Volume 41, Issue 16 p. 5904-5911
Research Letter
Open Access

Groundwater depletion during drought threatens future water security of the Colorado River Basin

Stephanie L. Castle

Stephanie L. Castle

UC Center for Hydrologic Modeling, University of California, Irvine, California, USA

Department of Earth System Science, University of California, Irvine, California, USA

Search for more papers by this author
Brian F. Thomas

Brian F. Thomas

UC Center for Hydrologic Modeling, University of California, Irvine, California, USA

Department of Earth System Science, University of California, Irvine, California, USA

NASA Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA

Search for more papers by this author
John T. Reager

John T. Reager

UC Center for Hydrologic Modeling, University of California, Irvine, California, USA

Department of Earth System Science, University of California, Irvine, California, USA

NASA Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA

Search for more papers by this author
Matthew Rodell

Matthew Rodell

Hydrological Sciences Laboratory, NASA Goddard Space Flight Center, Greenbelt, Maryland, USA

Search for more papers by this author
Sean C. Swenson

Sean C. Swenson

Climate and Global Dynamics Division, National Center for Atmospheric Research, Boulder, Colorado, USA

Search for more papers by this author
James S. Famiglietti

Corresponding Author

James S. Famiglietti

UC Center for Hydrologic Modeling, University of California, Irvine, California, USA

Department of Earth System Science, University of California, Irvine, California, USA

NASA Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA

Correspondence to: J. S. Famiglietti,

[email protected]

Search for more papers by this author
First published: 24 July 2014
Citations: 263

Abstract

Streamflow of the Colorado River Basin is the most overallocated in the world. Recent assessment indicates that demand for this renewable resource will soon outstrip supply, suggesting that limited groundwater reserves will play an increasingly important role in meeting future water needs. Here we analyze 9 years (December 2004 to November 2013) of observations from the NASA Gravity Recovery and Climate Experiment mission and find that during this period of sustained drought, groundwater accounted for 50.1 km3 of the total 64.8 km3 of freshwater loss. The rapid rate of depletion of groundwater storage (−5.6 ± 0.4 km3 yr−1) far exceeded the rate of depletion of Lake Powell and Lake Mead. Results indicate that groundwater may comprise a far greater fraction of Basin water use than previously recognized, in particular during drought, and that its disappearance may threaten the long-term ability to meet future allocations to the seven Basin states.

Key Points

  • Groundwater depletion in the Colorado River Basin is greater than we thought
  • As GW disappears, the basin will struggle to supply water to the seven basin states
  • It is time to bring groundwater under the water management umbrella

1 Introduction

Over a decade, drought in the Colorado River Basin (Basin; Figure 1) has exposed the vulnerability [Bureau of Reclamation, 1975; Barnett and Pierce, 2008] of the most overallocated river system in the world [Christensen et al., 2004]. Recently, the U.S. Bureau of Reclamation acknowledged the potential challenges [Bureau of Reclamation, 2012] to meeting future surface water allocations to the seven Basin states (Figure 1), noting that the contribution of local supplies, including groundwater withdrawals, will be required to offset anticipated shortages. While the need to exploit groundwater resources to meet Basin water demands has long been recognized [Bureau of Reclamation, 1975], withdrawals required to meet current demands remain undocumented and are uncertain in the future. In particular, water management under drought conditions focuses on surface water resources [Basin Interim Guidelines, 2007] without a regulatory framework to manage groundwater withdrawals outside of “river aquifer” systems [Leake et al., 2013]. At question is the potential impact of solely managing surface water allocations and diversions in the Basin, without regard to groundwater loss, on meeting future water demands.

Details are in the caption following the image
The Colorado River Basin of the western United States. The state and international boundaries are in light gray. The green and brown colors represent the high and low elevations, respectively [McKay et al., 2012]. The upper Basin is that portion of the Basin upstream of Lake Powell. The lower Basin is the remainder of the basin downstream of Lake Powell. The basin outlines are in dark gray. The river, its main tributaries, and Lake Powell and Lake Mead are shown in blue.

