1 Introduction

The understanding of the geologic and hydrologic settings of, and the controlling factors affecting, freshwater resources in Egypt is gaining increasing importance due to the challenges posed by natural and anthropogenic forcing factors. The natural factors might include, but are not limited to, changes in rainfall and/or temperature patterns, duration, and magnitude, whereas the anthropogenic factors could include population growth, overexploitation, and pollution. Given the Egyptian hyper-arid climate, Egypt is currently receiving very little annual rainfall, distributed as 50, 20, and 5 mm over the Sinai Peninsula, Eastern Desert, and Western Desert, respectively (Ahmed et al. 2013; Mohamed et al. 2015). These minimal rainfall amounts and the relatively higher air temperature (~ 30 °C) are extremely vulnerable to climate change. Recent climate change studies over Africa indicated a tendency toward higher extremes, where the arid or semiarid areas will becoming increasingly dry and the wetter areas will witness intensified precipitation and flooding (Vörösmarty et al. 2000; Hulme et al. 2001).

Egypt’s total population is on rise; it increased from 22 million in 1950 to 94 million in 2016 and is expected to continue for decades to come. The majority (95%) of the Egyptian population lives in the Nile River valley and Nile Delta (~ 10% of Egypt’s area), whereas Egypt’s deserts remain largely uninhabited, causing enormous pressures on the Nile River surface water resources. Given the fact that the Nile basin is largely extended over varying topographic and climatic regimes, it represents one of the most vulnerable river systems across the world. Recent studies have shown that Nile basin countries are expected to experience a water stress that will be manifested as changes in precipitation patterns, amounts, frequencies, and distributions along with changes in temperature, changes in river flow, and the occurrence of associated floods and drought events (e.g., Swain 2011). Egypt currently uses its total annual allocation (55 km3/year) of Nile River waters coming mainly (~ 85%) from the Blue Nile. In addition, some of the Nile River source countries declared that they will no longer abide by the treaty that regulated the distribution of the Nile River water. For example, Ethiopia just launched a major project to construct the Grand Ethiopian Renaissance Dam which will deprive Egypt of considerable portions of its Blue Nile water for several years. Recent studies (e.g., Sultan et al. 2014b) have shown that if the Grand Ethiopian Renaissance Dam reservoir (capacity: 70 km3) is to be filled in seven years, Egypt will lose, for each of the seven years following dam completion, a minimum of 15 km3 of its annual allocation to reservoir filling (10 km3), evaporation (3.5 km3), and infiltration (1.5 km3).

Egypt is seeking additional freshwater resources to overcome some of the aforementioned challenges and to pursue its plans for modernization and development. Currently, Egypt is planning to utilize more of its groundwater resources, at the expense of Nile River water, to support national reclamation projects; a minimum of 1.5 × 106 acres will be reclaimed during the coming five years. According to the hydrogeological map of Egypt (RIGW/IWACO 1988), Egyptian aquifer systems (Fig. 1) include (1) the Nile aquifer that occupies the Nile flood plain and desert fringes (Late Tertiary to Quaternary); (2) the Moghra aquifer which occupies the western edge of the Delta (Lower Miocene); (3) the coastal aquifer that is distributed over northern and eastern coasts (Late Tertiary to Quaternary); (4) the carbonate aquifer in the north and middle parts of the Western Desert (Upper Cretaceous to Eocene); (5) the fractured basement aquifer in the Eastern Desert and Sinai Peninsula (Precambrian); and (6) the Nubian aquifer covering the Western Desert, western parts of the Eastern Desert, and the middle parts of Sinai Peninsula (Cambrian to Upper Cretaceous). The majority of current Egyptian reclamation projects depend mainly on Nubian aquifer water resources.

Fig. 1
figure 1

Hydrogeologic map showing the spatial distribution of the major aquifer systems in Egypt (RIGW/IWACO 1988). Also shown are the spatial extension of the NAE (red polygon) and the spatial distribution of monitoring wells (colored crosses). Inset a The spatial extension of the Nubian aquifer sub-basins in Egypt, Libya, Chad, and Sudan as well as the major uplift. Also shown is the spatial extension of the Nubian recharge domains that extend over the aquifer outcrops and receive average annual rainfall (AAR) greater than 20 mm (hatched area). Inset b The spatial extension of the entire Nile basin along with the lower Nile basin (hatched area)

Full size image

The Nubian aquifer (area: 2 × 106 km2; Fig. 1, inset a) represents a transboundary aquifer system shared by four countries: Egypt (38%), Libya (34%), Sudan (17%), and Chad (11%), where the majority of the aquifer’s water resources are located in Egypt (41.5 vol%) and Libya (41.5 vol%), and less of it in Chad (12.8 vol%) and Sudan (9 vol%) (Thorweihe and Heinl 2002). The Nubian aquifer contains three major sub-basins: the Dakhla sub-basin in Egypt; the Kufra sub-basin in Libya, northeastern Chad, and northwestern Sudan; and the northern Sudan sub-basin in northern Sudan (Fig. 1, inset a). The Dakhla sub-basin in Egypt extends over two minor rift-related sub-basins: the Northwestern basin and the Upper Nile Platform (Fig. 1, inset a). In this study, the Dakhla sub-basin in Egypt was called the Nubian aquifer in Egypt (NAE). Given the fact that the health of the NAE affects the success of the Egyptian reclamation projects as well as the livelihood of many people, the ability to routinely observe the water resources of the NAE and make those observations publicly available to the decision makers is inevitable. In situ observations (e.g., groundwater levels) of the NAE suffer from delay, gaps, discontinuity, inconsistency, and poor quality. Moreover, these observations are sparse and do not adequately represent the entire aquifer (NAE area: 0.66 × 106 km2) averaged estimates. Satellite remote sensing observations offer an alternative and/or complement to local in situ measurements and could be used to monitor the aquifer health and longevity. Most of these observations are globally distributed, free and publicly available, and temporally and spatially homogeneous.

