Volume 59, Issue 7 e2022WR034099
Research Article
Open Access

Estimating Future Surface Water Availability Through an Integrated Climate-Hydrology-Management Modeling Framework at a Basin Scale Under CMIP6 Scenarios

Manqing Shao

Manqing Shao

Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA

Contribution: Conceptualization, Methodology, Software, Validation, Formal analysis, ​Investigation, Data curation, Writing - original draft, Visualization

Search for more papers by this author
Nelun Fernando

Nelun Fernando

Water Availability Department, Surface Water Division, Texas Water Development Board, Austin, TX, USA

Contribution: Conceptualization, Software, ​Investigation, Resources, Writing - review & editing, Project administration

Search for more papers by this author
John Zhu

John Zhu

Water Availability Department, Surface Water Division, Texas Water Development Board, Austin, TX, USA

Contribution: Methodology, ​Investigation, Resources, Data curation, Writing - review & editing

Search for more papers by this author
Gang Zhao

Gang Zhao

Department of Global Ecology, Carnegie Institution for Science, Stanford, CA, USA

Contribution: Methodology, Software, Validation, Data curation, Writing - review & editing

Search for more papers by this author
Shih-Chieh Kao

Shih-Chieh Kao

Environmental Sciences Division and Climate Change Science Institute, Oak Ridge National Laboratory, Oak Ridge, TN, USA

Contribution: Methodology, Resources, Data curation, Writing - review & editing

Search for more papers by this author
Bingjie Zhao

Bingjie Zhao

Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA

Contribution: Writing - review & editing, Visualization

Search for more papers by this author
Elizabeth Roberts

Elizabeth Roberts

Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA

Contribution: Writing - review & editing

Search for more papers by this author
Huilin Gao

Corresponding Author

Huilin Gao

Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA

Correspondence to:

H. Gao,

[email protected]

Contribution: Conceptualization, Resources, Writing - review & editing, Supervision, Project administration, Funding acquisition

Search for more papers by this author
First published: 27 June 2023

Nelun Fernando and John Zhu—Unless specifically noted, this manuscript does not necessarily reflect official Board positions or decisions.

Abstract

Climate change and increasing water demand due to population growth pose serious threats to surface water availability. The biggest challenge in addressing these threats is the gap between climate science and water management practices. Local water planning often lacks the integration of climate change information, especially with regard to its impacts on surface water storage and evaporation as well as the associated uncertainties. Using Texas as an example, state and regional water planning relies on the use of reservoir “Firm Yield” (FY)—an important metric that quantifies surface water availability. However, this existing planning methodology does not account for the impacts of climate change on future inflows and on reservoir evaporation. To bridge this knowledge gap, an integrated climate-hydrology-management (CHM) modeling framework was developed, which is generally applicable to river basins with geographical, hydrological, and water right settings similar to those in Texas. The framework leverages the advantages of two modeling approaches—the Distributed Hydrology Soil Vegetation Model (DHSVM) and Water Availability Modeling (WAM). Additionally, the Double Bias Correction Constructed Analogues method is utilized to downscale and incorporate Coupled Model Intercomparison Project Phase 6 GCMs. Finally, the DHSVM simulated naturalized streamflow and reservoir evaporation rate are input to WAM to simulate reservoir FY. A new term—“Ratio of Firm Yield” (RFY)—is created to compare how much FY changes under different climate scenarios. The results indicate that climate change has a significant impact on surface water availability by increasing reservoir evaporation, altering the seasonal pattern of naturalized streamflow, and reducing FY.

Key Points

  • The assessment of future available surface water is improved by explicitly accounting for projected reservoir evaporation and naturalized streamflow

  • An open-water evaporation module was used to represent the enhanced reservoir evaporation loss in future climate scenarios

  • An ensemble of future reservoir Firm Yield (FY) values under different climate change scenarios is generated for supporting water supply risk management decisions

Plain Language Summary

Under conditions of climate change, it is important to have actionable climate information for local water management and surface water planning. By investigating the impacts of climate change on surface water availability over a representative metropolitan area (and its water supply reservoirs), this research helps to bridge the gap between climate science and the necessary information for sound water management decisions. First, the coarse resolution information from global climate models was translated into the finer spatial scales suitable for application into the long-range water supply planning tools commonly used in Texas. Second, a climate-hydrology-management (CHM) modeling framework was developed to simulate future surface water availability under different climate scenarios. The results indicate that the changes in reservoir evaporation rates and streamflow patterns play an important role in reducing future surface water availability. The projected hydrological and water availability information can guide decisions related to local and long-range water planning. The study framework is also applicable to other river basins for improving the estimation of future water availability under climate change.

1 Introduction

Water availability—which is essential to human society, and to the natural environment—will be significantly influenced by global warming (Greve et al., 2018; Konapala et al., 2020). Climate change can alter hydrologic processes and lead to increases in surface temperature and evaporation, and to changes in precipitation and streamflow (Hostetler & Bartlein, 1990; Houghton et al., 2001; Huntington, 2006; Khan, 2022). On one hand, climate change can exacerbate water security by diminishing water supplies (Konapala et al., 2020; Kundzewicz et al., 2008). On the other hand, water demand continues to increase due to the global population growth (United Nations, 2019). Therefore, it is imperative to improve upon the estimation of future surface water availability under a changing climate (He et al., 2021; IPCC, 2022; Konapala et al., 2020; Soundharajan et al., 2016; Vorosmarty et al., 2000).

