We study the relation between surface infiltration and groundwater
recharge during managed aquifer recharge (MAR) with desalinated seawater in
an infiltration pond, at the Menashe site that overlies the northern part of
the Israeli Coastal Aquifer. We monitor infiltration dynamics at multiple
scales (up to the scale of the entire pond) by measuring the ponding depth,
sediment water content and groundwater levels, using pressure sensors,
single-ring infiltrometers, soil sensors, and observation wells. During a
month (January 2015) of continuous intensive MAR
(2.45
Managed aquifer recharge (MAR) is a common practice in water resources management in which excess water is stored in the aquifers for future consumption. Major techniques used for aquifer recharge include well injection, bank filtration, rainwater harvesting, and infiltration ponds (Dillon, 2005). In the Israeli Coastal Aquifer, MAR started in 1958 with Lake Kinneret water, surface runoff, and carbonate-aquifer groundwater as the recharge sources (Sellinger and Aberbach, 1973). Between the years 2000 and 2013 MAR in the Israeli Coastal Aquifer was mostly (88 %) from the soil aquifer treatment ponds at the Shafdan sites, where secondary effluents are delivered into infiltration ponds for tertiary treatment. The remaining MAR can be primarily attributed to storm runoff according to the seasonal rainfall (Israel Hydrological Service, 2013). Recently, desalinated seawater – a relatively new water source in Israel (Stanhill et al., 2015) – has been occasionally used as source water for MAR.
MAR using desalinated seawater (DSW) poses several scientific and operative challenges due to the unique water composition compared to natural groundwater. Yet, scientific publications on MAR with DSW are few. Field tests of MAR using DSW were performed during the 1970s and the 1990s in clastic and carbonate aquifers in Kuwait. Well clogging was identified as a major concern, especially in the clastic aquifers (Mukhopadhyay et al., 1994). These field tests were followed by laboratory studies focusing on clogging and geochemical processes using core experiments with DSW (Al-Awadi et al., 1995; Mukhopadhyay et al., 1998, 2004). A closely related study, on MAR with reverse-osmosis wastewater was conducted at the St André MAR site in Belgium. Reported work includes flow and transport modeling (Vandenbohede et al., 2008, 2009a; Vandenbohede and Van Houtte, 2012), isotope and geochemical analysis (Kloppmann et al., 2008; Vandenbohede et al., 2009b) and reactive transport modeling (Vandenbohede et al., 2013).
In this paper we focus on infiltration and recharge dynamics during MAR with
DSW. This work is part of a comprehensive study involving field and
laboratory investigations in order to better understand hydrological and
geochemical processes during MAR with surplus of reverse-osmosis DSW in
Israel (Ronen-Eliraz et al., 2017). The geochemical perspective of this field
study will be reported in a future publication (Ganot et al., 2017). The
results reported here are unique for several reasons. First, we monitored a
month (January 2015) of continuous MAR with 2.45
The purpose of this paper is to provide a detailed field-scale analysis of MAR with DSW from a hydrological perspective. Initially the monitoring system is described, the unsaturated zone is characterized, and several methods for calculating infiltration and recharge are presented. Next, infiltration dynamics (spatial and temporal) and its relation to the unsaturated zone lithology and to pond surface clogging is discussed. Finally, groundwater recharge estimations obtained from the analytical and numerical models are evaluated and compared.