The deployment of the Gravity Recovery and Climate Experiment (GRACE) mission and the collection of global temporal gravity fields over the past 15 years provide significant practical strategies to routinely observe and monitor the water resources of the NAE. The GRACE mission, sponsored jointly by the National Aeronautics and Space Administration (NASA) and the German Aerospace Center (DLR), is designed to map the temporal variations in Earth’s global gravity field on a monthly basis with unprecedented accuracy (Tapley et al. 2004a, b). The GRACE-derived variabilities in Earth’s gravity field can be used to make global estimates of the spatiotemporal variations in the total vertically integrated components (e.g., surface water, groundwater, soil moisture and permafrost, snow and ice, wet biomass) of terrestrial water storage (TWS) (Wahr et al. 1998).

The GRACE-derived TWS (TWSGRACE) data enabled the scientific community to address previously unresolvable hydrogeological questions (e.g., Ahmed et al. 2011, 2014b, 2016; Wouters et al. 2014; Fallatah et al. 2017). GRACE data have been extensively used to quantify aquifers’ recharge and depletion rates (e.g., Tiwari et al. 2009; Lenk 2013; Voss et al. 2013; Feng et al. 2013; Gonçalvès et al. 2013; Joodaki et al. 2014; Sultan et al. 2014a; Wouters et al. 2014; Castle et al. 2014; Döll et al. 2014; Al-Zyoud et al. 2015; Huang et al. 2015, 2016; Li and Rodell 2015; Chinnasamy and Agoramoorthy 2015; Chinnasamy et al. 2015; Ahmed et al. 2016; Huo et al. 2016; Jiang et al. 2016; Lakshmi, 2016; Long et al. 2016; Mohamed et al. 2016; Castellazzi et al. 2016; Veit and Conrad 2016; Wada et al. 2016; Yosri et al. 2016; Chinnasamy and Sunde 2016). Recent studies utilizing TWSGRACE data have shown that the NAE is witnessing an overall groundwater depletion. The reported groundwater depletion rates of the NAE varied with the examined period as well as the data sources. For example, groundwater depletion rates of 2.31 ± 1.00, 2.04 ± 0.99, 4.44 ± 0.42, and 2.58 ± 0.73 km3 were reported during the periods of April 2002 to November 2010, January 2003 to September 2012, January 2003 to December 2012, and April 2002 to December 2013, respectively (Sultan et al. 2013, 2014b; Ahmed et al. 2014a, 2015; Mohamed et al. 2015). Mohamed et al. (2016) have shown that the Nubian aquifer, from January 2003 to December 2012, is receiving an average annual recharge of 0.78 ± 0.49 and 1.44 ± 0.42 km3/year over the recharge domains in Sudan and Chad, respectively. By comparison, the Nubian aquifer in Libya and Egypt is witnessing a groundwater depletion of 0.48 ± 0.32 and 4.44 ± 0.42 km3/year, respectively. None of these studies has reported and/or quantified the amounts of natural recharge that the NAE is witnessing. Moreover, the recharge/discharge interaction of the NAE with artificial lakes, such as Lake Nasser and the Tushka Lakes, has not been clearly explained in many of these studies (Soltan et al. 2005; Sefelnasr 2007; Sefelnasr et al. 2015).

In this study, temporal (April 2002 to June 2016) TWSGRACE data along with the outputs of land surface models (LSMs) were used to provide improved estimates of recharge and depletion rates of the NAE. This study extends the previous investigations by (1) utilizing enhanced state-of-the-art TWSGRACE solutions, the global mass concentration solutions (mascons); (2) utilizing outputs from several LSMs; four versions of the Global Land Data Assimilation System (GLDAS) to isolate the GRACE-derived groundwater storage (GWS); (3) broadening the time interval by four years; and (4) investigating the interaction of the NAE with artificial lakes, such as Lake Nasser and the Tushka Lakes.

2 Data and Methods

In this section, a brief description of the GRACE, surface water, soil moisture, and rainfall datasets used in this study is provided. The procedures used to extract TWSGRACE, surface water storage (SWS), soil moisture storage (SMS) anomalies, and AAR are also presented in this section. All of the datasets used in this exercise are monthly and were acquired throughout the time period of April 2002 to June 2016.

2.1 GRACE-Derived TWS (TWSGRACE) Data

Three sources of GRACE data have been utilized in this study: spherical harmonics and mascon products of the University of Texas Center for Space Research (UT-CSR) and mascon solutions from the Jet Propulsion Laboratory (JPL). It is worth mentioning that, compared to the spherical harmonics fields, the mascon solutions provide higher signal-to-noise ratio, higher spatial resolution, and reduced error and do not require spectral (e.g., destriping) and spatial (e.g., smoothing) filtering or any empirical scaling techniques (Luthcke et al. 2013; Watkins et al. 2015; Save et al. 2016; Wiese et al. 2016; Scanlon et al. 2016). However, rescaling spherical harmonic solutions significantly increases the agreement with mascon solutions (Watkins et al. 2015; Scanlon et al. 2016).