Our study can provide insights about evaluating surface water availability in other urban areas globally. The number of urban dwellers facing water scarcity is expected to rise from 933 million (which is about one-third of the total urban population in 2016) to at most 2.373 billion people—representing nearly half of the global urban population—by 2050 (He et al., 2021). The situation could be especially dire in cities that rely primarily on surface water resources, which are inherently more susceptible to the consequences of climate change. Examples of such cities include Bangalore in India, Melbourne in Australia, Shanghai in China, and Los Angeles in the U.S. (Hegde & Chandra, 2012; Li et al., 2017; Pincetl et al., 2019; Timbal et al., 2015). Additionally, urban areas in Texas are facing significant challenges with water availability due to factors like a semi-arid climate, rapid population growth, and increased vulnerability to water availability variations under climate change (Nielsen-Gammon et al., 2020).

In Texas, the total water supply is around 8.9 million acre-feet per year, nearly two-thirds of which is from surface water resources (Texas Water Development Board, 2022). Most of the Texas surface water supply relies on the 114 major reservoirs that are spread across 23 river basins. The Trinity River Basin alone contributes more than 20% of the state's non-agricultural water supply (Texas Water Development Board, 2022). However, the storage of these reservoirs is expected to be significantly impacted by future alterations in precipitation, temperature, and other climatic variables due to a changing climate. Furthermore, with an estimated population growth of 52 million people by 2070 mainly occurring in urban regions such as the Dallas-Fort Worth (DFW) metroplex, accurately evaluating future water availability in the major reservoirs is imperative (Texas Water Development Board, 2022).

Reservoir storage is affected by two important factors: reservoir evaporation and stream inflow. Over the past few decades, reservoir evaporation has increased substantially (G. Zhao et al., 201820212022). Meanwhile, changes in seasonal precipitation due to climate change can result in significant variations in reservoir inflows (Konapala et al., 2020). In Texas, the average annual gross evaporation from 114 major reservoirs during 2001–2018 was 6.88 million acre-feet—and the annual evaporative losses during drought years, such as 2011, can surpass the annual municipal water use for the given year (Zhu et al., 2021). Understanding the impacts of reservoir evaporation on conservation storage—which can provide multi-purpose water supply—is critical when planning to address the potential future shortages from surface water resources.

With regard to the status-quo method for surface water planning, one major limitation is that the current water planning practice is normally based on historical hydrological data and current water availability by water management decision makers (such as those in California and Texas). Long-range water planning (i.e., 50 years into the future) in Texas currently uses the drought of record (e.g., the 1950s drought) as the benchmark drought event when estimating water supply shortages over the next five decades. However, the current planning methodology does not consider climate change, nor its impacts on naturalized streamflow and reservoir evaporation—both of which are critical elements of the water balance for surface water reservoirs (Texas Water Development Board, 2022). Future surface water availability for a given reservoir is expressed as a single reservoir Firm Yield (FY) estimate, without an associated uncertainty range assigned to the yield estimate.

It is challenging to consider climate change impacts in the water planning process, as knowledge gaps exist between climate change and water planning/management practices (Brekke et al., 2009; Orr et al., 2022). One of the gaps is related to the lack of reliable information co-produced through the engagement of all parties (e.g., governments, climate scientists, operational sectors). Another barrier is the need for translating coarse resolution climate information generated from Global Climate Models (also known as General Circulation Models, GCMs) to the fine spatial scale hydrologic variables necessary for supporting local water management.

In Texas, the key tool used for water planning is the Water Availability Modeling (WAM) system, which is a water accounting system that incorporates water rights information for Texas. In WAM, “Firm Yield” (FY) is a fundamental metric used to quantify surface water availability with respect to reservoirs, and has been applied widely for water planning purposes (Archfield & Vogel, 2005; Zhu et al., 2018). FY is used as the definition of surface water availability in this study. Although WAM encompasses different water users and can be directly applied to the decision-making process, it has not been able to adopt future surface water availability projections due to a lack of hydrological projections under climate change. To resolve this limitation, a fully distributed and high-resolution hydrological model can be used in conjunction with WAM to incorporate future hydrological information into the estimation of FY. One advantage of such models is the ability to simulate hydrological processes over spatially heterogeneous land cover, which can facilitate future projections of surface water availability over metropolitan areas. Additionally, the fine resolution hydrological models are normally driven by downscaled GCM forcings to simulate the spatially heterogeneous hydrological processes. Statistical downscaling methods are commonly employed due to their lower computational requirements and ease of implementation (Lanzante et al., 2018).

The lack of a reliable method for estimating reservoir evaporation when evaluating surface water availability is another challenge (Dai, 2016; Ito & Momii, 2021). The conventional approach is based on data collected from pan evaporation stations. For example, the gross lake evaporation data used in WAM for Texas river basins is calculated by applying the corresponding pan-to-lake coefficient to the evaporation data from a given pan (Texas Water Development Board, 2022; Zhu et al., 2021). However, this method has large errors, as only a few pan evaporation stations are typically located near a dam (G. Zhao & Gao, 2019). Furthermore, the 10-in deep water in evaporation pans cannot account for factors such as the heat storage and wind fetch effects (McMahon et al., 2013; G. Zhao & Gao, 2019). To address this issue, Finch (2001) developed a method that considers the heat storage effect and accurately quantifies reservoir evaporation losses. This method has been applied in a study that generated a long-term evaporation data set for 721 U.S. reservoirs (G. Zhao & Gao, 2019), and another that evaluated future projections of 678 major reservoirs over the CONUS under climate change (B. Zhao et al., 2023).

The objective of this study is to bridge the gaps between climate science and water management practices by developing a new modeling framework with an improved assessment of future available surface water under climate change at the basin scale. Through the integrated climate-hydrology-management (CHM) modeling framework, the advantages of two modeling approaches—the fully distributed, high-resolution Distributed Hydrology Soil Vegetation Model (DHSVM) model, and the WAM system—were leveraged. Specifically, statistically downscaled Coupled Model Intercomparison Project Phase 6 (CMIP6) climate projections were used to drive the models to simulate hydrological data and reservoir FY. A representative U.S. metropolitan area in Texas was used to demonstrate how future surface water availability is estimated in terms of reservoir FY through the modeling framework.