The Menashe MAR site is located on sand dunes, 28 m above mean sea level
(AMSL), overlaying the northern part of the Israeli Coastal Aquifer
(Fig. 1a). Climate is Mediterranean with annual mean precipitation of
566 mm yr
Monitoring of MAR activity was performed in the southern infiltration pond (herein referred to as “the pond”) where DSW discharged occasionally according to operational considerations of Mekorot (Israel national water company) and the Israeli Water Authority. A dedicated monitoring system including observation wells, soil sensors, and infiltration rings was installed at the south part of the pond (Fig. 1c, e). The two groundwater observation wells (OA and OB) are 30 m deep, perforated at the lower part of the well (10 m from the bottom) and penetrating the saturated zone. Both were monitored by loggers (CTD-Diver, Eijkelkamp) measuring pressure head and electrical conductivity (EC). The shallow unsaturated zone includes eight soil sensors (5TE and GS3, Decagon Devices) at depths of 0.3, 0.5, 1, 1.5, 2, 2.5, 3, and 4 m below the pond surface, measuring volumetric water content (WC) and bulk EC (a measure of the electric conductivity of the bulk soil, which includes soil, water and air). The monitoring system continuously operated since October 2014 and measurements were obtained regularly every 15–30 min and at a finer resolution of 1–5 min during MAR or infiltration tests. In addition to the permanent monitoring system, ponding depth was monitored by three pressure loggers installed on the pond surface for the January 2015 MAR event at the north, center, and south part of the pond (Fig. 1c). All pressure head measurements were compensated by on-site logging barometer (BARO-Diver, Eijkelkamp).
Disturbed sediment samples (from auger) were taken during the drilling of the observation wells. Relatively undisturbed continuous core samples from the unsaturated zone were obtained by a direct-push rig (9700-VTR PowerProbe, AMS). Cores were taken at the following locations: next to the soil sensors (0–12 m depth), the southern road (0–9 m depth), and east of the pond (0–6 m depth, Fig. 1d). All sediment samples (both disturbed and undisturbed) were analyzed for particle size distribution by sieving for gravel (> 2 mm), sand (2–0.045 mm), and hydrometer for silt and clay. Bulk densities were calculated only for the undisturbed samples from the mass and volume of the cores and their water content.
The sediment profiles of observation wells OA and OB are similar. The top
30 m includes two repeated sequences, each consisting of a sand layer
overlaying a sandy-clay-loam (SCL) layer, down to
Infiltration rates were calculated at the pond scale by pond draining rate, and at local scale by single-ring infiltrometers and wetting-front propagation (Dahan et al., 2007). Details of each method are given in what follows.
Ponding depth data were used to calculate the pond-scale infiltration rates. This method represents the infiltration rate of the whole pond, which is an average of the local infiltration rates (that may vary spatially due to sediment heterogeneity). The average pond infiltration rates were calculated by linear regression of ponding depth, which declined due to intermittent inlet discharge during the January 2015 MAR event. Each observation point of infiltration rate was calculated from a large number (tens to hundreds) of ponding-depth measurements. Two conditions must be met in order to calculate infiltration rates by this method: (1) ponding depth is declining solely due to infiltration (i.e., no other inlet/outlet source or surface flow) and (2) the time span of the descending ponding-depth data is sufficiently long (usually at least a few hours) in order to obtain regression with low-error slope (which is an estimate of the pond-scale infiltration rate).
In order to capture local infiltration rate variability, we used an array of 24 single-ring infiltrometers (100 cm long, 20 cm diameter) hammered 60 cm into the ground at different locations (Fig. 1e). Sediment samples taken from each infiltration ring location (outside the ring) at depth of 5 cm (undisturbed) and 60 cm (disturbed, with an auger, divided into four sections) were analyzed for bulk density (undisturbed only) and particle size distribution (both).
Infiltration tests were performed under relatively dry conditions (average WC
of 0.09 m
Recharge models.
Monitoring of water content variation in the unsaturated zone by the soil sensors provides information on the wetting/drying-front propagation velocity. Infiltration (or drainage) rates are evident from the lag in wetting (or drying) front between different sensors at various depths. Infiltration rates were estimated from the velocity of the fronts and the difference in water content on both sides of the wetting front.
The material properties used for the layers in the analytical
models and uncalibrated numerical model (Fig. 2): soil texture and
van Genuchten–Mualem hydraulic functions parameters – residual and saturated
water contents,
We describe the flow from the surface to the water table during a MAR operation using three different models: two simple analytical models (i.e., one using water table data, the other ponding-depth data) and one numerical model (in which both data sets were used). The simple analytical models are useful not only when there are not enough data to calibrate a numerical model; they also provide a first approximation which can be used as a preliminary test for a numerical model. In all three models we assign similar sediment profile layers and saturated hydraulic conductivity (Table 1), evaluated from pedotransfer functions (PTFs) using bulk density and particle size distribution data (Schaap et al., 2001). Only the saturated hydraulic conductivity of the top SCL layer was modified during calibration of the numerical model. Temporal and cumulative infiltration/recharge were obtained using 5 min resolution data measured during the January 2015 MAR event.