The spherical harmonics of the UT-CSR GRACE solution (Level 2; Release 05; degree/order: 60; available at: ftp://podaac.jpl.nasa.gov/allData/grace/L2/CSR/RL05) were used in this study. The GRACE-derived C20 and degree-1 coefficients were replaced with those estimated by Cheng et al. (2011) and Swenson et al. (2008), respectively. The glacial isostatic adjustment (GIA) correction was applied using the GIA model developed by A et al. (2013). The temporal mean (April 2002 to June 2016) was removed from these solutions, and systematic and random errors were reduced by applying destriping and Gaussian (half-width: 200 km) filters, respectively (Wahr et al. 1998; Swenson and Wahr 2006). Following the procedures advanced by Wahr et al. (1998), the spherical harmonic coefficients were then converted to TWSGRACE grids of equivalent water thickness.

The generated TWSGRACE grids were then rescaled to minimize the attenuation in the amplitude of the TWSGRACE time series due to the application of GRACE post-processing steps (e.g., Landerer and Swenson 2012; Long et al. 2015). The approach described in Velicogna and Wahr (2006) was adapted to scale TWSGRACE. Two synthetic mass distributions were assumed across the NAE, converted into Stokes coefficients (up to degree 60), filtered using destriping and Gaussian (200 km) filters, and then reconverted to mass distributions. The ratio of the recalculated mass distribution to the original mass distribution is called the scaling factor. One of the selected synthetic mass distributions was chosen to represent the global TWSGRACE trend results, while the other was set to be 1.0 inside the NEA spatial domain and 0.0 outside it. These synthetic mass distributions were selected to reflect a real picture of what the TWSGRACE trends look like across the NAE spatial domain. The final scaling factor (1.90 ± 0.80) was selected as the average of the two scaling factors, generated by using the two different synthetic mass distributions, whereas the difference was used to quantify errors associated with that scaling factor. Raw TWSGRACE estimates over the NAE were then multiplied by the generated scaling factor to calculate the scaled TWSGRACE estimates.

The JPL mascon data (Release 05; version 2; 0.5° × 0.5° grid; available at: ftp://podaac.jpl.nasa.gov/allData/tellus/L3/mascon/RL05/JPL/) provide monthly gravity field variations for 4551 equal areas of 3° spherical caps. The coastline resolution improvement (CRI) filtered data, utilized to determine the land and ocean fractions of mass inside every land/sea mascon, were used in this study (Watkins et al. 2015; Wiese et al. 2016). The UT-CSR mascon solutions (Release 05; version 1; 0.5° × 0.5° grid; available at: http://www.csr.utexas.edu/grace/RL05_mascons.html) approach uses the geodesic grid technique to model the surface of the Earth using equal area gridded representation of the Earth via 40,962 cells (40,950 hexagons + 12 pentagons) (Save et al. 2012, 2016). The size of each cell is about equatorial 1°, the number of cells along the equator is 320, the average area of each cell is 12,400 km2, and the average distance between cell centers is 120 km. The UT-CSR mascon does not suffer from oversampling at the poles like an equiangular grid (Save et al. 2016).

The secular trend in TWSGRACE data was extracted by simultaneously fitting a trend and seasonal (e.g., annual and semiannual) terms to each TWSGRACE time series. The trend solutions are displayed in Fig. 2. Errors associated with monthly TWSGRACE and calculated trend values were then estimated (Tiwari et al. 2009; Scanlon et al. 2016): (1) monthly TWSGRACE time series were fitted using annual, semiannual, and trend terms and residuals (R1) were calculated; (2) R1 were smoothed using a 13-month moving average, a trend was removed, and the residuals (R2) were calculated; (3) the standard deviation of R2 represents the maximum uncertainty in monthly TWSGRACE values; (4) Monte Carlo simulations (e.g., Hastings 1970; Vrugt et al. 2009) were performed by fitting trends and seasonal terms for many (n = 20,000) synthetic monthly datasets, each with values chosen from a population of Gaussian-distributed numbers having a standard deviation similar to that of the examined population; and (5) the standard deviation of the extracted synthetic trends was interpreted as the trend error for TWSGRACE. The generated trend data were then statistically analyzed by using parametric techniques (i.e., Student t test) to identify trends that are statistically significant at 95 and at 65% levels of confidence.

Fig. 2
figure 2

Secular trend images of monthly (April 2002 to June 2016) TWSGRACE estimates generated, from a UT-CSR spherical harmonics, b UT-CSR mascons, c JPL mascons, and d average solution over the NAE and surroundings. Also shown is the spatial extension of the NAE (hatched area)

Full size image

Given the fact that TWSGRACE data have no vertical resolution, since GRACE cannot distinguish between anomalies resulting from different components of TWS (e.g., surface water, soil moisture, and groundwater), the contributions of SWS and SMS need to be quantified and removed from TWSGRACE time series (Fig. 3) to calculate GWS according to the following equation:

$$ \Delta {\text{GWS}}_{\text{GRACE}} =\Delta {\text{TWS}}_{\text{GRACE}} -\Delta {\text{SMS}} -\Delta {\text{SWS,}} $$
(1)

where ΔGWS, ΔSWS, and ΔSMS represent the change, with respect to the temporal (April 2002 to June 2016) mean, in groundwater, surface water, and soil moisture storage, respectively. Errors in GWS are then estimated by adding, in quadrature, errors associated with TWSGRACE, SWS, and SMS according to the following equation:

$$ \sigma_{{{\text{GWS}}_{\text{GRACE}} }} = \sqrt {(\sigma_{{{\text{TWS}}_{\text{GRACE}} }} )^{2} + \left( {\sigma_{\text{SMS}} } \right)^{2} + \left( {\sigma_{\text{SWS}} } \right)^{2} } , $$
(2)

where \( \sigma_{{{\text{TWS}}_{\text{GRACE}} }} \), \( \sigma_{\text{SMS}} \), and \( \sigma_{\text{SWS}} \) represent errors in ΔTWS, ΔSWS, and ΔSMS, respectively.

