasonal-varying (i.e. in terms of timing and amount) reservoir conservation pool reallocation, which is mostly viable for real-world reservoir adaptive management.We applied the generic framework and methods in this study to reservoir systems in Texas as a user case due to the complexity of the hydroclimatic regime and the significance of mitigating hydrologic extremes (e.g.floods and droughts).In Texas, surface water makes up almost two-thirds of the total existing water supply (8.9 MAF per year) for municipal, manufacturing, steam-electric, and mining users (Texas Water Development Board 2021).The operation target may switch from flood control to water conservation within a short period (e.g.weeks) due to the quick shift from flood season to dry season.The aridand semi-arid climate also exemplifies the importance of evaporative loss in reservoir operation.

Data-driven reservoir operation models under hydroclimatic and anthropogenic forcings
In this study, we focus on a set of 21 reservoirs located in eastern Texas, which are owned and operated by the USACE Tulsa District and Fort Worth District (details in supplementary materials table S1).To construct the data-driven reservoir operation models that mimic the existing regulatory rules, we utilize the standardized database for historical daily reservoir levels and operations of USACE reservoirs developed by Patterson and Doyle (2018).These observed records include daily reservoir storage volume (acre-feet, af), inflow (cubic feet per second, cfs), and release (cubic feet per second, cfs) for each reservoir.Gaps in the dataset are filled using nearest neighbor interpolation.The output variables of our data-driven models include reservoir release, evaporative loss, and water storage.The model forcings include reservoir inflows, air temperature (as a proxy for potential evaporation; Livneh et al 2015), and reservoir water level (as a system state variable).Specifically, we include the current inflow, the previous 7 day inflow history, and accumulated inflow volumes for the preceding 15, 30, 90, 180, and 365 d.Recent instantaneous inflows represent the short-term hydrologic variability to capture flood mitigation-related release decisions, while accumulated inflow volumes represent the trend of water availability for water supply operations.We also include static inputs including reservoir conservation capacity and dead pool capacity to represent the reservoir infrastructure constraints.All water-related variables are converted to million cubic meters (mcm).We apply the long shortterm memory (Hochreiter and Schmidhuber 1997) to produce three daily outputs: storage, evaporation, and release.While some studies use reservoir water storage as an input in data-driven models (Yang et al 2016, 2021, Longyang and Zeng 2023), it is noted that reservoir water storage level (a state variable) needs to be explicitly represented in the reservoir model to capture reservoir performance under future climate.This study updates daily reservoir storage in the simulation loop (i.e.simulated storage in the previous day is fed as one of the inputs for the current day's reservoir operation decision).This approach avoids calculating storage based on reservoir water balance where some water budget items (e.g.diversion and leakage) are not available.60% of data records are used during the training process, 10% of them for validation, and the rest for testing.The Nash-Sutcliffe efficiency (NSE) (Nash and Sutcliffe 1970) of the three outputs time series (storage, evaporation, and release) is applied to evaluate model performance.The well-trained reservoir operation models represent the actual reservoir operation decisions under historical hydroclimatic variability and existing infrastructure capacity constraints.
The future climate forcings are dynamically downscaled six GCMs outputs (i.e.ACCESS-CM2, BCC-CSM2-MR, CNRM-ESM2-1, MRI-ESM2-0, MPI-ESM1-2-HR, and NorESM2-MM, details in the supplementary materials table S2) in Coupled Model Intercomparison Project Phase 6 (CMIP6) under the Shared Socioeconomic Pathways (SSPs) 585 emission scenarios (Rastogi et al 2022).The biascorrected 3-hourly climate forcings are fed into the calibrated Variable Infiltration Capacity model version 5 (Hamman et al 2018) with 1/24 • (∼4 km) spatial resolution to simulate runoff (Kao et al 2022).We aggregate runoff generated in the upstream watershed of reservoirs to generate reservoir inflow.Future air temperature is obtained from Rastogi et al (2022).It is noted that the future climate from SSP 585 represents a high CO 2 emission scenario with no mitigation policies (O'Neill et al 2016).Therefore, we primarily focus on the effectiveness of current reservoir management strategies under the most extreme future scenario in this study.In the assessment under future scenarios, all reservoirs retain their conservation pool capacity the same as that in 2023.We also include sedimentation scenarios following the projected rating curve developed by Zhu et al (2018) to represent the reduction in the reservoir storage capacity.
Reservoir reallocation is simulated by allocating 5% of reservoir storage capacity from the flood control pool to the water conservation pool to enhance water supply performance at the expense of a potential increase in flood risk.We test reallocation scenarios each month and year-round to illustrate the reallocation strategies under interactions between hydroclimatic and operational factors.