2 Study Region

The study region is the Upper Trinity River Basin (Figure 1), which provides a large amount of water supply to the Region C Regional Water Planning Area in Texas (Texas Water Development Board, 2020). The DFW metropolitan area experienced the largest population gain from 2010 through 2019 among all metropolitan areas in the U.S., with an increase of 1,206,599 people (19.0%) (United States Census Bureau, 2020).

Details are in the caption following the image

The map of the study region and basic information about the reservoirs. The 24,000 km2 study domain includes the Dallas-Fort Worth metropolitan area, six reservoirs, and four control points.

Six reservoirs around the metropolitan area were selected as the target reservoirs—Lake Bridgeport, Eagle Mountain Lake, Lake Worth, Grapevine Lake, Lake Ray Roberts, and Lewisville Lake (Figure 1). Basic information about the reservoirs is shown in Figure 1. Lake Bridgeport, Eagle Mountain Lake, and Lake Worth are located on the West Fork of the Trinity River, and they operate as a system for local water supply, water storage, flood control, and recreation. Grapevine Lake is located along the Trinity River floodplain and provides water supply to Dallas (via Dallas Water Utilities) and to two residential areas in central Dallas County. Lake Ray Roberts and Lewisville Lake make up the primary water supply for the cities of Dallas and Denton, and the surrounding communities. In addition, four WAM control points located in the main river channels are selected.

3 Methodology and Data

3.1 Models and Data

3.1.1 Downscaled CMIP6 Climate Projections

The downscaled climate projections were derived from nine selected CMIP6 GCMs (Eyring et al., 2016) under the Shared Socioeconomic Pathways-Representative Concentration Pathway (SSP-RCP, van Vuuren et al., 2014) emission scenarios. Specifically, two standard scenarios—SSP245 and SSP585—were selected. These two scenarios are commonly used for climate change studies because they represent two contrasting possibilities (Balu et al., 2023; Reid et al., 2021). SSP245 represents a future in which there is some progress toward mitigating greenhouse gas emissions and increasing energy efficiency, while SSP585 represents a continuation of the current trends in economic and population growth—leading to a future with high greenhouse gas emissions and high radiative forcing.

Overall, nine GCMs (Table 1) were selected based on data availability (when the research started). The selection was also based on how well the GCMs reproduced drought characteristics during the historical baseline period (Text S1 in Supporting Information S1). The Palmer Drought Severity Index generated from Penman-Monteith potential evapotranspiration calculations (PDSI-PM; Sheffield et al., 2012) was derived during the historical period from the nine GCMs. By comparing the cumulative probabilities between the GCM-based PDSI-PM and the observation-based PDSI-PM during the historical period (1950–2018), the nine GCMs with the best performance were selected.

Table 1. Basic Information About the Nine Coupled Model Intercomparison Project Phase 6 Global Climate Models
Number Model name Resolution (longitude/latitude) Variant id Institution
1 ACCESS-CM2 1.87°/1.25° r1i1p1f1 CSIRO-ARCCSS
2 BCC-CSM2-MR 1.12°/1.12° r1i1p1f1 Beijing Climate Center
3 CanESM5 2.8°/2.8° r1i1p1f1 Canadian Centre for Climate Modelling and Analysis
4 CNRM-ESM2-1 1.4°/1.4° r1i1p1f2 CNRM-CERFACS
5 EC-Earth3 0.7°/0.7° r1i1p1f1 EC-Earth Consortium
6 MPI-ESM1-2-HR 0.94°/0.94° r1i1p1f1 Max Planck Institute for Meteorology, Deutsches Klimarechenzentrum and Deutscher Wetterdienst
7 MRI-ESM2-0 1.12°/1.12° r1i1p1f1 Meteorological Research Institute
8 NorESM2-MM 1.25°/0.94° r1i1p1f1 Norwegian Climate Center
9 UKESM1-0-LL 1.9°/1.2° r1i1p1f2 Met Office Hadley Center and Natural Environment Research Council

To assess the climate change impacts on surface water availability over the study region, appropriate downscaling techniques are needed to translate the coarse resolution information from the GCMs to fine spatial scales (Ashfaq et al., 2016; Maurer et al., 2013). Here, the Double Bias Correction Constructed Analogues (DBCCA) method (Werner & Cannon, 2016) was utilized to downscale daily precipitation, maximum temperature, minimum temperature, and average wind speed from the nine selected GCMs. DBCCA is based on the widely used Bias Correction Constructed Analogues (BCCA) method developed by Maurer et al. (2010). It was found to be one of the two best-performing methods among a group of popular statistical downscaling approaches (Werner & Cannon, 2016), and (additionally) it was used in a recent CMIP6 downscaling effort over the contiguous U.S. (Rastogi et al., 2022).

We adopted the same DBCCA setup employed by Rastogi et al. (2022) and used the 1/16° grid (∼6 km) (as per Livneh et al., 2013) as the reference meteorological observation data set for the downscaling. The downscaling was conducted for the entire period of 1950–2099, in which 1950–2018 was treated as the historical baseline for model verification and 2019–2099 was treated as the projected future period for water resources assessment. After downscaling, the Mountain Microclimate Simulation Model (MTCLIM; Thornton & Running, 1999) version 4.2 was used to generate the full set of DHSVM meteorological forcings (i.e., air temperature, wind speed, relative humidity, incoming shortwave radiation, incoming longwave radiation, and precipitation).