Because the pond water depth is much smaller than its horizontal dimensions, we consider 1-D vertical infiltration (perpendicular to the layers) – e.g., see Philip (1992). This assumption neglects lateral water flow, which is mainly relevant at the pond boundaries and during early and late stages of MAR (when only a portion of the pond surface is covered with water). However, during most of the January 2015 MAR event the whole pond area was covered with water and therefore the 1-D flow is a reasonable approximation. A main advantage of the 1-D model, apart from its simplicity, is that it can capture the whole-pond MAR processes in a single representative 1-D sediment profile.
In this lumped model, we use two measured data sets – water table levels and
saturated hydraulic conductivities. We consider the following transient
boundary conditions: flux at the top and water table level (head) at the
bottom. At the top, the recharge flux is equivalent to the flux in a
saturated layered column under unit gradient flow. The lower boundary is the
level of the moving water table. Calculations begin from the time that the
water table starts rising, with initial conditions of fully saturated
sediment profile. Assuming an initially saturated profile allows using the
saturated hydraulic conductivity (
The second analytical model assumes seepage flow through a perched water
surface together with ponding depth data. We consider steady-state seepage
through the topmost low-permeability SCL layer (Fig. 2a, 4–6 m depth), and
that both this layer as well as the sand layer above it (0–4 m) are
saturated under ponding, justified by the disparate hydraulic conductivities
of the sand (high) and SCL (low) layers. By the same reasoning, the sand
layer below the restrictive SCL layer remains unsaturated, maintaining
steady-state flow (Fig. 2b). With the above, the 1-D steady-state flux
through the saturated layers
Water ponding depth at the south, center, and north locations
inside the pond during the January 2015 MAR event. Observation well OA shows
a sharp increase of the water table during MAR (17 m). Minor ticks on the
Infiltration through the unsaturated zone was simulated with the HYDRUS-1-D software (version 4.16), a finite-element code for 1-D uniform water movement in variably saturated rigid porous media (for a detailed description of the governing equations see Šimůnek et al., 2009). HYDRUS-1-D was recently used to evaluate recharge in natural settings (Assefa and Woodbury, 2013; Neto et al., 2016; Turkeltaub et al., 2015). In this work HYDRUS-1-D was used to evaluate infiltration through the unsaturated zone and groundwater recharge using the 1-D Richards equation, with negligible root water uptake (sink term).
The model domain includes 10 layers within 0–30.5 m depth, with eight different compositions based on the sediment core samples, discretized into 1000 elements of thickness of 0.1–4 cm (average of 3 cm) varied according to the sediment type and location, and to the original groundwater level (Fig. 2c).
The van Genuchten–Mualem model (van Genuchten, 1980; Mualem, 1976) was used
for the water retention curves and unsaturated hydraulic conductivity
functions of the different sediments. The hydraulic parameters (Table 1) were
calculated from the measured particle size distribution and bulk density
using PTFs (ROSETTA; Schaap et al., 2001), which is incorporated in
HYDRUS-1-D. An exception was made for the kurkar rock (layers 8 and 10), for
which we used the hydraulic parameters of material 5 (sand) since the kurkar
was crushed during drilling and its structure was destroyed. The saturated
hydraulic conductivities (
Boundary conditions at the soil surface (top) were the ponding depth (monitored at its surface when filled) and no flow when the pond is empty, and groundwater level (measured in well OA) at the bottom. The variable-head boundary conditions were applied at fixed locations at the top and bottom of the domain, and were updated at a 1 h resolution. Variable-head boundary conditions were selected in both boundaries in order to capture the highly dynamic behavior of the system during the January 2015 MAR event (rise of 2.2 and 17 m in ponding depth and groundwater level, respectively). The code output is the flow at the domain boundaries: infiltration flux at the pond surface (top) and groundwater recharge (bottom). The recharge flux at the fluctuating groundwater table is similar to that at the bottom of the domain, because HYDRUS-1-D neglects the storage term in the variably saturated flow equation. Initial water content profile was obtained by field data (unsaturated at depth of 0–4 m and saturated below the water table, at 24.4 m) and by running a simulation for 88 days (September to December 2014) before the beginning of MAR that incorporates daily precipitation and evaporation.