Characterizing hydrologic extremes and reservoir operation performance
Since most reservoirs in the study area serve both flood control and water conservation purposes, we investigate both flood mitigation and reservoir storage as reservoir operation performance indices.Flood control performance is evaluated by comparing reservoir inflow and release.We define a flood event as when the streamflow exceeds the 90th percentile of historical data, with a prescribed time lag of at least 10 d between such events.The start and end of an event are marked by the discharge exceeding the threshold before the peak occurrence and falling below the threshold after peak occurrence, respectively.For each flood event, we define its characteristics including peak discharge (cubic meters per second, cms) and duration (days) (for an illustration see supplementary materials figure S2).
The water supply performance at each reservoir is summarized by the reliability, resiliency, and vulnerability of the reservoir storage time series defined in Hashimoto et al (1982).We use 70% of the conservation pool water supply target, aligning with the Texas Water Development Board's (TWDB) Texas water conditions reports, which label reservoirs as in a 'Normal to High' condition when over 70% of their conservation storage capacity is filled.Reliability measures the likelihood that a reservoir can fulfill its water supply target calculated as the percentage of days when reservoir storage is at least 70% of its conservation pool.Resiliency measures how quickly a reservoir recovers from failure states when the storage level is less than 70% of its conservation storage capacity.Vulnerability quantifies how severe the consequences of failure may be (how unfilled the reservoir is), calculated as the proportion of the shortfall (amount of water required to reach 70% of the conservation pool) to the total capacity of the conservation pool.
The trained data-driven models (representing the existing operation policy) are driven by each projected climate forcings in the historical period  to represent baseline performance and future period (2015-2059) to represent future performance.This allows the evaluation of reservoir performance under consistent climate change scenarios.

Structure equation modeling (SEM) to quantify the tradeoffs between reservoir storage shortages and high inflow conditions
We utilize structural equation modeling (SEM) to disentangle the complex relationships among hydrologic variability, operation strategies, evaporation, and reservoir storage shortages.As a robust multivariate statistical tool, SEM enables the simultaneous examination of multiple observed and latent variables (Kline 2015).It combines factor and path analysis for comprehensive cause-effect modeling (Bollen and Pearl 2013).In our SEM model, we establish 'StorageDrought' as a latent variable, measured by three observed metrics: the frequency, duration, and magnitude of occasions when reservoir water levels fall below 70% of the conservation capacity.We particularly focus on the role of high flow (defined as flows exceeding the 70th percentile of historical inflow) in potentially mitigating storage drought.For this, we introduce 'HydroCondition' as another latent variable, measured by the frequency, duration, and magnitude of high inflow events.Reservoir management practices during flood seasons aim to conserve water and lessen storage drought risk.We thus introduce a third latent variable, 'OperationStrategy' , measured by the frequency, duration, and magnitude of water-release events.Reservoir evaporation is included as a factor exacerbating storage drought, with all variables analyzed on a monthly scale.

Results
The trained reservoir operation models demonstrate robust performance across most reservoirs investigated in this study, as detailed in supplementary materials figure S1.The median NSE values on the test set for reservoir storage (or reservoir water level), evaporative loss, and water release are 0.991, 0.984, and 0.938, respectively.To evaluate the performance of the current reservoir operation policy in terms of flood control and water supply under future climate, we run the trained reservoir model with baseline and future streamflow from the downscaled hydroclimatic models.