3.1.2 Distributed Hydrology Soil Vegetation Model (DHSVM)

The DHSVM model (Wigmosta et al., 1994) is an open-source, physics-based, fully distributed model that computes the energy and water balances at each grid cell. The model can be set up at a high spatial resolution (e.g., 10–200 m) and at a sub-daily time step (e.g., 1–24 hr). In this study, the calibrated soil and vegetation parameters were adopted from G. Zhao et al. (2018) for the Trinity River Basin at 200 m spatial resolution. The simulated streamflow agrees well with observations at the selected USGS gages, with R2 values ranging from 0.70 to 0.85 and NSEs ranging from 0.43 to 0.84. The calibrated model also performs well for lake water elevation, with R2 values ranging from 0.60 to 0.95 and the NSEs ranging from 0.46 to 0.95.

There are two DHSVM components which are important for this study: (a) An explicit, high-resolution urban module (Cuo et al., 2008) simulates the spatial heterogeneity of the urban sprawl (e.g., more urbanization along the highway) for the metropolitan area (Shao et al., 2020); and (b) An open water evaporation submodule. DHSVM accounts for the heat storage effect (Edinger et al., 1968; McMahon et al., 2013; G. Zhao & Gao, 2019) and incorporates a generally applicable wind function (McJannet et al., 2012). As a result, the model is capable of calculating reservoir evaporation rates using standard land-based meteorological data.

3.1.3 Water Availability Model (WAM)

The WAM model is used for predicting the amount of water that would be in a river or stream under a specified set of conditions, and it has been configured for each individual Texas river basin (R. A. Wurbs et al., 2005). The Texas WAM system was implemented from 1997 to 2004 following comprehensive water management legislation enacted by the Texas legislature in 1997 (Sokulsky et al., 1998; R. A. Wurbs, 2001). WAM serves the purpose of authorizing water permits and calculating water availability, while conforming to the prior appropriations doctrine (Texas Water Code, 2022). It contains the Water Rights Analysis Package (WRAP) executable programs, WRAP input files for 23 river basins in Texas, a geographic information system, and other tools and databases. WRAP was developed for the main purpose of assessing the capabilities of reservoir/river systems to meet specified requirements from water management and regulation (R. Wurbs, 2019). WRAP simulations involves three steps: First, the sequences of monthly naturalized streamflows and reservoir net evaporation rates are used to represent river basin hydrology. Second, water allocation is performed—with water being allocated to each water right based on the prior appropriations doctrine. Lastly, the simulation results and the summary statistics are organized and calculated. A detailed description of WRAP input data is provided in Text S2 in Supporting Information S1.

3.1.4 Firm Yield (FY)

FY is defined as the maximum annual diversion from a reservoir or a reservoir system that is available at 100% reliability during an extreme drought. It is a fundamental and deterministic target for long-term water planning in Texas (Archfield & Vogel, 2005; Nielsen-Gammon et al., 2020; Zhu et al., 2018). FY is normally regarded as the worst-case scenario for local water supply, and can decrease when severe droughts (as compared to previous ones) occur (Nielsen-Gammon et al., 2020). Specifically, FY in WAM refers to the simulated maximum annual diversion volume from a reservoir (or a reservoir system) that can be supplied without a shortage during a hydrologic period-of-analysis simulation (R. Wurbs, 2019).

A new term, “Ratio of Firm Yield” (RFY), is introduced in this study to quantify the change of reservoir FY in a consistent manner. RFY is defined as the simulated FY in the future period divided by the FY in the historical period (Equation 1):
urn:x-wiley:00431397:media:wrcr26721:wrcr26721-math-0001(1)
where the Historical Period represents 1980–2019, Period 1 represents 2020–2059, and Period 2 represents 2060–2099. These same three periods are also used in the following sections.

3.2 Study Design

This study presents a CHM modeling framework (Figure 2) for estimating future surface water availability under climate change scenarios. In this framework, DHSVM has a unique advantage of simulating future naturalized streamflow and reservoir evaporation rates under climate change scenarios. These DHSVM simulation results are then used to drive the WAM model for future reservoir FY estimates, which are essential for long-term water management planning.

Details are in the caption following the image

The flowchart of the climate-hydrology-management (CHM) modeling framework.

The study design encompasses a trend analysis followed by the CHM modeling, which is described below in detail.

First, the trends of the multi-model ensemble means for the four climatic variables (temperature, vapor pressure deficit [VPD], precipitation, and wind speed) during the period of 2020–2099 were evaluated using the Mann-Kendall test and Sen's slope method. The significance level, α (alpha), of the Mann-Kendall test was defined as 0.05 (Kendall, 1975; Mann, 1945). The slope of the time series data (i.e., change per unit time) was predicted using Sen's slope estimator (Sen, 1968). The Mann-Kendall test was used to determine whether the time series had monotonic trends. A positive or negative value of Sen's slope indicated an upward or downward trend.

Second, DHSVM was forced by a series of DBCCA downscaled climate projections to simulate future naturalized streamflow and reservoir evaporation rates. The naturalized streamflow was obtained by running DHSVM with the parameters from G. Zhao et al. (2018) but without using the reservoir module. The reservoir net evaporation rate was obtained by subtracting the average precipitation rate from the DSHVM simulated reservoir gross evaporation rate.

Third, the seasonal statistical analysis was conducted using the simulated monthly naturalized streamflow at the four WAM control point locations (Figure 1). For each season in each of the two future periods, the 25th, median, and 75th percentile values of the monthly naturalized streamflow were calculated from the nine GCMs under SSP245 and SSP585, respectively. The differences of these values between Period 1 and Period 2 were calculated under each climate change scenario.