Fig. 3
figure 3

Temporal variations in TWSGRACE estimates, along with the associated uncertainties, extracted from the UT-CSR mascons (blue line), JPL mascons (green line), UT-CSR spherical harmonics (red line), and average (black line) solutions over the NAE

Full size image

2.2 SWS Data

Two main surface water reservoirs within the NAE were examined: Lake Nasser and the Tushka Lakes (Fig. 1). Both reservoirs are expected to affect TWSGRACE and therefore GWS estimates over the NAE. Figure 4, for example, shows the GRACE average sensitivity kernel function over the Lake Nasser. The sensitivity kernel function is generated by converting 1 cm uniformly distributed Lake Nasser height into monthly Stokes coefficients up to degree 60.

Fig. 4
figure 4

GRACE unscaled dimensionless averaging sensitivity kernel function generated over the Lake Nasser area within the NAE

Full size image

The Lake Nasser surface levels time series was extracted from averaging two main surface water-level datasets: (1) the U.S. Department of Agriculture’s Foreign Agricultural Service (USDA-FAS) global reservoir and lake monitoring database (GRLM; available at: https://www.pecad.fas.usda.gov/cropexplorer/global_reservoir/), and (2) the Hydroweb database at Laboratoire d’Etudes en Geophysique et Oceanographie Spatiales (LEGOS/GOHS; available at: http://www.legos.obs-mip.fr/fr/soa/hydrologie/hydroweb/) (Crétaux et al. 2011). The Lake Nasser monthly level anomalies were then generated, with respect to the temporal mean (April 2002 to June 2016), over the entire NAE (Fig. 5a). The accuracy of lake levels derived from these two databases has been estimated at 3–4 cm for the largest lakes (Birkett 1995; Shum et al. 2003; Swenson and Wahr 2009). In this study, the monthly errors in Lake Nasser level estimates were calculated as the standard deviation of sub-monthly water levels collected from Hydroweb and GRLM databases (e.g., Muala et al. 2014). Trends in Lake Nasser levels and associated trend errors were estimated using the procedures described for TWSGRACE.

Fig. 5
figure 5

Temporal variations, averaged over the NAE, in a Lake Nasser level anomalies, b Tushka Lakes level anomalies, and c SMS. Also shown are the uncertainty limits associated with each monthly value

Full size image

The Tushka spillway was constructed in 1978 to protect the Aswan High Dam from cyclic flooding events. In 1990, Lake Nasser started to rise; it reached a height of 182 m and spilled over into the Tushka spillway in September 1998. This event resulted in the formation of the first Tushka Lake. In subsequent years, as water continued to flow through the spillway, five additional lakes were created (Fig. 1). Examination of temporal satellite images indicates a gradual decrease in areas and volumes of the Tushka Lakes. Over 80% of the loss is via evaporation given the low permeability of the variegated shale underlying much of these lakes (e.g., Sultan et al. 2013). In this study, the monthly volume, area, and water height decrease in the Tushka Lakes were quantified. Two Landsat images (paths: 176 and 175; row: 44; spatial resolution: 30 m; available at: https://earthexplorer.usgs.gov/) were mosaicked and used to quantify the temporal variations, in water covered areas, of the Tushka Lakes in a geographic information system (GIS) environment. Landsat 5 Thematic Mapper (TM) images, Landsat 7 Enhanced Thematic Mapper (ETM +) images, and Landsat 8 images were used for the 2002, 2003–2012, and 2013–2016 periods, respectively. The Tushka Lakes’ volumes were calculated using the areas and water depth information. The water depth estimates were quantified using a digital elevation model (DEM; spatial resolution: 90 m), acquired prior to the formation of the Tushka Lakes, and validated using data extracted from topographic sheets (scale 1: 100,000). A 10% error estimate for the Tushka Lakes level time series was assumed (e.g., Castle et al. 2014). Moreover, the Tushka Lakes’ volumes during the period from 2002 to 2010 were compared to, and validated against, volumes extracted using different remote sensing images [e.g., MODerate-resolution Imaging Spectroradiometer (MODIS) and Advanced Very High-Resolution Radiometer (AVHRR) for area calculations and Geoscience Laser Altimeter System (GLAS) for water-level estimation] and techniques (e.g., Chipman and Lillesand 2007; Lillesand et al. 2015). The Tushka Lakes’ monthly level anomalies were then generated with respect to the temporal mean (April 2002 to June 2016) and averaged over the entire NAE (Fig. 5b). Trends in the Tushka Lakes’ levels and associated trend errors were estimated using the procedures described for TWSGRACE.