The ability to observe changes in water resources at large scales has been greatly facilitated by the deployment of recent Earth-observing satellites. One such satellite mission, the NASA Gravity Recovery and Climate Experiment (GRACE) [Tapley et al., 2004], has measured the temporal variations in the Earth's gravity field since March 2002. These observations are now routinely applied to estimate the monthly changes in terrestrial or total land water storage (i.e., all of the snow, surface water, soil moisture, and groundwater) in regional areas that are 200,000 km2 or larger [Wahr et al., 2004] (Figure 2). Several studies have now demonstrated that GRACE observations, when combined with coincident data sets for snow water equivalent (SWE), surface water storage, and soil water content in a mass balance, can quantify changes in groundwater storage with sufficient accuracy [e.g., Rodell et al., 2009; Famiglietti et al., 2011] to influence regional water management decisions [Famiglietti and Rodell, 2013].

Details are in the caption following the image
Monthly anomalies (deviations from the mean of the study period) of the total water storage (TWSA) for (a) the entire Basin, (b) the upper Basin, and (c) the lower Basin, from January 2003 to November 2013 (i.e., the full GRACE RL05 record available at writing). The three TWSA estimates were calculated independently using basin-specific scaling. The anomaly errors are shown in light blue shading. There are inconsecutive gaps in the GRACE data record, increasing in number toward the end of the time period due to recent declines in satellite power supply. Subsequent analyses focus on the period of prolonged drought extending from December 2004 to November 2013.

Our goal in this report is to identify changes in freshwater storage, including surface reservoir and groundwater storage, to assess the influence of conjunctive surface water and groundwater use on water availability in the Colorado River Basin during the recent drought. We evaluate the terrestrial water storage anomalies (TWSA) using GRACE observations during a 9 year period (December 2004 to November 2013) that begins 4 years into a prolonged drought in the southwestern United States, after water levels in Lake Powell and Lake Mead had declined precipitously [Piechota et al., 2004] (see Methods section). In particular, we estimate the changes in groundwater storage during the 9 year drought period, when reservoir volumes were intensively managed to maintain hydropower production and to meet surface water allocations to the Basin states.

2 Methods

We used the Release 05 of the University of Texas Center for Space Research GRACE data [Tapley et al., 2007] (ftp://podaac.jpl.nasa.gov/allData/grace/L2/CSR/RL05/). Average water storage changes for the Colorado River Basin were computed as anomalies of terrestrial water storage in equivalent water height (in millimeters, converted to cubic kilometers here using the area of the study basins) following Swenson and Wahr [2009] (Figure 2). Processing methods include filtering GRACE data to reduce noise [Swenson and Wahr, 2006] and later restoring the associated lost signal over a specific region by scaling the data correctively [Velicogna and Wahr, 2006]. This processing results in estimates of satellite measurement error and leakage error from out-of-basin signal, both of which are included in a Basin-specific time-invariant error estimate [Wahr et al., 2006]. Figure 2 shows the Basin time series of terrestrial water storage changes from January 2003 to November 2013, nearly the complete available GRACE data record.

Because our focus here is on quantifying groundwater storage changes versus surface water storage changes during drought, we restrict our analyses to the 9 year period from December 2004 to November 2013. Prior to December 2004, the Basin had experienced four additional years of drought, effectively limiting surplus inflows that replenish Lake Powell and Lake Mead. This caused steep declines in reservoir storage prior to December 2004. Late 2004 also marked the beginning of a clear drought signal in the GRACE data, relative to its launch date in March 2002 (Figure 2).