Fourth, the DHSVM simulated naturalized streamflow forced by the downscaled GCMs was bias-corrected at each WAM control point using the Quantile Mapping method (Wood et al., 2002; Wood & Lettenmaier, 2006). Specifically, biases were calculated at each percentile in the cumulative distribution function between the WAM monthly naturalized streamflow and the DHSVM simulated naturalized streamflow using the GCMs during the historical baseline period. These biases were then applied to correct the DHSVM simulated monthly naturalized streamflow during the future periods.

Finally, the adjusted projected hydrological data (i.e., naturalized streamflow and reservoir net evaporation rate) from the previous steps were adopted as input for WAM to facilitate the FY simulations. The FY was simulated for the four individual reservoirs—Lake Bridgeport, Eagle Mountain Lake, Lake Worth, and Grapevine Lake—and a reservoir system—the Lake Ray Roberts-Lake Lewisville system using each GCM and climate change scenario during the Historical Period, Period 1, and Period 2, respectively. The RFY was also calculated for each GCM and each reservoir/reservoir system under two climate scenarios.

4 Results

4.1 Downscaled CMIP6 Climate Projections

In this section, the time series of four climatic variables—temperature, VPD, precipitation, and wind speed—are compared under SSP245 and SSP585 from 2020 to 2099. For each climatic variable, the trends for the mean values of the annual multi-model ensembles, and the associated uncertainty ranges, are illustrated in Figure 3. The statistical results are summarized in Table 2. In general, the time series of the annual multi-model ensemble means of temperature and VPD have statistically significant increasing trends under both SSP245 and SSP585 scenarios. Specifically, the annual multi-model ensemble means of temperature are expected to increase by 0.036° per year under SSP245, and 0.082° per year under SSP585. In contrast, the annual multi-model ensemble mean of wind speed has only a slightly increasing trend—0.001 m per second per year—under both scenarios. Of the four variables, only precipitation does not show a significant trend under both scenarios.

Details are in the caption following the image

The annual multi-model ensemble time series and uncertainty ranges of four climate variables from the nine Coupled Model Intercomparison Project Phase 6 GCMs from 2020 to 2099 under future scenarios: (a) air temperature under SSP245; (b) vapor pressure deficit (VPD) under SSP245; (c) precipitation under SSP245; (d) wind speed under SSP245; (e) air temperature under SSP585; (f) VPD under SSP585; (g) precipitation under SSP585; (h) wind speed under SSP585. The results are averaged over the study region, with the ensemble means in orange and the trends in dotted-lines.

Table 2. Results of the Mann-Kendall Trend Tests for Four Climate Variables Under SSP245 and SSP585
Scenario Variables Units P-value (two-tailed) Alpha Test interpretation Sen's slope
SSP245 Temperature °C <0.0001 0.05 Reject H0 0.036
VPD Pa <0.0001 Reject H0 3.716
Precipitation mm/year 0.633 Accept H0 0.006
Wind Speed m/s 0.009 Reject H0 0.001
SSP585 Temperature °C <0.0001 Reject H0 0.082
VPD Pa <0.0001 Reject H0 8.847
Precipitation mm/year 0.627 Accept H0 −0.006
Wind Speed m/s <0.0001 Reject H0 0.001

4.2 Projected Hydrological Data

4.2.1 Projected Naturalized Streamflow

In this section, the naturalized streamflows (averaged by season) at the four control point locations are compared between the two future periods. At each control point, the differences between Period 1 and Period 2 are illustrated for all four seasons at the 25th, median, and 75th percentiles of the multi-model ensemble projections under the SSP245 and SSP585 scenarios (Figure 4).

Details are in the caption following the image

The differences of the simulated monthly naturalized streamflow between Period 1 and Period 2 during the four seasons at the 25th, median, and 75th percentiles at the four Water Availability Modeling streamflow control points: (a) 8WTJA; (b) 8WTBO; (c) 8ETLA; and (d) 8ELLE.

The results show that the differences are positive for 79% of the cases, indicating that the simulated naturalized streamflows in Period 1 are projected to be larger than those in Period 2. Most of the differences are found in autumn. Under SSP245, the naturalized streamflows in autumn are projected to be larger in Period 1 than in Period 2 in most cases. Under SSP585, the naturalized streamflows in the spring are projected to be larger in Period 1 than in Period 2 in most cases. At each control point and climate change scenario, the absolute values of the differences at the 75th percentile are normally larger than those at the 25th percentile and the median. In conclusion, the seasonal patterns of the naturalized streamflow are projected to be different during the two future periods, with varying degrees of discrepancy under the two scenarios.

4.2.2 Projected Reservoir Evaporation Rate

In this section, the simulated reservoir evaporation rates are compared among the target reservoirs. Figure 5 shows the simulated annual average gross and net evaporation rates, which are projected to increase from Period 1 to Period 2 under both scenarios. In general, both the gross and net evaporation rates under SSP585 are larger than those under SSP245 in the two future periods. This suggests that climate change can impose significant effects on boosting open water evaporation through increasing air temperature, VPD, and wind speed (see Figure 5).

Details are in the caption following the image

Bar plots of the multi-model ensemble medians of the annual average (a) gross evaporation rates and (b) net evaporation rates for each reservoir. Each colored bar represents the median evaporation rate under different emission scenarios in different periods. The uncertainty range represents 25%–75%.