2.3 SMS Data

Soil moisture data (Fig. 5c) were extracted from GLDAS model (version 1; available at: ftp://hydro1.sci.gsfc.nasa.gov) given that previous studies over Saharan Africa (Ahmed et al. 2016) have shown that GLDAS provides more reasonable estimates of soil moisture in arid areas when compared to estimates from other LSMs. GLDAS is a NASA-developed land surface modeling system which performs advanced simulations to quantify optimal fields of land surface states (e.g., soil moisture, snow, surface temperature) and fluxes (e.g., evapotranspiration, ground heat flux) using ground and satellite-based observations (Rodell et al. 2004). Soil and elevation data inputs are based on high-resolution globally distributed datasets. The Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) data represents the precipitation inputs, whereas the radiation inputs are generated from the observation based downward radiation products, derived using Air Force Weather Agency (AFWA) fields and procedures. Other meteorological forcing data are produced by the National Oceanic and Atmospheric Administration (NOAA) Global Data Assimilation System (GDAS) atmospheric analysis system (Rodell et al. 2004). The GLDAS model simulates soil moisture through four versions: variable infiltration capacity (VIC), community land model (CLM), Noah, and Mosaic (Koster and Suarez 1996; Liang et al. 1996; Koren et al. 1999; Dai et al. 2003; Rodell et al. 2004). The soil moisture time series over the NAE was calculated by averaging the soil moisture estimates from the four GLDAS model versions (i.e., VIC, CLM, Noah, and Mosaic). The associated errors for GLDAS-derived monthly soil moisture estimates were calculated as the mean monthly standard deviation from the four GLDAS model simulations (e.g., Tiwari et al. 2009; Voss et al. 2013; Castle et al. 2014; Joodaki et al. 2014). Errors in soil moisture trends were calculated as the standard deviation of the trends computed from the four GLDAS simulations (Voss et al. 2013; Castle et al. 2014).

2.4 Rainfall Data

Rainfall data were utilized to explore the climatic controls on the temporal variation in TWSGRACE and GWS observed over the NAE. The CMAP data (available at: https://www.esrl.noaa.gov/psd/data/gridded/data.cmap.html) were utilized in this study. CMAP data provide global (88.75°N to 88.75°S) merged precipitation estimates from a variety of satellite- and ground-based sources from January 1979 to January 2017 with spatial and temporal resolutions of 2.5° and one month, respectively (Xie and Arkin 1997). The satellite sources used to produce monthly CMAP data include the Geostationary Operational Environmental Satellite (GOES) Precipitation Index (GPI), the Outgoing Longwave Radiation (OLR)-based Precipitation Index (OPI), the Special Sensor Microwave/Imager (SSM/I), and the microwave sounding unit (MSU). In addition to the aforementioned satellite sources, CMAP also integrates the National Centre for Atmospheric Research (NCAR) reanalysis precipitation data along with the Global Precipitation Climatology Centre (GPCC) data of ~ 200,000 routinely operating precipitation gauges (Xie and Arkin 1997). Over northern and eastern Africa, it has been demonstrated that CMAP data provide the best spatial and temporal correspondence with gauge-based measurements of rainfall and stream flow (e.g., Adeyewa et al. 2003; Dinku et al. 2008; Beighley et al. 2011; Sylla et al. 2013).

3 Results and Discussion

3.1 Temporal Variations in TWS Anomalies

Figure 2 shows the secular trend images in TWSGRACE estimates generated from CSR spherical harmonics (Fig. 2a), CSR mascon (Fig. 2b), JPL mascon (Fig. 2c), and the average of the three solutions (Fig. 2d) over NAE. Inspection of Fig. 2d indicates that the NAE is experiencing an average negative (− 3 mm/year) TWSGRACE trends during the investigated period. Higher TWSGRACE depletion rates (< − 6 mm/year) are observed over the northeastern parts of NAE. The northern coastal areas as well as the southern parts of the NAE, close to recharge areas in Sudan, are witnessing lower TWSGRACE depletion rates (> − 2 mm/year).

The spatial distributions of negative TWSGRACE trends slightly vary with the source of TWSGRACE data. For example, in case of the CSR mascon solutions (Fig. 2b), areas witnessing a uniform negative (− 3 mm/year) TWSGRACE trend are centered over the NAE; the location of these areas is shifted to the east and to the west in case of CSR spherical harmonics (Fig. 2a) and JPL mascon (Fig. 2c) solutions, respectively. Over the entire NAE, comparison between TWSGRACE trends generated from the three solutions (Fig. 2a-c) with the mean trend (Fig. 2d) indicates that JPL mascon-derived trends (Fig. 2c) are slightly (10%) lower than mean trends (Fig. 2d). However, CSR mascon (Fig. 2b) and spherical harmonics-derived (Fig. 2a) trends explain 96 and 93%, respectively, of the mean trend variabilities (Fig. 2d). This is probably related to the way that the TWSGRACE products have been generated. For example, JPL mascons were generated from 3° spherical caps, UT-CSR mascons were generated from 1° hexagons, and UT-CSR spherical harmonics were smoothed using a 200 km Gaussian filter (spatial resolution: ~ 125,000 km2).

Figure 3 shows the temporal variations in TWSGRACE time series generated over the NAE. Inspection of Fig. 3 shows an excellent agreement in amplitudes, phases, and trends of TWSGRACE extracted from UT-CSR mascon, JPL mascon, and UT-CSR spherical harmonic solutions. Moreover, the minute observed differences lie mostly within the uncertainty limits of each TWSGRACE estimate. Figure 3 also shows an overall depletion in TWSGRACE estimates extracted from the three different TWSGRACE solutions. Depletion rates of − 3.26 ± 0.16 mm/year (− 2.15 ± 0.11 km3/year), − 3.73 ± 0.23 mm/year (− 2.46 ± 0.15 km3/year), and − 3.15 ± 0.30 mm/year (− 2.08 ± 0.20 km3/year) were observed in TWSGRACE estimates of UT-CSR mascon, JPL mascon, and UT-CSR spherical harmonic solutions, respectively. The average depletion rate over the entire NAE, as calculated from the mean of the three solutions (black line; Fig. 3), is estimated at − 3.38 ± 0.21 mm/year (− 2.23 ± 0.14 km3/year).