Performance of current reservoir operation policies for flood risk mitigation and water supply under future climate
The change of high flow conditions under future climate is evaluated based on the differences between future (near-future, 2020-2039 and mid-term future, 2040-2059) and baseline (1990-2019) periods' reservoir inflows.There is a modest change (within a range of ±20%) for most reservoirs in terms of the number and duration of flood events (figure 1(a)) compared to the baseline period.However, all reservoirs are projected to experience larger flood peaks (rising by between 2% and 62%), indicating a higher risk of more severe flood events upstream in the future.
We compare the reservoir inflow and release to quantify the changes in flood risks to evaluate the reservoir's performance in mitigating flood risks.Existing operation rules are largely effective in reducing flood risks across the baseline, near-future, and mid-term future periods (figure 1(b)).Reservoirs manage to decrease the frequency of flood events (around −10% to −80% across all 21 reservoirs for all periods) and reduce the magnitude of flood peak flows (around −2% to −90%) by eliminating smaller floods and reducing the magnitude of flood peak flows.The flood water is temporally saved in the reservoir and released after the event, as indicated by extending the duration of events.However, Pat Mayse Lake stands as an exception.It may be attributed to the largest shifts in flow timing in the future (as detailed in supplementary materials figure S3), specifically a higher concentration of water flow in the flood season, particularly in the summer.This could also highlight potential limitations in current management strategies for future conditions.
The impact of climate change on the water supply performance differs across reservoirs (figure 1(c)).In the near future (2020-2039), water supply reliability will experience mild changes (within ±15%), a trend poised to persist and amplify in the mid-term (2040-2059), potentially benefiting from increasing incoming water quantities.The reservoir in Texas tends to maintain the water level at the designated top of conservation pool elevation as stream flows and water demands allow (Wurbs 2021).Minor changes in reliability are expected and align with the purposes of these reservoirs to provide a consistent water supply.While reliability remains fairly stable, the resiliency and vulnerability of the 21 reservoirs are noticeably affected.In the near and mid-term future, with existing operational rules, some reservoirs are projected to experience increased resiliency and decreased vulnerability.However, not all reservoirs are resistant to future droughts.Some, like Canyon Lake and Bardwell Lake, face a future with declining resiliency and heightened vulnerability, leading to severe failures and recovery difficulties.This suggests an urgent need for managers to revisit and update the existing operation rules to better equip themselves against the anticipated higher risk of extreme events and potential degradation in water supply performance.