The three pressure loggers monitored the local ponding depth during January 2015 MAR event, showing the filling period as water flows from the south to the north part of the pond (Fig. 3). A uniform water level was reached after the whole pond surface was covered with water and the three pressure loggers measured similar levels (beginning on 3 January 2015), with a constant difference between the pressure loggers that represents ponding depth difference due to a shallower pond surface (topography) at the northern part. Next, water ponding depth increases sharply and reaches a maximum of 2.2 m (11 January 2015). At this stage a small dam was opened at the north part of the pond allowing water to flow freely to the connecting channel and to the northern infiltration ponds (Fig. 1b) and the ponding depth was stabilized at 1.8 m for 17 days. During this period, on 21 January 2015, a flooded area of 10.7 ha was mapped with a GPS device to obtain the ponded area (Fig. 1c). After 31 days, inlet discharge was stopped and the water ponding depth decreased; finally on 2 February 2015 the pond was drained completely.
The pressure loggers in the observation wells captured a substantial rise of 17 m in groundwater level after 1 month of continuous MAR during January 2015 (Fig. 4a). Note that this rise represents the local conditions beneath the pond, while the influence on regional groundwater levels is damped farther away from the pond (e.g., a well, located 600 m to the north of the pond margins, showed a maximal groundwater-level rise of 4 m; data not shown). The lag in groundwater rise (Fig. 3) is due to the infiltration process through the unsaturated zone, which takes around 3 days until water reaches the water table, since the beginning of discharge on 29 December 2014. However, groundwater drops almost immediately (5 h) after inlet discharge was stopped and pond water level declines. This fast response implies fully connected flow (Brunner et al., 2009) between the high-level groundwater and the pond. Note that the relatively sharp decline in groundwater level during the first month after discharge was ended is followed by a gradual decline in the following months (Fig. 4a).
The EC monitoring of groundwater at the observation wells is shown in Fig. 4a
to emphasize the response of groundwater to MAR with DSW
(
Infiltration rate measurements at various scales and perspectives:
Volumetric water content (WC) measurements in the vadose zone also capture
the January 2015 MAR event showing constant WC during most of the ponding
period and also later during the dry period. Changes in WC are most notable
during wetting and drying of the upper sand layer at the beginning and end of
infiltration, respectively (Fig. 4b). The bulk EC at depth of 3 m is
increasing at the beginning of the MAR event and then decreases and
stabilizes at 0.08 mS cm
Pond infiltration rates show a general decrease during the January 2015 MAR
event (Fig. 5a). Infiltration rates of 9
Results of the single-ring infiltration tests under the different conditions
show some degree of spatial and temporal variability of infiltration rates
(Fig. 5b). Spatial variability was evaluated from differences in rates in
different locations, and was found to be moderate (coefficient of variation,
CV
Sharp wetting and drying (drainage) fronts, typical for sandy and coarse
sediments, were observed at the beginning and end of infiltration,
respectively (Fig. 5c, d). The estimated infiltration rates between couples
of soil sensors (i.e., between 0.3 and 0.5 m, 0.5 and 1 m, etc.) change
with time: the infiltration rate generally decreases as the wetting front
advances deeper (Fig. 5e), as expected from theory (Assouline, 2013) and the
impact of the SCL layer, while the drying front shows a more complex trend
(Fig. 5f). In order to compare with infiltration rate measurements by other
methods, the average infiltration and drainage rate of the top sand layer was
estimated between the soil sensors at 0.3 and 4 m depth, as 12.8 and
0.16 m d
The lumped and seepage models provide cumulative recharge of 20.2 and 16.4 m
(2.2
To capture temporal variations in drainage and groundwater recharge, we
simulated the January 2015 MAR event from 25 December 2014 to 5 October 2015.
