Managed aquifer recharge with reverse-osmosis desalinated seawater: modeling the spreading in groundwater using stable water isotopes

Abstract

Abstract. The spreading of reverse-osmosis desalinated seawater (DSW) in the Israeli Coastal Aquifer was studied using groundwater modeling and stable water isotopes as tracers. The DSW produced at the Hadera seawater reverse osmosis (SWRO) desalination plant is recharged into the aquifer through infiltration pond at the managed aquifer recharge (MAR) site of Menashe, Israel. The distinct difference in isotope composition between DSW (δ18O = +1.41; δ2H = +11.34 ‰) and the natural groundwater (δ18O = −4.48 to −5.43 ‰; δ2H = −18.41 to −22.68 ‰) makes the water isotopes a preferable tracer compared to widely-used chemical tracers, such as chloride. Moreover, this distinct difference can be used to simplify the system to a binary mixture of two end members: desalinated seawater and groundwater. This approach is especially robust when spatial data of stable water isotopes in the aquifer is scarce. A calibrated groundwater flow and transport model was used to predict the DSW plume distribution in the aquifer after 50 years of MAR with DSW. The results show that after 50 years 94 % of the recharged DSW was recovered by the production wells at the Menashe MAR site. The presented methodology is useful for predicting the distribution of reverse-osmosis desalinated seawater in various downstream groundwater systems.

Exploring Predictive Uncertainty at a Double‐Source Managed Aquifer Recharge Site via Stochastic Modeling

“…These high fractions of DSW in downgradient wells are close to the deterministic prediction for these wells predicted by Ganot, Holtzman, Weisbrod, Bernstein et al. (2018). The similarity is due to the fact that in Ganot, Holtzman, Weisbrod, Bernstein et al.…”

“…The Transition Probability (T‐ProGS) method, which is a Markov chain approach to geostatistical analysis and simulation of spatial distributions of categorical variables (Carle, 1999), was applied to generate the hydrogeological model of the subsurface in a single (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018) and multirealization approach, all conditioned to the log data. In the single realization approach subsurface model, constructed in a previous study (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018; Section 2.3.4), the model domain includes two subdomains (∼2/3 western and 1/3 eastern, location of separation plane is marked with a dashed black line; Figure 1b) in order to increase hydrogeological realism, in which an evident difference exists in characteristics between two statistical subregions, as reflected in well logs' data (i.e., higher % of clay in the east; Figure 1c and Table 3). The western and eastern subdomains were generated separately, as following: (a) geostatistical properties of the categorical variables of the eastern and the western subdomains were calculated using log data collected from boreholes to the east and west of the subdomain separation plane, respectively (Figure 1b).…”

“…The subsurface is represented by four categorical variables (i.e., hydrogeological units, hydrofacies, Table 3), as interpreted from over 100 borehole logs drilled during numerous geological studies conducted in the region over the decades (Figure 1b). The Transition Probability (T‐ProGS) method, which is a Markov chain approach to geostatistical analysis and simulation of spatial distributions of categorical variables (Carle, 1999), was applied to generate the hydrogeological model of the subsurface in a single (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018) and multirealization approach, all conditioned to the log data. In the single realization approach subsurface model, constructed in a previous study (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018; Section 2.3.4), the model domain includes two subdomains (∼2/3 western and 1/3 eastern, location of separation plane is marked with a dashed black line; Figure 1b) in order to increase hydrogeological realism, in which an evident difference exists in characteristics between two statistical subregions, as reflected in well logs' data (i.e., higher % of clay in the east; Figure 1c and Table 3).…”

“…Thus, the eastern part of the single validated realization from Ganot, Holtzman, Weisbrod, Bernstein et al. (2018) was inserted in all realizations to the east of the separation plane, replacing the coinciding generated subset of cells (Figure 2). Furthermore, the calibrated values of the hydrogeologic properties were adopted from the single realization model (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018) to the multirealization model.…”

“…The hydraulic and geostatistical properties of the generated subsurface realizations (Figure 2 and Table 3) were adopted (Section 2.3.1) from a single realization that was calibrated to 2015–2017 data (Ganot, Holtzman, Weisbrod, Bernstein, et al., 2018) and validated herein with the new 2018–2019 data. The single realization flow and transport model were run with MODFLOW (Harbaugh et al., 2017) and MT3DMS (Zheng & Wang, 1999) codes, respectively, with daily stress periods and the following boundary conditions: specified hydraulic head of 0 was set at the aquifer western boundary at sea level (due to the nonsensitivity of the model to western head boundary conditions, tidal conditions were not implemented) and a time‐varying head was specified at northern, southern, and eastern aquifer boundaries according to head observations at wells located in the vicinity of the boundaries.…”