Impact of feedback between operation strategy and hydrologic condition on storage drought and evaporative loss
The reservoir storage droughts shown in figure 1(c) would be caused by intertwined factors including inflow regime, evaporative loss, and reservoir release.For example, more frequent high flows would lead to spilled flood water, which decreases the water stored in the reservoir even with more inflow quantity.High water levels under the conservation operation target lead to a larger reservoir surface area and enhanced evaporative loss, which may result in reservoir storage drought.The SEM analysis provides quantitative insights from the machine learning reservoir model to attribute the hydroclimatic and operational factors on reservoir storage drought.The SEM analysis implies that current management practices adopted in Texas generally help in mitigating the risks of water shortages, thus facilitating a stable water supply, while evaporation emerges as a deteriorating factor for the system performance in many reservoirs.For example, in the case study of Grapevine Lake (figure 2(a)), operation strategies for managing high inflow effectively alleviate storage drought (with a path value of −0.71); conversely, evaporation exacerbates storage drought (with a path value of 0.09).The path value in figure 2 measures the direct impact of one variable on another within the model, where its magnitude reflects the strength of this effect and its sign (positive or negative) denotes the nature of the relationship, similar to a regression coefficient in linear regression.Figure 2(b) further summarizes the quantitative influence of hydrological conditions, operation strategies, and evaporation on storage drought of selected reservoirs.For most reservoirs, the current operation strategies stand as the paramount factor in reducing storage drought occurrences.However, there are exceptions; a few reservoirs either do not exhibit a statistically significant relationship with any of the factors examined (e.g.Lake Waco) or show a higher correlation with other factors like evaporation (e.g.Canyon Lake) or hydrological conditions (e.g.Navarro Mills Lake).This may indicate that operation strategies may be less effective in these reservoirs, requiring an update of the current regulation rules to better capitalize on high inflow conditions.
The non-beneficial reservoir evaporative loss represents the water cost of reservoir operation and will affect the tradeoffs among various operation targets.In a specific reservoir, meteorological conditions and the reservoir water level (or surface area) jointly influence the evaporative process over seasons.Monthly correlation analysis identifies dominant factors influencing reservoir evaporation across seasons under the six climate-forcing scenarios, as shown in figure 3. We categorize the reservoirs into three groups based on the dominant influencing factors: (1) storagedominated (such as the Pat Mayse Lake), where the water storage level is the principal determinant of evaporation; (2) temperature-dominated (such as the Jim Chapman Lake), where temperature mainly governs evaporation all year round; (3) seasonally-hybrid (such as the Lake Waco), where storage and temperature alternately have a significant influence over different seasons.The different dominant factors on reservoir evaporative loss also provide insights for adaptive reservoir operation under changing climate.For seasonally-hybrid reservoirs, temperature is the primary control from May to August, which implies that in summer, curbing evaporation by managing water levels is largely unfeasible.Conversely, from September to December, usually the post-flooding season, water level becomes a more important factor influencing evaporation.Therefore, any decision to alter the conservation level for storing more water during these months should be approached with caution to avoid inadvertently increasing the evaporative loss.

Opportunity and strategies of reservoir reallocation to tradeoff between water supply and flooding control
Based on the analysis in sections 3.1, and 3.2, reallocation strategies by adjusting the existing conservation pool's upper limits may bring additional benefits for water supply performance without substantially raising the risk of downstream flooding and evaporation loss.For example, for Lake Waco (figure 4(a)), elevating the year-round conservation capacity enhances the reservoir's resiliency, with slight improvements in reliability and vulnerability.This positive effect is notably observed in August, around the end of flood seasons.Increasing the conservation pool level for the whole year only yields a slightly better performance than the August adjustment.The adjustment in August enables the harnessing of more floodwater during the final peak flows, subsequently delaying the onset of storage recession below the original top of the conservation pool (figure 4(c)).It facilitates a more effective recovery from storage drought (i.e. the water level falling below 70% of the conservation pool).Meanwhile, the elevated conservation pool does not significantly escalate the threat of downstream flooding, with fluctuations remaining within a ±0.2% range, highlighting the feasibility of adjusting the conservation capacity on a monthly or annual basis (figure 4(b)).Figures S5 and S6 in supplementary materials demonstrate the feasibility of reallocation in other reservoirs, with nuanced differences among cases.The optimal strategy for Lake Waco and Pat Mayse Lake is to temporarily elevate the designated top of the conservation pool in August; however, Jim Chapman Lake benefits from a February adjustment, given its additional flood events during the winter months.This adjustment for Jim Chapman Lake allows the reservoir to retain more floodwater following winter flood events, with slightly reducing the flooding risk downstream (supplementary materials figure S6).
Quantifying the tradeoff between high inflow harvesting and evaporative losses is vital for reallocation strategies to safeguard reservoir efficiency and stable water supply.The water saving from reallocation would be counterbalanced by the high evaporation loss, leading to negligible increases in the water supply performance.Taking Lake Waco as an example, compared to figure 4(a), supplementary materials figure S4 reveals that the elevated conservation capacity level in September is counteracted by an increase in reservoir evaporation.For Lake Waco (seasonally-hybrid type), September is a critical period following the flood season, where the water storage predominantly controls the reservoir evaporation, overshadowing the influence of temperature (see figure 3).Therefore, the analysis of dominant factors in figure 2 and their temporal pattern in figure 3 provides a quantitative simulation approach to develop reservoir reallocation strategies by considering the tradeoffs and synergies among the temporal characteristics of hydroclimatic and operational factors.It is noted that sedimentation (such as reduction in active reservoir storage) stands as an additional threat to reservoir operation performance besides evaporation.For Lake Waco, the reservoir sedimentation-induced capacity loss tends to adversely affect both resiliency (−13.2%) and vulnerability (+1.6%).This drawback can negate the benefits offered by reallocation strategies.While sedimentation slightly elevates the flooding risk downstream by intensifying its frequency, this increment is relatively minor, recorded at +3.8%.