Piecewise trend analysis of the average of the three TWSGRACE solutions (black line; Fig. 3), shown in Table 1, is conducted over four distinctive periods: April 2002 to February 2006 (Period I), March 2006 to March 2008 (Period II), April 2008 to December 2012 (Period III), and January 2013 to June 2016 (Period IV) (black dashed lines; Fig. 3). Examination of Fig. 3 and Table 1 shows that the NAE is witnessing a TWSGRACE decline during Periods I and III at almost the same rate [Period I: − 6.40 ± 0.98 mm/year (− 4.23 ± 0.65 km3/year); Period III: −6.61 ± 0.48 mm/year (− 4.36 ± 0.32 km3/year)] and with a much lower rate during Period II [Period II: − 0.37 ± 1.14 mm/year (− 0.25 ± 0.75 km3/year)]. However, during Period IV the NAE witnesses a TWSGRACE increase (Period IV: 9.38 ± 1.34 mm/year; 6.19 ± 0.88 km3/year). The temporal variability in TWSGRACE is related to temporal variations in one or more of the TWSGRACE compartments (e.g., SWS, SMS, and GWS). Below, the temporal variations in each of the TWSGRACE compartments are discussed.

Table 1 Trends in water budget components averaged over the NAE during the investigated period (April 2002 to June 2016)
Full size table

3.2 Temporal Variations in SWS Anomalies

Figure 5a shows the temporal variations in Lake Nasser level anomalies during the examined period. It is worth mentioning that during the investigated period Lake Nasser’s maximum height (179.71 m) was observed in November 2007, whereas its minimum height (169.52 m) was observed in July 2012. Lake Nasser general fluctuation is mainly attributed to the seasonality in the Nile River. It has been reported that the Nile River is witnessing 64-, 19-, 12-, and 7-year climate cycles (e.g., Kondrashov et al. 2005). Examination of Fig. 5a shows that Lake Nasser is witnessing an overall height increase of 0.42 ± 0.30 mm/year. Piecewise trend analysis (Table 1) indicates that the Lake Nasser level anomalies are declining (− 9.18 ± 0.84 mm/year) during Period I, increasing (23.79 ± 0.73 mm/year) during Period II, declining (− 8.66 ± 0.78 mm/year) during Period III, and increasing (6.19 ± 1.17 mm/year) during Period IV. Analysis of Lake Nasser trends indicates that the temporal variations in lake levels during Periods I, III, and IV are in phase and consistent with, and largely driving, the temporal variations in TWSGRACE. This assumption is supported by the fact that the TWSGRACE trends, observed during these periods, are correlated with increases and decreases of the level of Lake Nasser during the same periods (Figs. 3 and 5a and Table 1).

The temporal variations in the Tushka Lakes level anomalies are shown in Fig. 5b. Starting in 2002, the Tushka Lakes have witnessed a dramatic decrease in volume, area, and water level. For example, the Tushka Lakes’ volumes (areas) are estimated at 27.11 km3 (1669.62 km2), 11.81 km3 (972.42 km2), 5.55 km3 (512.72 km2), and 0.36 km3 (130.18 km2) in Januaries of 2002, 2006, 2010, and 2016, respectively. Analysis of temporal variations in the Tushka Lakes’ volumes indicates that they cumulatively lost 56, 80, and 98% of their volumes in 2006, 2010, and 2016, respectively, compared to their volume in 2002. The loss in the Tushka Lakes’ volumes and areas is believed to be an evaporation loss (e.g., Chipman and Lillesand 2007; Sultan et al. 2013). Examination of Fig. 5b shows that the levels of the Tushka Lakes are experiencing an overall systematic decrease in water levels of − 1.94 ± 0.01 km3/year that is equivalent to − 1.36 ± 0.01 km3/year (− 2.06 ± 0.30 mm/year) if distributed over the entire NAE.

3.3 Temporal Variations in SMS Anomalies

Figure 5c shows the temporal variations in GLDAS-derived SMS time series extracted over the NAE. Inspection of Fig. 5c shows that the SMS is witnessing an overall depletion of − 1.19 ± 0.02 mm/year (− 0.79 ± 0.01 km3/year). Inspection of piecewise trend analysis results (Table 1) and Fig. 5c indicates that the SMS is always declining during the four investigated periods but at varying rates (Period I: − 0.83 ± 0.09 mm/year; Period II: − 1.71 ± 0.09 mm/year; Period III: − 1.26 ± 0.03 mm/year; and Period IV: − 0.62 ± 0.10 mm/year).

3.4 Temporal Variations in GWS Anomalies

Figure 6a shows the GWS time series extracted, according to Eqs. (1) and (2), over the NAE. Inspection of Fig. 6a shows that the NAE is witnessing an overall GWS decline of − 0.55 ± 0.27 mm/year (− 0.36 ± 0.18 km3/year). Piecewise trend analysis results (Table 1) shows that the NAE is experiencing a GWS increase (5.16 ± 1.23 mm/year; 3.41 ± 0.81 km3/year) during Period I followed by a sharp GWS decrease (− 20.37 ± 1.24 mm/year; − 13.45 ± 0.82 km3/year) during Period II, then a GWS increase during periods III and IV (Period III: 6.52 ± 0.85 mm/year, 4.31 ± 0.56 km3/year; Period IV: 6.07 ± 2.45 mm/year, 4.00 ± 1.68 km3/year).