To assess the accuracy of the GRACE data used here, we performed independent water budget analyses using regional precipitation (P) data from the PRISM system [Daly et al., 2008] (http://prism.oregonstate.edu/recent/), satellite-based evapotranspiration (ET) from Moderate Resolution Imaging Spectroradiometer (MODIS) [Tang et al., 2009], and the U.S. Bureau of Reclamation dam releases (Q) (usbr.gov; accessed December 2013) on the Colorado River. Uncertainty in the water balance estimate [Rodell et al., 2004a, 2004b] was calculated assuming relative errors of 15% for P [Jeton et al., 2005] and 5% in Q [Rodell et al., 2004b]. A 15% bias on the daily ET was determined by Tang et al. [2009]; we assume the relative error increases to 25% on a monthly time scale. We computed the monthly storage changes, dS/dt, as P − ET − Q, and compared them to dS/dt derived from the GRACE terrestrial water storage anomalies using a discrete backward difference. Results illustrate a good agreement between dS/dt derived from the water budget and that observed by the GRACE, for the entire Basin, and the upper and lower Basins (Figure S1 in the supporting information). Our comparisons were limited to March 2005 to March 2010 owing to the availability of ET estimates. Numerous additional studies have shown strong correspondence between GRACE water storage changes, hydrologic fluxes, and observations [see, e.g., Swenson et al., 2006; Famiglietti et al., 2011].

Accessible water storage changes (the combination of surface reservoir and groundwater storage changes) in the Basin are quantified using a water mass balance approach. Studies [e.g., Rodell and Famiglietti, 2002; Rodell et al., 2009; Famiglietti et al., 2011; Scanlon et al., 2012] have shown that GRACE-observed water storage changes, in combination with additional data sets, can be used to isolate individual components of the terrestrial water balance. We assume that the total water storage in a region is composed of soil moisture (SM), snow water equivalent (SWE), surface water (SW), and groundwater (GW):
urn:x-wiley:00948276:media:grl51963:grl51963-math-0001(1)
where the subscript t indicates a function of time, and changes in these components balance in their sum. We apply GRACE observations of variations from the long-term mean of this total with estimates of soil moisture and SWE to quantify changes in accessible water. We simplify equation 1 by defining accessible water as the sum of groundwater and surface water storage:
urn:x-wiley:00948276:media:grl51963:grl51963-math-0002(2)
where Δ indicates a variation from the time mean in an individual variable, and TWSA is the terrestrial water storage anomaly.

Soil moisture anomalies in equation 2 were estimated from the NASA Global Land Data Assimilation System (GLDAS) [Rodell et al., 2004a] (http://disc.sci.gsfc.nasa.gov/) due to the lack of observational soil moisture data on large scales and for consistency with the previous studies [Rodell et al., 2009; Famiglietti et al., 2011]. We average the results of three land surface models from GLDAS (Variable Infiltration Capacity [Liang et al., 1994], Noah [Chen et al., 1996], and Community Land Model 2 [Dai et al., 2003]) and apply the mean monthly standard deviation as an error estimate based on model structural biases (Figure S2 in the supporting information).

Data obtained from the Snow Data Assimilation System (SNODAS) [National Operational Hydrologic Remote Sensing Center, 2004] (http://nsidc.org/data/polaris/) were used for SWE in equation 2 (Figure S2 in the supporting information). SNODAS is the only gridded observation-based SWE product that assimilates ground, airborne, and satellite snow observations into its model structure and consequently has been used to represent SWE in other regional hydrologic studies [Famiglietti et al., 2011; Barlage et al., 2010]. Previous studies documented error of approximately 11% between SNODAS and snowpit observations in the Rocky Mountains [Rutter et al., 2008] and 15% error for basin-wide analysis [Famiglietti et al., 2011]. For this study, we assume 20% error due to the topographic and terrain heterogeneity throughout the Basin [U.S. Geological Survey, 2004].

We further separated the components of accessible water (Figure S3 in the supporting information) into surface water reservoir storage and groundwater storage (Figure 3). Reported reservoir storage time series from Lake Powell and Lake Mead were obtained from the U.S. Bureau of Reclamation [usbr.gov; accessed December 2013]. We assume that Lake Powell and Lake Mead account for the majority of the observed surface water change as they comprise approximately 4 times the annual flow of the river and make up 85% of surface water in the Basin [Rajagopalan et al., 2009]. The U.S. Geological Survey (USGS) errors for hydrologic measurements ranging from “excellent (5%)” to “fair (15%)” [Sauer and Meyer, 1992] were used to provide error estimates for surface water reservoir storage. A two sample t test could not reject the null hypothesis that sample means were different using the USGS ranges in error, and throughout the rest of the analysis, we used a 10% error estimate for the surface water reservoir storage time series.