Climate change uncertainty
The SSP585 scenario used in this study represents the most severe climate projection with an anticipated additional radiative forcing of 8.5 W m −2 by 2100 (O'Neill et al 2016).Therefore, the results reported here examine reservoir management in mitigating flood risks and securing water supply under extreme climate conditions.Characterized by a high reliance on fossil fuels and resource-heavy lifestyles (Masson-Delmotte et al 2021), SSP585 presents pressing challenges for water resource management.Expanding the scope, other scenarios such as SSP126 ('Sustainability' , adding 2.6 W m −2 radiative forcing by 2100), SSP245 ('Middle of the road' , with 4.5 W m −2 ), SSP370 ('Regional rivalry' , reaching 7 W m −2 ), and SSP460 ('Inequality' , with 6 W m −2 ), cover a spectrum of possible future conditions for water management.Further investigation under these scenarios can provide a broader probability range with different levels of climate protection measures, spanning from optimistic to more moderate future outcomes.However, the efficacy of such investigations hinges on the reliability of climate models.The six GCMs selected for this study have relative independence and low structural similarity, which should enhance the degree of confidence in climate projections (Pathak et al 2023).Despite this, as noted by Ashfaq et al (2022), these models have limitations in simulating seasonal mean precipitation and temperature for the South and West US regions.Furthermore, although dynamic and statistical downscaling methods are similarly effective in correcting historical climate simulations, the projected precipitation extremes based on Livneh (Livneh et al 2015) as the reference data tend to be underestimated (Rastogi et al 2022), potentially leading to underestimated flood risk assessments in our study.Additionally, bias correction, which primarily adjusts the monthly distribution of precipitation and temperature, may overlook interannual variability (Kao et al 2016).Coupled with the potential biases in hydrologic modeling due to variations in spatial resolution and parameterization (Kao et al 2022), the uncertainty in hydrological outcomes may be further compounded.This highlights the importance of cautious interpretation in applying these models to water resource management, considering the uncertainties inherent in climate change projections.

Reservoir operation targets
Evolving operational targets and water demand can present a mix of opportunities and challenges for future reservoir management.This study assesses whether current reservoir operation rules can continue to meet their targets amid a changing climate.It is assumed that the designated operational purposes of the reservoirs will remain unchanged in the future.However, these purposes may evolve, either realigning operational priorities or expanding to include new objectives (e.g.adding a hydropower component for hydroelectric generation, or allocating environmental flow for downstream ecosystem protection).Hadjerioua et al (2012) highlight that several non-power dams in Texas (including Belton Lake, Livingston Dam, etc) have the potential to generate significant amounts of clean, reliable hydropower, with the estimated annual average generation ranging from 59 153 to 327 233 MWh.As a case in point, Livingston Dam underwent modifications by adding three 8 MW Kaplan-type turbine-generator units, which began commercial operations on 14 July 2020, and now produce sufficient clean electricity to supply around 12 000 households in East Texas (Chase-Israel 2022).Such modifications could be implemented in other dams in the future as well, potentially leading to a shift in the original purposes and operational strategies of these reservoirs.While this could ease pressure on the power system, it may also bring conflicts between hydropower generation and flood control (Opperman et al 2022).Due to data and model limitations, this study refrains from directly modeling specific demands such as crop requirements or municipal and industrial (M&I) water demand.Instead, we assume that these aspects are implicitly embedded in the historical records, which have guided operators to meet water supply targets in the past.In fact, per capita water usage does not always align predictably with population growth, making it challenging for data-driven models to reasonably represent these complex dynamics.An example of this is seen in Texas, where remarkable water conservation efforts have led to a notable decrease in average municipal water consumption per person, from 175 gallons daily in 2000 to 138 gallons in 2015 (a reduction of around 21%), as reported by the TWDB.Continuing water conservation efforts in the future could help alleviate stress on water supply systems.A caveat is that we do not account for the evolving nature of water rights, regulations, and laws, which can change in response to various environmental and societal factors.It is promising for future work to incorporate details from region-specific water management models, such as the water availability model (Wurbs et al 2005), which contains the Water Rights Analysis Package executable programs for individual Texas river basins.