Fig. 6
figure 6

a Temporal variations in the GWS estimates generated over the NAE along with their uncertainty limits. b Validation of the GWS anomalies (black solid thick line) against the available monthly well observations (individual well: colored circles; average: blue thick line) over the NAE

Full size image

Given the paucity of the groundwater-level measurements in the study area, monthly (April 2005 to April 2008) water-level data of six monitoring wells, distributed in the Western Desert of Egypt (refer to Fig. 1 for well locations), were used to validate the GWS variability over the NAE. The water level in these wells ranges from 200.30 m to 246.6 m. Unfortunately, the specific yield information for these wells is not available. Hence, the water-level time series for each well was normalized by its standard deviation following the approach advanced by Castle et al. (2014) where we subtracted the temporal mean from each monthly water-level value and then divided by the temporal standard deviation. The normalized water-level and GRACE-derived GWS measurements are shown in Fig. 6b. Inspection of Fig. 6b indicated that the GRACE-derived GWS estimates generally capture the observed temporal groundwater-level variability during Periods I and II as indicated by the analysis of the available water level data.

3.5 Groundwater Recharge and Depletion Rates and Conditions

Examination of Fig. 6a and Table 1 reveals that the NAE is receiving natural recharge during Periods I (GWS trend: 3.41 ± 0.81 km3/year), III (GWS trend: 4.31 ± 0.56 km3/year), and IV (GWS trend: 4.00 ± 1.68 km3/year). However, the NAE is experiencing GWS depletion of − 13.45 ± 0.82 km3/year during Period II. The sharp GWS decline rate during Period II is largely related to exceptional drought conditions in Period I compared to the previous periods. Examination of the AAR and Lake Nasser levels indicates a decline during Period I (AAR: 120 mm; Lake Nasser level: 174.6 m) compared to the preceding years (1998–2002; AAR: 133 mm; Lake Nasser level: 178.2 m). Another common contributing factor, for the observed GWS depletion, could be the baseflow recession. Normal baseflow recession occurs naturally during periods of extended drought and could cause extensive water-level declines over time periods of weeks to months that result in volumetrically significant storage depletion (e.g., Alley and Konikow 2015). It is worth mentioning that during Period II the trend of the combined GWS and SWS components still positive (0.88 ± 0.75 km3/year), suggesting the possible usage of Lake Nasser surface water resources during periods similar to Period II. In other words, during the periods where the GWS trends are declining, Lake Nasser surface water could be used to augment the running irrigation projects and hence minimize the impacts of the GWS decline. One other possible solution could be channeling the encroaching Lake Nasser water, in high flood years such as Period II, across the western plateau reaching the lowlands west of the plateau, that are largely underlain by the Nubian aquifer, to recharge the NAE (e.g., Sultan et al. 2013).

To quantify the recharge rates during Periods I, III, and IV, the discharge rate (natural discharge + anthropogenic groundwater extraction) was added to the GWS trends using the following equation:

$$ {\text{Recharge}} =\Delta {\text{GWS}} + {\text{Discharge}}. $$
(3)

The sum of the average annual anthropogenic groundwater extraction and the average annual natural discharge for NAE was estimated at 2.85 km3/year (Sultan et al. 2007; Mohamed et al. 2016). The recharge rates for the NAE are estimated at 6.26 ± 0.81, 7.16 ± 0.56, and 6.85 ± 1.68 km3/year during Periods I, III, and IV, respectively. The total recharge during Periods I, II, and IV is estimated at 20.27 ± 1.95 km3; however, approximate average annual recharge rate of 1.66 km3/year is estimated during the three periods.

The recharge rates of the NAE during Periods I, III, and IV are partially attributed to increasing the AAR during the investigated periods compared to the preceding periods. Comparing AAR approach was used instead of examining the trends in rainfall given the fact that a rainfall value is already a rate and so it corresponds to a trend signal in TWSGRACE, whereas a trend in rainfall corresponds to an increase in TWSGRACE rate. However, a one-to-one correspondence, in magnitudes, is not to be expected, given that AAR could be redistributed as runoff and evapotranspiration that could affect the spatial and temporal distribution of the precipitated water, and hence the recharge locations and magnitudes. Moreover, a progressive shift in timing between the AAR and recharge should be also expected given the time rainfall takes to feed the groundwater in shallow and deep aquifers (e.g., Owor et al. 2009; Ahmed et al. 2011, 2014b; Hocking and Kelly 2016). It is worth mentioning that, over-arid and semiarid regions such as the NAE, only a small portion of any episodic rainfall event will reach the aquifer’s water table in the time of this event. However, this portion will increase gradually with time. The time lag between the rainfall and the recharge events depends on magnitude and duration of rainfall, soil type, texture, and hydrologic properties, and density and types of vegetation (e.g., Vogel and Van Urk 1975; Mauth et al. 2003; Döll and Flörke 2005; Keese et al. 2005; Thomas et al. 2016). Over the NAE, for example, the groundwater response to local rainfall events is delayed by 1–4 months; highest rainfall occurs in January; the peak rise in the groundwater level is recorded in April to May (e.g., Sultan et al. 2011; 2013; Gad 2009). These estimates are compared to, and correlated with, those from other arid and semiarid environments of similar geologic and climatic settings (e.g., Döll and Flörke 2005; Thomas et al. 2016; Tirogo et al. 2016). It is worth mentioning that the effective recharge and groundwater flow within the NAE could take thousands of years. For example, a progression of groundwater ages was observed within the NAE from southwest (< 0.03 × 106 years) to northeast (1 × 106years). This age progression suggests that the NAE received autochthonous recharge events that were primarily occurred at the foothills of the Uweinat mountainous area (Fig. 1). The spatial distribution of the age progression indicates relatively high groundwater flow velocities (~ 2 m/year) toward the north and low velocities (0.2 m/year) toward the east (e.g., Sturchio et al. 2004; Patterson et al. 2005; Sultan et al. 2013).