The medians of the projected gross evaporation rates over Lake Bridgeport, Eagle Mountain Lake, and Lake Worth are in the range of 3.73–3.87 mm/day from 2020 to 2099 under both scenarios. However, the projected gross evaporation rates over Grapevine Lake, Ray Roberts Lake, and Lewisville Lake are in the range of 3.65–3.75 mm/day (again from 2020 to 2099, under both scenarios), which is smaller than that of the first three reservoirs. The projected net evaporation rates share the same pattern across these reservoirs. The three reservoirs with larger evaporation rates are located on the western side of the study domain, where the air is drier and the precipitation level is smaller. We also found that the uncertainty ranges of the net evaporation rates are larger than those of the gross evaporation rates. This may be caused by the large uncertainties associated with the precipitation projections in the climate models, as the net evaporation rate is defined as the gross evaporation rate minus the precipitation rate.

4.3 Projected Firm Yield (FY)

In this section, the WAM simulated FY during the two future periods are compared with the historical values under SSP245 and SSP585. The results for the four reservoirs, and the Ray Roberts Lake-Lewisville Lake system, are illustrated in Figure 6.

Details are in the caption following the image

Comparisons between the projected Firm Yield (FY) and the historical FY at (a) Lake Bridgeport; (b) Eagle Mountain Lake; (c) Lake Worth; (d) Grapevine Lake; and (e) the Ray Roberts Lake-Lewisville Lake system. (f) The total FY for all of the reservoirs and the reservoir system from (a) to (e).

In general, the FY values of all reservoirs are projected to decrease, with more changes in Period 2 than in Period 1. For example, at Lake Bridgeport, the median value of FY under SSP245 is 46,000 ac-ft/year (5.7 × 107 m3/year) and 45,800 ac-ft/year (5.6 × 107 m3/year) in Period 1 and Period 2, respectively. Meanwhile, the projected FY values under SSP585 are always smaller than those under SSP245. Across the entire system of the six reservoirs, the multi-model ensemble median of the total FY is 13.6% and 16.8% smaller than the baseline period values under SSP245 and SSP585, respectively. However, the changes at Grapevine Lake are smaller than the changes at the other reservoirs.

The RFY from the nine GCMs are illustrated for all of the reservoirs in Figure 7. Overall, the RFY under SSP585 is smaller than that under SSP245 for most GCMs in the same future period. For example, at Lake Bridgeport, the RFY ensemble mean is 83% under SSP245 and 67% under SSP585 in Period 2. At Lake Worth, the RFY for Period 1 under SSP245 (projected by the UKESM1-0-LL) is shown as “no data” because the simulation could not reach FY—meaning that a shortage will always exist in the water supply under this projection. Also, the RFY values projected by some GCMs are very small at Lake Worth, indicating that this lake will suffer from severe water shortages in the future.

Details are in the caption following the image

Ratio of Firm Yield values by reservoir/reservoir system from each General Circulation Model under four scenarios: (a) SSP245 in Period 1; (b) SSP245 in Period 2; (c) SSP585 in Period 1; and (d) SSP585 in Period 2. Each number to the right is the average of its corresponding row, while each value on the top is the average of the corresponding column.

While the uncertainties associated with FY and RFY vary among different reservoirs, they are always larger under SSP585 than under SSP245 in both future periods. There are three possible reasons for this: First, SSP585 assumes a more fossil-fuel-intensive future with higher greenhouse gas emissions than SSP245, which leads to greater uncertainty in the climate response. Second, there could be more extreme events under SSP585 than under SSP245. Third, SSP585 assumes a society with greater variability in socio-economic development, which can lead to larger uncertainties in the projections of climate change impacts.

5 Discussion

The most important contribution of our research is that an integrated framework of hydrology and water resources management is provided to guide water management planning under climate change. While we agree that many papers have explored how to leverage watershed-scale climate projections to guide water resources management, existing literature sources are typically focused on either the impacts on natural water availability or water management (but not both). For example, Matonse et al. (2013) used future climate projections as inputs into the Generalized Watershed Loading Functions—Variable Source Area (GWLF-VSA) watershed model to simulate future inflows into local water supply reservoirs. Pumo et al. (2017) used the physics-based and spatial distributed hydrological model, the TIN-based Real-time Integrated Basin Simulator to investigate the impacts of climate change and urbanization on watershed hydrology at a river basin in Oklahoma, USA. Forero-Ortiz et al. (2020) used the Hydrologiska Byråns Vattenbalansavdelning hydrological model to evaluate potential future droughts under climate change in Barcelona City, Spain.

Some recent studies have started to use both hydrological and water management models to investigate the impacts of climate change. For example, Muttiah and Wurbs (2002) used the Soil and Water Assessment Tool (SWAT) and WRAP models to investigate the impacts of climate change on water supply reliability in the San Jacinto River Basin, Texas. Khoi et al. (2021) studied the impacts of climate change on water availability in the upper Dong Nai River Basin using the SWAT and Water Evaluation and Planning models. However, the SWAT model is a semi-distributed hydrological model that does not consider the physical process of reservoir evaporation—which is of great importance in water resources management, but is often overlooked in the current modeling communities. It is challenging to reliably estimate reservoir evaporation, even though freshwater loss through reservoir evaporation is substantial and increasing. Therefore, our work—which demonstrates an integrated framework of hydrology and water resources management practices—stands out in the current literature, and is a useful example for future studies to better guide water management planning under climate change.