Forecast-informed reservoir operation (FIRO)
Effective management strategies to enhance the water supply performance may further benefit from incorporating streamflow forecasting information.The storage reallocation strategies examined in this study are based on historical data, overlooking real-time meteorological and hydrological forecasts.In the future, the actual flood season patterns may deviate from historical expectations.This deviation could pose water management challenges, such as unanticipated flood damages (i.e. if the flood season is delayed and the actual flood season is longer than a regular one, a flood event occurs beyond a regular flood season, Ding et al 2015) or prolonged droughts (i.e. if the actual flood season ends earlier than a regular one, the dry season will be extended).FIRO is a management strategy that leverages watershed monitoring data along with advanced weather and streamflow forecasts to aid water managers in making informed decisions based on current and anticipated conditions (Delaney et al 2020, Jasperse et al 2020).Short-term forecasts are informative for a water reservoir system primarily operated to fulfill the shortterm objective (e.g.flood control).They equip the system operators with the foresight needed to prepare buffer storage for mitigating the upcoming flood peak and thus minimizing flood damages (Wang et al 2012, Raso et al 2014, Zhao et al 2014).Medium to longterm forecasts offer valuable insights for reservoirs serving the purpose of water supply, which is typically characterized by slower dynamics that unfold over extended periods (Denaro et al 2017).Recent initiatives in FIRO have demonstrated the potential of utilizing forecasted rainfall and real-time reservoir inflow data to improve reservoir system management (Konieczki et al 2017, Delaney et al 2020, Zarei et al 2021).By taking advantage of accurate forecasts, the reservoir system in Texas can more effectively utilize the last flood event of each season to replenish the conservation pool instead of releasing water downstream, thereby enhancing water storage benefits.This strategy can simultaneously meet flood control objectives and water supply needs.
While FIRO is viewed as a cost-effective method for ensuring water supply and reducing flood impacts, its adoption in Texas is limited due to specific challenges, including limitations in forecast accuracy and specificity, infrastructural and operational constraints facing reservoir operators, and information gaps between forecast providers and operators (Fernando et al 2020).The expansive geographical scale and climatic diversity of Texas present a huge challenge for accurately predicting storm systems that lead to runoff and flooding over extended timeframes (Texas Water Development Board 2020).In Texas, the unpredictability of precipitation is substantially influenced by weather patterns like El Niño and La Niña (Hoerling et al 2013, Cheng et al 2018), and further compounded by the erratic nature of thunderstorms and tropical disturbances during the warm season (Nielsen-Gammon et al 2020).This unpredictability particularly impacts streamflow forecasts, where the accuracy in predicting high flows sensitively depends on the quality of the precipitation forcing (Kim et al 2018).Perhaps out of a risk-averse inclination, many reservoir operators in Texas may typically rely on absolute statements, not probability.Yet, constraints from infrastructure and operations demand skillful streamflow predictions.For those non-federal water supply reservoirs in Texas, which secure M&I water use in large cities, they lack a designated flood control pool and have limited capacity to regulate, store, and manage floodwaters.Therefore, pre-releases from conservation pools before potential flood events rely heavily on the accuracy of forecasts.To promote the adoption of FIRO, it is critical to ensure reliable forecasts for effective reservoir management, maintaining steady revenue and a continuous water supply for consumers.On the other hand, it is imperative to determine how to modify operation strategies to counteract the risks posed by climate change in these reservoirs.Furthermore, infrastructural challenges, such as the design of reservoirs for water conservation and the potential need for retrofitting or adding additional flood gates, must be tackled to fully harness the benefits of proactive forecast usage (Fernando et al 2020).