Figure 7a shows the AAR for the investigated Periods I, II, III, and IV as well as two preceding periods (1979–1997 and 1998–2002), averaged over the Nubian aquifer outcrops in Chad and Sudan (layout in Fig. 1, inset a) that are receiving AAR greater than 20 mm. A conservative 20 mm/years value was selected based on the fact that, in Egypt a rainfall event of less than 5 mm is not expected to produce a significant recharge and/or runoff events (e.g., Milewski et al. 2009). Examination of Fig. 7a indicates that the AAR over the Nubian aquifer recharge domains increased from 91 mm during 1979–1997 period to 133 during 1998–2002 period. This AAR increase is partially responsible for the groundwater recharge events during Period I. The annual recharge rates as well as the recharge-to-rainfall ratio are displayed in Fig. 7b. Similarly, the increase in the recharge rates during Periods III and IV is related to the increase in the AAR during Periods II and III, respectively. For example, the AAR increased from 120 mm during Period I to 155 mm and 121 mm during Periods II and III, respectively (Fig. 7a, b).

Fig. 7
figure 7

a Temporal variations in AAR generated over the Nubian recharge domains (solid black line) and the lower Nile basin (dashed black line). Average Lake Nasser level (solid gray line) is also shown. b Temporal variations in average annual recharge (dashed black line) and recharge-to-rainfall ratio (dashed gray line) over the NAE during the investigated periods

Full size image

It is worth mentioning that the recharge rates of the NAE during Periods IV are also partially attributed to the increase in Lake Nasser levels. The average Lake Nasser level was estimated at 174.6, 175.7, 174.5, and 175.6 m during Period I, Period II, Period III, and Period IV, respectively. The Lake Nasser level during Period IV is 1.1 m higher than the average level during Period III. The increase in Lake Nasser levels is supported by increasing the AAR over the lower Nile basin (layout in Fig. 1, inset b) during the same periods (Period I: 706 mm; Period II: 763; Period III: 748 mm; and Period IV: 769 mm). The AAR over the lower Nile basin during Period IV is 21 m higher than the AAR during Period III.

Given the current overall GWS depletion rate (− 0.55 ± 0.27 mm/year; − 0.36 ± 0.18 km3/year) during the entire investigated period (April 2002 to June 2016), the longevity of the NAE can be estimated. Based on modeled recoverable groundwater volumes (5180 km3) (Bakhbakhi 2006), the NAE could last for more than 10,000 years assuming a constant GWS depletion and recharge rates. Increasing depletion rates and/or decreasing the recharge rate would reduce the NAE’s longevity.

4 Summary and Conclusions

Egypt is currently seeking additional freshwater resources to pursue its plans for modernization and development. Decision makers are planning to utilize more of Egyptian groundwater resources, at the expenses of the limited Nile River surface water, to support national reclamation projects (1.5 × 106 acre in 5 years). Almost all of the reclamation areas are planned to utilize NAE groundwater resources. The NAE needs to be routinely and continuously monitored because of its importance.

Previous studies that utilized TWSGRACE to monitor the NAE reported groundwater depletion rates varied with the examined period as well as the data sources. In addition, the reported groundwater depletion rates were extracted from the entire time series, ignoring the temporal variability and cyclicity occurred at different time intervals. None of these studies reported groundwater recharge within the NAE. In this study, temporal (April 2002 to June 2016) TWSGRACE data along with the outputs of LSMs were used to provide improved estimates recharge and depletion rates of the NAE and to investigate the interaction of the NAE with the artificial lakes.

Results indicate that during Periods I, III, and IV the NAE is receiving a total recharge of 20.27 ± 1.95 km3. Recharge events of the NAE occur only under excessive precipitation conditions over the Nubian recharge domains (in Sudan and Chad) and/or under a significant rise in Lake Nasser levels. The sharp GWS decline rate during Period II (− 13.45 ± 0.82 km3/year) in the NAE is largely related to exceptional drought conditions in Period I compared to the previous periods. Another common contributing factor for the observed GWS depletion could be the normal baseflow recession. However, during this period the trend in the combined GWS and SWS components is still increasing (0.88 ± 0.75 km3/year) suggesting the possible usage of Lake Nasser surface water resources in development plans.

Findings indicate that Egyptian decision makers are facing a real challenge to provide and maintain sustainable water resource management. However, they are highly recommended to use a conjunctive surface water and groundwater management plan given the fact that in periods where the GWS is declining (e.g., Period II), the SWS of Lake Nasser could be utilized.

The study results demonstrate that global monthly TWSGRACE solutions can provide a practical, informative, and cost-effective approach for monitoring aquifer systems located in any geologic or hydrologic setting across the globe. However, it is worth mentioning that, in the calculations of TWSGRACE and GWS trends, the temporal variations that are related to semiannual, annual, multi-annual, and decadal climatic cycles were assumed to be represented in the examined TWSGRACE record, while the semiannual, annual, and multi-annual cycles are likely to be represented in the available TWSGRACE records, the decadal cycles might not be, given the short GRACE operational period (15 years). The acquisition of TWSGRACE data over the upcoming years by the GRACE-FO (expected in 2017/2018; nominal/expected lifetime: 5/10 years) and GRACE-II (planned in 2025; nominal/expected lifetime: 5/10 years) missions will in part address these limitations by enabling the acquisition of continuous and lengthy TWSGRACE records.