Details are in the caption following the image
Monthly anomalies (km3) of groundwater storage (black) and of surface reservoir storage (green) for (a) the entire Basin (trend: −5.6 ± 0.4 km3 yr−1) and Lake Powell and Lake Mead combined (trend: −0.9 ± 0.6 km3 yr−1), (b) the upper Basin (trend: −1.7 ± 0.4 km3 yr−1) and Lake Powell (trend: −0.6 ± 0.6 km3 yr−1), and (c) the lower Basin (trend: −2.6 ± 0.3 km3 yr−1) and Lake Mead (trend: −0.1 ± 0.6 km3 yr−1), from December 2004 to November 2013. The anomaly errors are shown in light gray shading for groundwater storage and in light green shading for reservoir storage. All trends are summarized in Table 1.
We rearranged equation 1 to isolate the contribution of groundwater storage changes (Figure 3) to changes in the total water storage (Figure 2). We used the reservoir storage changes in Lake Mead and Lake Powell with soil moisture and snow water equivalent data as described above:
urn:x-wiley:00948276:media:grl51963:grl51963-math-0003(3)
where ΔSWt indicates the surface water anomaly from the reservoirs (Lake Powell and Lake Mead combined for the entire Basin: Lake Powell for the upper Basin and Lake Mead for the lower Basin). Equation 3 was solved each month, and errors in the groundwater storage were estimated by propagating the errors of TWSA, SM, SWE, and SW, following Rodell et al. [2004b].

We compared our GRACE-based estimates of groundwater storage changes to groundwater level observations at 74 monitoring wells located throughout the Basin. These data were obtained from the USGS [USGS Groundwater Climate Response Network, 2014] and from the Arizona Department of Water Resources (ADWR; https://gisweb.azwater.gov/waterresourcedata/GWSI.aspx, accessed May 2014). The selection of wells for comparison was limited to the locations with observations that were concurrent with GRACE. Of these, 7 USGS and 65 ADWR were located in the lower Basin, and 2 USGS monitoring wells were identified in the upper Basin. GRACE-derived groundwater estimates generally capture the observed behavior well (see Results section and Figure 4).

Details are in the caption following the image
Entire Basin comparison between the GRACE groundwater storage anomalies (black line) in km3 and the monthly USGS well observations. Because specific yield information is not available for all wells, we normalize each well time series by its standard deviation and then average (in blue). Selected well observations were only available from March 2005 to October 2012; thus, we calculated the average over this time period.

The trends reported in the text and summarized in Table 1 were estimated employing a method that accounts for residual serial correlation and time series error, and subbasin trends may not sum linearly [Johnston and DiNardo, 1997]. We identified several significant trends over the entire 108 month time period studied, and in shorter time periods, from December 2004 to January 2010 and from February 2010 to November 2013 (Table 1).