Conclusions
In this study, we evaluate the effectiveness of current operational rules in addressing future flood risks and water supply performance.While current practices effectively counter many flood risks, climate change amplifies challenges in reservoir resilience and vulnerability.In Texas, most operational strategies adeptly manage high inflow conditions, mitigating storage drought and ensuring a reliable water supply.Yet, certain reservoirs may not benefit sufficiently from these guidelines in the future, emphasizing the need for regulatory updates.The adverse impact of reservoir evaporation on the performance of many water supplies should be factored into these revisions.
We further analyze the potential benefits of adjusting the upper limits of reservoir conservation pools.Our findings highlight the feasibility of storage reallocation between the flood control and conservation pools to bolster water supply resilience without amplifying downstream flood risks.Most reservoirs investigated in this study would benefit from the reallocation by conserving water at the end of flood seasons, with negligible flood risk changes.This straightforward way can pave the way for broader FIRO adoption, which in turn offers more accurate dynamic storage adjustments to store more floodwater.Moreover, reservoir sedimentation reduces resilience and heightens vulnerability, underscoring the importance of sediment removal in ensuring sustainable operations.

Figure 1 .
Figure 1.(a) Changes in flooding indices calculated from reservoir future inflows relative to baseline period; (b) percentage changes in indices in flooding indices from reservoir inflow and release during baseline and future period; (c) percentage change in storage indices attributable to climate change.The semi-violin plots illustrate the distribution of percentage changes, while the semi-box plots indicate median values, as well as the first and third quartiles (depicted by upper and lower box hinges).Whiskers show the range of maximum and minimum values, excluding outliers.Individual dot points represent raw data for each reservoir.'n = 21' represents the number of selected reservoirs.

Figure 2 .
Figure 2. (a) Influence of hydrologic conditions, operation strategies, and evaporation on storage drought at Grapevine Lake, as evaluated by a structural equation model (SEM).Latent variables are represented by ellipses, while observed variables appear in rectangles.Standardized path values are displayed along each pathway; positive values are enclosed in black boxes and negative values are in red boxes.Paths with a statistically significant p-value below 0.05 are enclosed in a solid line box, while those with non-significant p-values are enclosed in a dashed line box.(b) Summary of SEM path values for select reservoirs investigated in this study.In the x-axis labels, 'H' represents hydrologic conditions, 'O' denotes operation strategies, and 'E' signifies evaporation.Statistically insignificant values are shown as not a number.

Figure 3 .
Figure3.Influential determinants of reservoir evaporation, segmented by month and informed by six GCMs, for the selected reservoirs examined in this study.Gray indicates that storage is the dominant factor affecting evaporation for the corresponding month, while red means that temperature plays a more significant role.The black solid line displayed on the secondary axis represents the multi-year average storage.