While the findings and conclusions of this research are based upon a specific metropolitan area, the insights and methods can be generalizable to other places with comparable social and environmental contexts (such as semi-arid or arid regions with reservoirs as one of their major surface water resources). There are three key insights that can be used to other places: (a) It explicitly accounts for future open water evaporation and future naturalized streamflow, which is critical to accurately assessing future surface water availability; (b) It adopts an open water evaporation module that can be implemented into other hydrological models and/or water management models; and (c) It generates a range of future water availability values (i.e., FY) under different scenarios using an ensemble of GCMs, which supports water planning under uncertainties. Our approach should be applicable to multiple western U.S. states with similar geographical, hydrological, and water rights conditions—such as California, Colorado, New Mexico, Nevada, Arizona, and Nebraska. In particular, multiple western U.S. states also use prior appropriation doctrines like Texas (Sea Grant, 2023). Considering the increasing evidence that the southwest U.S. is expected to become warmer and drier in the future (USGCRP et al., 2014), it is important to consider climate change-induced uncertainties in long-term water resources planning. Similarly, in other places across the world—like Chile, some parts of Australia, and south Africa—water rights are also allocated based on the prior appropriation doctrine, and the open water evaporation rate is relatively high (Donoso, 2018; Godden, 2005; Grafton & Horne, 2014). Therefore, our modeling framework—which includes both a WRAP based on prior appropriation doctrine, and a fully distributed hydrological model with an advanced reservoir evaporation module—is broadly applicable, and can provide best available estimates of projected future surface water availability at a local scale.

Both the DHSVM model and the reservoir evaporation module are publicly available. While the WAM model was developed for Texas, a region-specific water management model can always be adopted to replace WAM in other applications. Also, other places can incorporate future projections of the relevant reservoir evaporation rate into their local or regional water resources management framework, and further improve the water resources management component (such as the rational allocation of local water resources, the construction of new reservoirs, and the management and operation of reservoirs; B. Zhao et al., 2023). Additionally, our framework can be used to project evaporation rates and FY values under other climate scenarios, such as SSP126 and SSP370—which could aid policymakers in developing environmental protection policies.

Even though there are already WAM techniques in some of the above areas, our methods can help to improve the existing modeling approach to one that can more accurately simulate the local water availability. For example, WaterSim is a water balance model that assesses the available water conditions of a region or watershed, and has been applied to the Phoenix Metropolitan Area (Arizona) and 25 other regions within the seven states associated with the Colorado River Basin—Arizona, California, Colorado, Nevada, New Mexico, Utah, and Wyoming (Gober et al., 2011). As there is no specific calculation of open water evaporation in WaterSim, our evaporation calculation method can be applied to WaterSim to improve the estimation of reservoir evaporation.

Our findings are valuable for future water management planning, especially for arid/semi-arid regions. With increasing water demand (due to a fast-growing population) and more extreme drought events (due to climate change), water supply systems at local or regional scales will face significant challenges in the future (Dallison et al., 2021). Our research highlights the impacts of increasing reservoir evaporation on surface water availability, which can exacerbate existing water shortages. To address this issue, we suggest that incorporating future projections of reservoir evaporation into other local or regional water management models can lead to improved long-range water planning across the world.

One significant contribution of our study is improving the estimation of reservoir-specific evaporation rates by considering the heat storage effect and fetch effects. In Texas, the reservoir evaporation rate is currently calculated for 1° × 1° grid cells (“quadrangles”), which provides a single gross evaluation rate value for each month based on the pan evaporation and a monthly pan-to-lake coefficient for each “quadrangle” (Texas Water Development Board, 2023). However, this method does not represent the true “over-water” evaporation rate, and does not have an associated uncertainty range. Our study provides an improved method that calculates the associated uncertainty range of the reservoir evaporation rate for each studied reservoir, under different climate scenarios within two future periods (i.e., SSP245 and SSP585, near-future and far-future). Given the fine spatial resolution adopted in the study, the reservoir evaporation rates are suitable for application at local scales (i.e., the county scale or an urban study area in semi-arid regions). Studies have found that the improved calculation of reservoir evaporation is necessary to be accounted for in the water supply systems (Vano et al., 2010; VanRheenen et al., 2011; Wiley & Palmer, 2008).

One of our findings is that the FY values of all studied reservoirs are projected to decrease from historical to future periods under both scenarios, which is mainly driven by the rising trends of temperature, VPD, and wind speed in the study region. Considering the projected increase in average global temperatures in the future (IPCC, 2013), it is likely that the FY of reservoirs in other areas will also be affected. Additionally, studies project that droughts with greater intensities and prolonged durations will become more frequent under climate change, and the number of extended droughts will also increase dramatically (Huang et al., 2022). Moreover, the FY of reservoirs in other regions around the world could be further impacted by the projected increase in global water demand for all purposes, which is expected to rise from 4,600 km3 per year currently to approximately 5,500–6,000 km3 per year by 2050 (Boretti & Rosa, 2019). Therefore, our research framework will provide valuable insights for future studies that aim to quantify the projected FY of reservoirs in other regions. Our study also suggests that the far future (i.e., Period 2) will experience a larger surface water deficit than the near future, especially under SSP585. These results are in line with the findings by Touseef et al. (2021), which predict potential water shortages after 2050 over the Hongshui River Basin in China, and are more likely to happen under the high climate change emission scenario.

There are over 50 GCMs in the latest CMIP6, and the resolution of most is insufficient to reliably assess the climate change impacts at policy-relevant regional and local scales (Ashfaq et al., 2022). Thus, an appropriate downscaling method—along with the use of a sub-selection of GCMs from the large CMIP ensemble—are important for climate change related studies at the local scale. The DBCCA method was selected as it was found to be one of the two best-performing downscaling methods among a few popular statistical approaches (Werner & Cannon, 2016). Although only nine GCMs were used (due to the limited number of data resources when the research started), it has been proven that these nine GCMs can adequately represent the historical drought patterns, and thus are suitable for the projection of future surface water availability. Also, it is crucial for studies about climate change impacts to understand the uncertainties involved with climate projections, which arise from many factors—such as GCM structure, the choice of which observations to use, and the downscaling technique (Lopez et al., 2009; Rastogi et al., 2022).