We calibrated the numerical model to the whole-pond infiltration rate data
(Fig. 5a) in order to generalize the local sediment profile into a whole-pond
representative profile. Only saturated hydraulic conductivities of the top
SCL section were modified during the calibration (4–6 m, layers 2 and 3,
calibrated
The model calibration shows a good fit for 90 % of the infiltration period
(4–31 January 2015) with a relative root mean square error of 4.8 %
(Fig. 6a). Relatively poor fits between the model and the whole-pond
infiltration data were obtained for the first two observations points (25 and
30 December 2014). These two observations were measured at early stages, when
the pond was partly filled, which may overestimate infiltration rates due to
surface flow. Checking the calibrated model against WC data (from the vadose
zone monitoring system at 2 m depth) shows reasonable validation as the
model was calibrated against whole-pond data, while the WC represents
point-specific data. In terms of wetting/drying front, the model
underestimates the arrival time of the front. A better fit to the WC data was
achieved during the calibration process using the built-in HYDRUS-1-D inverse
modeling, by fitting the van Genuchten–Mualem parameters
Testing the numerical model with different PTFs shows an expected variation in the results of infiltration and recharge rates. These variations clearly demonstrate the need for calibration when using PTFs to estimate deep vadose zone hydraulic parameters (Zhang et al., 2016). In this case, the simulation with the PTFs of Tóth et al. (2015) was closest to the calibrated simulation results (Fig. S1 in the Supplement).
Cumulative infiltration and recharge of the various models during
2015. According to the numerical model, most of the infiltrated water
(
Testing models assumptions.
Change in water storage in the unsaturated zone (i.e., above the original water table) during ponding (stage 1 and 2) and drainage (stage 3). Simulated rates of surface infiltration and groundwater recharge are also shown as a reference.
Our numerical simulation results highlight the transient nature of
groundwater recharge. At the end of the January 2015 MAR event when the pond
was empty (2 February 2015), the estimated total groundwater recharge was
17.1 m, vs. 22.4 m of surface infiltration (Fig. 7). That is, during
ponding
For a flooded pond surface of 10.7 ha the total surface infiltration gives
roughly a total water volume of 2.4
Spatial infiltration variability depends on the soil type and structure and
its spatial distribution in the pond. Single-ring dry infiltration rates
showed significant correlation with bulk density (
Temporal infiltration variability is evident from the single-ring tests as
infiltration rates decrease from 6–16 m d
High spatial and temporal variability of infiltration rates, measured with thermal and pressure probes, was reported by Racz et al. (2012) during several months of MAR to an infiltration pond with an area of 3 ha. They postulated that small differences in the percentage of fine material in the relatively homogeneous shallow soil, clogging of the pond surface, and deeper unsaturated zone processes can explain this variability. Mawer et al. (2016) used fiber optic distributed temperature sensing to monitor infiltration rates with high spatial resolution during MAR to an infiltration pond. They concluded that 80 % of the recharged water infiltrated through the most permeable 50 % surface area of the pond which was explained by heterogeneous clogging. In our study the relatively deep unsaturated zone sampling and infiltration rate data show that the spatial and temporal variability of infiltration rates is suppressed (and controlled) by the low-permeability layers. Probably for the same reason, together with the high-quality source water (DSW), there was no field evidence in our study for clogging of the top sand layer.