Table 1. Trends in Water Budget Components Were Calculated Employing a Method Which Adjusts a Linear Model for Residual Serial Correlation and Time Series Error [Johnston and DiNardo, 1997]a
Trends in Terrestrial Water in km3/yr
Time Component Entire Colorado River Basin (CRB) Upper CRB Lower CRB
Entire time period TWSA −7.18 ± 0.75 −2.34 ± 0.59 −3.90 ± 0.47
December 2004 to November 2013 SWE 0.00 ± 0 0.00 ± 0 0.00 ± 0
SM −1.29 ± 1.8 −0.861 ± 0.85 −0.905 ± 0.24
Reservoirs −0.865 ± 0.60 −0.638 ± 0.63 −0.057 ± 0.63
GW −5.56 ± 0.44 −1.66 ± 0.40 −2.63 ± 0.30
AW −5.40 ± 0.47 −1.13 ± 0.44 −3.02 ± 0.30
Time
Piecewise analysis 1 TWSA −10.6 ± 1.4 −3.41 ± 1.1 −7.49 ± 0.90
December 2004–January 2010 SWE 0.00 ± 0 0.00 ± 0 0.00 ± 0
SM −2.67 ± 4.2 −1.74 ± 1.9 −1.45 ± 2.2
Reservoirs −0.428 ± 0.34 1.31 ± 0.13 −1.20 ± 0.05
GW −6.23 ± 0.91 −1.91 ± 0.80 −4.06 ± 0.60
AW −6.29 ± 0.96 −1.37 ± 2.2 −5.27 ± 0.62
Time
Piecewise analysis 2 TWSA −19.2 ± 2.1 −11.5 ± 2.0 −9.14 ± 1.3
February 2010 to November 2013 SWE 0.00 ± 0 0.00 ± 0 0.00 ± 0
SM −6.82 ± 1.2 −2.88 ± 0.76 −3.64 ± 0.62
Reservoirs −8.42 ± 4.7 −3.22 ± 1.2 −0.085 ± 2.0
GW −10.9 ± 1.5 −6.10 ± 1.5 −5.83 ± 0.89
AW −11.2 ± 1.6 −7.48 ± 1.6 −4.85 ± 0.90
  • a The approach identified several significant trends (shown in bold) in accessible water (AW) in the Basin over the entire time period from December 2004 to November 2013 and a piecewise trend analysis conducted from December 2004 to January 2010 and from February 2010 to November 2013. The Basin TWSA estimates are calculated independently, and there is no assumption that subbasin trends will sum linearly.

3 Results

We find that during the 108 month study period, the entire Colorado River Basin lost a total of 64.8 km3 of freshwater (−7.2 ± 0.8 km3 yr−1, where ± represents the standard error of the slope coefficient) (Figure 2a) with a more severe rate of loss since February 2010 (−19.2 ± 2.1 km3 yr−1). The upper Basin (Figure 1) lost 21.6 km3 of water during the entire study period, with more severe loss rates after February 2010 (−11.5 ± 2.0 km3 yr−1) (Figure 2b). Study period losses in the lower Basin of 34.7 km3 were greater than in the upper Basin and declined at a faster rate (−3.9 ± 0.5 km3 yr−1) (Figure 2c). All trends are listed in Table 1. As described in the Methods section, we compared our GRACE-derived water storage estimates to independent water balances for the entire, upper, and lower Basins with good agreement (Figure S1 in the supporting information). This comparison lends additional confidence to the results reported here.

Further analysis of trends in groundwater storage (Figure S4 in the supporting information) revealed two distinct phases of depletion prior to and following 2009–2010. From December 2004 to January 2010, groundwater storage declined more rapidly in the lower Basin (−4.1 ± 0.6 km3 yr−1) compared to the upper Basin (−1.9 ± 0.8 km3 yr−1). Groundwater losses from February 2010 to November 2013 were found to be even greater in the upper (−6.1 ± 1.5 km3 yr−1) and lower Basins (−5.8 ± 0.9 km3 yr−1).

A brief recovery in groundwater storage is apparent from June 2009 to March 2010, when moderately wetter conditions provided a combination of potential groundwater recharge and temporarily alleviated the need to augment surface water supplies. The steepest rate of groundwater storage decline (in the upper Basin in 2013) follows exceptional drought conditions in 2012 and record low Rocky Mountain snowpack (U.S. Drought Monitor, 2012; see Figure S2 in the supporting information). Such behaviors highlight the close connection between surface water availability and groundwater use [Famiglietti et al., 2011].

We find that water losses throughout the Basin are dominated by the depletion of groundwater storage (Figure 3). Renewable surface water storage in Lake Powell and Lake Mead showed no significant trends during the 108 month study period, more recent declines (since 2011) and currently low (<50% of capacity) storage levels notwithstanding. Groundwater storage changes however accounted for the bulk (Table 1) of the freshwater losses in the entire Basin (50.1 km3 and −5.6 ± 0.4 km3 yr−1), the majority of which occurred in the lower Basin (Figure 3c). As mentioned in the Methods section, we examined the USGS and ADWR monitoring wells in the Basin during the study period. The observed behavior in these wells showed a good agreement with our GRACE-based estimates. Figure 4 shows the comparisons for the USGS wells. A Sen's slope trend comparison to the ADWR wells showed that measured groundwater table changes closely matched our GRACE-based estimates. These comparisons help confirm the groundwater depletion rates reported here.