Even though our work can deliver useful information for local water resources management and regional water planning purposes, some limitations need to be addressed in future studies:
  1. Our research focuses on a semi-arid metropolitan area in Texas where reservoirs serve as the primarily source of water supply. Our findings can be applied to metropolitan areas, or other non-urban areas with arid/semi-arid climate. However, for regions with large surface water-groundwater interactions, a groundwater accounting model would need to be integrated into our framework.

  2. The WAM we employed is based on a priority water rights system. Therefore, if we were to apply our modeling framework to other locations, it is important to account for the specific water rights doctrines that are relevant in those places, and to adjust the water allocation scheme accordingly.

  3. The WAM we used did not take into account potential modifications in future water management approaches. It is recommended that future studies incorporate different water management scenarios in their water availability models. This will enable more effective adjustments to water policies under varying climate change scenarios.

  4. Reservoir depth is a crucial factor in accurately assessing the impact of heat storage. Generally, there is a linear relationship between depth and heat storage when the depth is less than 20 m. However, for this study, a fixed depth was assigned to each reservoir, which may result in some inaccuracies when estimating heat storage.

  5. The advective heat flux is an important component of the reservoir heat balance (Friedrich et al., 2018). However, this term was not included in the calculation of reservoir evaporation in this study due to the unavailability of related data.

6 Conclusions

Driven by fast population growth and economic development, global freshwater withdrawals have increased over three-fold from 1.23 trillion m3 in 1950 to 3.99 trillion m3 in 2014 (Roser & Rodés-Guirao, 2013; United Nations, 2019). Some metropolitan areas may experience surging immigration, leading to higher demand for freshwater resources. Meanwhile, water supply shortage could be further exacerbated by a higher open water evaporation rate under warming (B. Zhao et al., 2023).

In this study, an integrated CHM modeling framework was developed, which is generally applicable to river basins with geographical, hydrological, and water right settings similar to those in Texas. The framework leverages the advantages of the DHSVM and WAM models. Additionally, the DBCCA method was utilized to downscale future climate forcings from nine CMIP6 GCMs for driving the DHSVM model. A range of future reservoir FY and RFY values were projected under two emission scenarios for guiding water risk management decisions. Our major findings include:
  1. The Upper Trinity River Basin is projected to experience rising trends in terms of temperature and VPD under both the SSP245 and SSP585 scenarios. The time series of wind speed only has a slightly increasing trend, and there are no obvious trends in precipitation.

  2. The seasonal pattern of the naturalized streamflow is projected to change from near-future to far-future, and most changes are projected to occur in the autumn. The 25th, median, and 75th percentiles of the projected seasonal streamflows were calculated at all four streamflow stations under the SSP245 and SSP585 scenarios. The seasonal naturalized streamflow values in the far-future are lower than those in the near-future in 79% of the cases.

  3. The ensemble medians of both gross and net evaporation rates are projected to increase from near-future to far-future under both scenarios. The projected gross evaporation rate averaged over all six reservoirs during the period of 2020–2099 ranges from 3.28 to 4.29 mm/day, and from 3.32 to 4.37 mm/day under SSP245 and SSP585, respectively. The uncertainty ranges of the net evaporation rates are larger than those of the gross evaporation rates, which is attributed to the large uncertainties from the CMIP6 precipitation projections.

  4. The FY values of all reservoirs are projected to decrease, with the values under SSP585 smaller than those under SSP245 in both future periods. The median of the total FY in the far-future period is 13.6% and 16.8% smaller than the baseline period under SSP245 and SSP585, respectively. This suggests that under climate change, the water users in the study region are very likely to experience inadequate water supply in the future.

  5. For the same reservoir in the same future period, the RFY under SSP585 is projected to be smaller than that under SSP245 for most GCMs, indicating that the decrease in water availability is more significant under SSP585 than SSP245. The GCM-based uncertainties of both FY and RFY vary among different reservoirs—but are always larger under SSP585 than under SSP 245—during both future periods.

Acknowledgments

This work was supported by the U.S. National Science Foundation's Non-Academic Research Internship Program, the U.S. National Science Foundation (Grant CBET-1454297), and U.S. Geological Survey (USGS) graduate student research grants through Texas Water Research Institute. It was also partially supported by the U.S. Department of Energy (DOE) Water Power Technologies Office as a part of the SECURE Water Act Section 9505 Assessment. One of the authors is an employee of UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the DOE. DHSVM is maintained jointly by the Hydrology Group at Pacific Northwest National Laboratory (PNNL) and the Civil Engineering Department at the University of Washington. We thank Dr. Zhuoran Duan at PNNL for assisting with the DHSVM model. This work has benefitted from the usage of the Texas A&M Supercomputing Facility (http://hprc.tamu.edu). The climate model downscaling was conducted at the Oak Ridge Leadership Computing Facility, which is a Department of Energy Office of Science User Facility. We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF.

    Data Availability Statement

    The data of nine CMIP6 Global Climate Models used in this paper are available at Bentsen et al. (2019), Dix et al. (2019), EC-Earth Consortium (2020), Jungclaus et al. (2019), O'Connor (2019), Seferian (2019), Swart et al. (2019), Yukimoto et al. (2019), and Zhang et al. (2019). The observation based historical forcings are publicly available at Livneh et al. (2013). The reservoir Firm Yield data generated from this research is publicly available at Shao (2023).

    Software Availability Statement: The parallel version of DHSVM used in this research is available at Perkins et al. (2019) and DHSVM source code can be obtained from its Github repository (https://github.com/pnnl/DHSVM-PNNL). The code used for this study is in the “parallel” branch. The executable programs of Water Rights Analysis Package Modeling System are available at R. A. Wurbs (2021). Information about WAM and WRAP is available at Texas Commission on Environmental Quality (2023).