Figure 4 .
Figure 4. (a) Percentage change in water supply performance (reliability, resiliency, vulnerability) in different storage reallocation scenarios in Lake Waco.(b) Percentage change in downstream flood risk (duration, frequency, intensity) in different storage reallocation scenarios in Lake Waco.(c) Changes in the onset of storage recession in different storage reallocation scenarios in Lake Waco.The onset of storage recession is defined as the first day the reservoir's storage falls below the conservation level after a prolonged period of sustaining at this level, and this recession should last for at least 14 d.(d) Hydrograph of reservoir storage fluctuations between 1 May and 30 September 2012.'All Year' refers to a scenario involving year-round elevation of the conservation pool's upper limit.Labels from 'Jan' to 'Dec' indicate scenarios where the pool elevation is raised only during the corresponding month.'Sed' represents a condition where sedimentation decreases available storage.
and Pearl J 2013 Eight myths about causality and structural equation models Handbook of Causal Analysis for Social Research (Springer) pp 301-28 Caretta A M M A, Arfanuzzaman R B M, Morgan S M R and Kumar M 2022 Water Climate Change 2022: Impacts, Adaptation, and Vulnerability: Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change Carriere P E and Wurbs R A 1988 Evaluation of Storage Reallocation and Related Strategies for Optimizing Reservoir System Operations (Texas Water Resources Institute) Castle S L, Thomas B F, Reager J T, Rodell M, Swenson S C and Famiglietti J S 2014 Groundwater depletion during drought threatens future water security of the Colorado River Basin Geophys.Res.Lett.41 5904-11 Chase-Israel J 2022 Building new hydro at an existing water supply lake in Texas.Hydro.org(available at: www.hydro.org/powerhouse/article/building-new-hydro-at-an-existingwater-supply-lake-in-texas/) Cheng L, Hoerling M, Smith L and Eischeid J 2018 Diagnosing human-induced dynamic and thermodynamic drivers of extreme rainfall J. Clim.31 1029-51 Cosgrove W J and Loucks D P 2015 Water management: current and future challenges and research directions Water Resour.Res.51 4823-39 Croley T E and Raja Rao K N 1979 Multiobjective risks in reservoir operation Water Resour.Res. 15 807-14 Delaney C J et al 2020 Forecast informed reservoir operations using ensemble streamflow predictions for a multipurpose Schewe J et al 2014 Multimodel assessment of water scarcity under climate change Proc.Natl Acad.Sci. 111 3245-50 Texas Water Development Board, University of Texas at Arlington & National Integrated Drought Information System 2020 Forecast-informed reservoir operation (FIRO) and water resources management in Texas Texas Water Development Board 2021 2022 state water plan: water for Texas Wang F, Wang L, Zhou H, Saavedra Valeriano O C, Koike T and Li W 2012 Ensemble hydrological prediction-based real-time optimization of a multiobjective reservoir during flood season in a semiarid basin with global numerical weather predictions Water Resour.Res.48 Woodhouse C A, Gray S T and Meko D M 2006 Updated streamflow reconstructions for the Upper Colorado River basin Water Resour.Res.42 Wurbs R A 2021 Storage and regulation of river flows by dams and reservoirs Texas Water J. 12 10-39 Wurbs R A, Carriere P E and Johnson W K 1990 Management strategies for increasing reservoir yields Water Int. 15 168-75 Wurbs R A, Muttiah R S and Felden F 2005 Incorporation of climate change in water availability modeling J. Hydrol.Eng. 10 375-85 Yang T, Gao X, Sorooshian S and Li X 2016 Simulating California reservoir operation using the classification and regression-tree algorithm combined with a shuffled cross-validation scheme Water Resour.Res.52 1626-51 Yang T, Zhang L, Kim T, Hong Y, Zhang D and Peng Q 2021 A large-scale comparison of artificial intelligence and data mining (AI&DM) techniques in simulating reservoir releases over the Upper Colorado Region J. Hydrol.602 126723 Zarei M, Bozorg-Haddad O, Baghban S, Delpasand M, Goharian E and Loáiciga H A 2021 Machine-learning algorithms for forecast-informed reservoir operation (FIRO) to reduce flood damages Sci.Rep. 11 24295 Zhao G, Li Y, Zhou L and Gao H 2022 Evaporative water loss of 1.42 million global lakes Nat.Commun.13 3686 Zhao T, Zhao J, Lund J R and Yang D 2014 Optimal hedging rules for reservoir flood operation from forecast uncertainties J. Water Resour.Plan.Manage.140 04014041 Zhu J, Yang Y, Holmquist H and Leber N 2018 Projected reservoir rating curves based on sedimentation surveys and its application in water planning in Texas World Environmental and Water Resources Congress 2018 (American Society of Civil Engineers) pp 349-62