4 Discussion

Drought in the Basin has effectively limited the surplus inflows that replenish Lake Powell and Lake Mead since the beginning of the 9 year study period, while active surface water management has prevented further declines in reservoir levels. Consequently, reservoirs show insignificant trends in storage levels (−0.9 ± 0.6 km3 yr−1), while groundwater has been significantly depleted (−5.6 ± 0.4 km3 yr−1). The vast difference may well be attributed to the regulatory framework already in place to manage surface waters, and to the general need for more active and enforceable groundwater management throughout the Basin, in particular, during drought.

The large, net negative change in groundwater storage is a clear indication that groundwater withdrawals are not balanced by recharge and must be greater than the observed depletion rate. The additional loss of 5.6 km3 yr−1 of groundwater, relative to the annual Basin surface water allocations of 18 km3 yr−1, indicates further that the Basin water supply was overallocated by at least 30% during the study period. Thus, we observe that groundwater is already being used to fill the gap between Basin demands and the annual renewable surface water supply.

Groundwater is typically used to augment sparse surface water supplies in the arid, lower Basin, and across the entire Basin during drought [Hutson et al., 2004; Kenny et al., 2009]. More generally, water managers around the world rely on groundwater to mitigate the impacts of drought on water supply [Leblanc et al., 2009; Famiglietti et al., 2011; Famiglietti and Rodell, 2013; Taylor et al., 2013]. Groundwater represents the largest supply of water for irrigation within the Basin [Hutson et al., 2004; Kenny et al., 2009], while irrigated acreage in the Basin has increased during our study period [Ward and Pulido-Velazquez, 2008; Cohen et al., 2013]. Furthermore, prolonged drought across the southwestern U.S. has resulted in overreliance on groundwater to minimize impacts on public water supply [Famiglietti and Rodell, 2013]. Long-term observations of groundwater depletion in the lower Basin (e.g., in Arizona—despite groundwater replenishment activities regulated under the 1980 Groundwater Code—and in Las Vegas [Konikow, 2013]) underscore that this strategic reserve is largely unrecoverable by natural means and that the overall stock of available freshwater in the Basin is in decline.

Future water management scenarios that account for both population growth and climate change also point to the inability of reservoir storage alone to meet the Basin allocations [Barnett and Pierce, 2008; Bureau of Reclamation, 2012]. These scenarios indicate that additional stresses will be placed upon the groundwater system, beyond those described here, to meet future Basin water demands. We believe that the combination of reduced surface water availability resulting from decreasing future snowpack [Barnett et al., 2008] and groundwater depletion poses a significant threat to the long-term water security of the region. As groundwater supplies reach their limits, the ability to supply freshwater during drought, or to fill the predicted, increasing gap between supply and demand [Bureau of Reclamation, 2012], will be severely constrained.

The challenge to policy makers and water managers in the Colorado River Basin is to reliably meet freshwater demand under these dynamic conditions. Our work suggests that a conjunctive surface water and groundwater management plan is essential for sustainable water management in the Basin. Despite commendable efforts to craft solutions to meet required surface water allocations [Bureau of Reclamation, 2012], consideration of the ability of groundwater withdrawals to meet current and future demands remains dormant. We hope that the heightened awareness of the rates of the Basin groundwater depletion highlighted here will foster urgent discussion on conjunctive management solutions required to ensure a sustainable water future for the Colorado River Basin and for the western United States.

Acknowledgments

This research was funded by grants from the NASA GRACE Science Team and from the University of California Office of the President, Multicampus Research Programs and Initiatives. Jason Christensen of the U.S. Bureau of Reclamation provided the reservoir capacities and dam releases. We thank Q. Tang of the University of Washington for the use of the MODIS ET data and Min-Hui Lo of the National Taiwan University for the assistance with the GRACE validation. Access to all data sets used in this study is available through the links provided in the main text. Authors J.T. Reager and J.S. Famiglietti conducted some of this research at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA.

The Editor thanks Sandra Postel and an anonymous reviewer for their assistance in evaluating this paper.