Clogging of the infiltration surface is the major operational concern in most
MAR systems (Bouwer, 2002; Martin, 2013). The extent of clogging during MAR
with DSW is questionable due to the low turbidity, organic matter, and total
dissolved solids (TDS) of the source water (in this case, the DSW turbidity
To further examine the impact of DSW on clogging, we performed preliminary
infiltration column experiments in the laboratory with DSW and sand taken
from the pond surface (top 0.4 m). Results showed a reduction by a factor of
1.5 compared to the initial infiltration rate, probably due to
compaction clogging (see Sect. S2 in the Supplement). Similar results were
obtained by Lado and Ben-Hur (2010) in a column experiment with sandy soil
leached with reverse-osmosis effluent. They suggested that the relatively
large average pore size in the sandy soil prevented pore clogging and
It is worth noting that infiltration of low-salinity water into natural
settings may cause clogging due to clay swelling, dispersion and colloidal
release and deposition, which can lead to
Clearly, the simplified models cannot capture groundwater recharge dynamics
as the numerical model does, but they can serve as a first approximation for
recharge when no other data are available or as complementary recharge
estimation when other methods are used. The lumped model suffers from
practical and theoretical limitations compared to the seepage model. The need
for drilling a monitoring well inside (or very close to) the infiltration
pond and in addition continuously monitoring groundwater level is the main
operational limitation. Moreover, there is no field evident that supports the
assumption of a saturated profile with a unit gradient along the
heterogeneous sediment profile. Yet, a main advantage of the lumped model is
its ability to predict groundwater recharge using only pedotransfer-based
The unit-gradient assumption in the simplified models was tested using the results of the calibrated numerical model. Checking the hydraulic gradients as calculated from the calibrated numerical model, for the lumped model (between the pond surface and the groundwater table) and for the seepage model (at 6 m depth, below the upper SCL layer), shows that the unit-gradient assumption is not always valid (Fig. 8a). This is due to the significant water table rise, the layered sediment profile, and the variably saturated conditions. These factors, together with the lack of calibration of the simplified models, provide a possible explanation for the differences between the simplified models and the calibrated numerical model.
The calibrated 1-D numerical model is a more complex tool compared to the simplified models that were presented or to other approximated methods, yet it is still simpler compared to 2-D or 3-D variably saturated models. Sumner et al. (1999) and Morel-Seytoux (2000) discussed the validity of 1-D flow along the unsaturated zone for estimating groundwater mounding during MAR. We tested our numerical model results using the analytical solution of Morel-Seytoux et al. (1990), which assumes 1-D vertical infiltration along the unsaturated zone and radial flow along the saturated zone. The calculated groundwater level below the pond (Eq. 27 in Morel-Seytoux et al., 1990), using the infiltration and recharge rate results from the calibrated 1-D numerical model, shows a reasonable fit with the observed groundwater levels, supporting our assumption that flow along the unsaturated zone is mainly vertical (Fig. 8b). The differences between the calculated and observed groundwater levels can be attributed to the analytical model assumptions and to errors associated with the estimated model parameters (0.25, 185, and 70 m for specific yield, equivalent pond radius, and saturated aquifer thickness, respectively).
The main advantage of our 1-D numerical model is its ability to capture infiltration and recharge dynamics of the MAR system based on only one representative sediment profile. The obvious main drawback of the 1-D model is its inability to capture lateral flows, both at the unsaturated and saturated zones. This limitation, to some extent, is compensated by the use of data-based variable-head boundary conditions which were employed to better estimate surface infiltration and groundwater recharge (for comparison, applying a constant-head lower boundary condition as an alternative will overestimate recharge). However, when using these boundary conditions the model is inadequate for predicting water table or ponding depth evolution during future MAR events. This limitation can be overcome by changing the model boundaries, as discussed, for example, by Neto et al. (2016).
Predicting the recharge dynamics during MAR by numerical simulations is a
valuable tool for planning successive MAR events, and as an input for
regional groundwater models. Groundwater recharge is governed by the
boundaries of the system and the unsaturated zone hydraulic properties. In
MAR sites with unsaturated zone of intermediate depth (normally
Groundwater level under a sandy infiltration pond in the Israeli Coastal
Aquifer rose by 17 m during 1 month of continues MAR with surplus
desalinated seawater. Measured infiltration rates were relatively uniform
spatially, however highly variable in time: during continuous discharge of
2.45
The data of this paper are available upon request from the first author via email (yonatan.ganot@mail.huji.ac.il).
The authors declare that they have no conflict of interest.
The research leading to these results received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 619120 (Demonstrating Managed Aquifer Recharge as a Solution to Water Scarcity and Drought – MARSOL). We thank Amos Russak and Raz Amir for their technical assistance and Eline Futerman-Hartog for conducting the dry infiltration experiments. Edited by: Bill Hu Reviewed by: three anonymous referees