Integration of Artificial Recharge and Recovery Systems for Impaired Water Sources in Urban Settings: Overcoming Current Limitations and Engineering Challenges

Hydrological and physical limitations

Aquifer storage capability is mainly determined by the spatial extent and composition of the aquifer, as well as its degree of confinement and connectivity to streams and other aquifers (David and Pyne, 1995; National Research Council, 2008; Maliva and Missimer, 2010). Subsurface conditions in urban settings can be particularly challenging due to a higher degree of soil compaction leading to reduced permeability, channel structures due to existing or removed subsurface infrastructure (e.g., pipes, sewer or power lines), high anisotropy, and heterogeneity (Hibbs and Sharp, 2012). If different hydrogeological parameters like the hydraulic gradient, permeability, and mineralogy of the aquifer as well as ambient groundwater quality are not well understood (Bouwer, 2002; Sharma and Amy, 2011), an insufficient site characterization can lead to the poor performance of an ARR system, that is, low recharge rates, low storage capacity, and low yield (Maliva and Missimer, 2010).

Kloppmann et al. (2012) defines recovery efficiency (RE) of an ASR system as the percentage of water that can be recovered: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}RE = \frac { Vol_ { rec } } { Vol_ { inj } } \tag { 1 } \end{align*} \end{document}

where Vol_rec is the cumulative volume of water recovered after some storage time and Vol_inj is the total volume of water injected. While this is a reasonable approach, quantifying RE for an ARR system is challenging since recovery must be qualified by quantity and quality characteristics that distinguish contributions from native groundwater and recharged water (Pyne, 1995; Reese, 2002; Sheng, 2005).

Recharging water by spreading water in ponds or basins for infiltration through the vadose zone has become a common practice in the United States because the installation and operation is cheaper than injection wells. In addition, unconfined aquifers, where surface infiltration is practiced, may have a larger storage capacity than confined aquifers (National Research Council, 2008). However, the surface infiltration system may create localized water mounding, which is a rise in the water table under the pond that is particularly pronounced if the transmissibility of the saturated zone is low. In addition, the infiltration system may also create perched water tables, which signify the appearance of saturated zones above the regional water table as water accumulates on top of low-conductivity layers or lenses (Bouwer et al., 1999). Stratification in the vadose zone may induce significant lateral flows that may be hard to predict or control. Mounding and perched aquifers may in some cases result in reduced infiltration rates as the gradient in the potential that drives water flow is reduced. The mounding effect can change the direction and magnitude of regional groundwater flow. In this case, stored water may flow back into a neighboring stream, discharge into brackish or saline waters in coastal regions, or actuate recharge of other connected aquifers with the consequence that some portion of the injected water is effectively lost for augmenting the water supply or cannot be recovered without installation of additional recovery wells. Analogous challenges are encountered in the design and operation of injection wells.

Reliable and cost-effective operation of surface infiltration requires monitoring of the hydrogeological structure, the recharge rates around the recharge locations as well as the hydraulic gradient over an ARR site. This can be accomplished through infiltration tests (Bouwer et al., 1999), geophysical methods (Haines et al., 2009; Maliva et al., 2009; Nenna et al., 2011), or thermal sensors (Anderson, 2005; Healy and Scanlon, 2010; Saar, 2011). In practice, groundwater mounds can be alleviated by reducing recharge rates, arranging infiltration ponds in long, narrow strips instead of compact round or square areas, dispersing ponds over large areas, and operating nearby groundwater pumping wells (National Research Council, 2008). Injection wells are used for recharging water directly into aquifers, where available land is scarce or infiltration through the vadose zone is not feasible due to the presence of low-permeability layers overlying a water-stored formation.

During water infiltration into an ARR system, physical, biological, chemical, and/or gas-introduced clogging of soils and compaction of the created clogging layer (Pérez-Paricio and Carrera, 1998) can dramatically reduce the hydraulic conductivity of in-fill pore spaces in the aquifer (Bouwer, 2002). Gas entrainment caused by chemical reactions or subsurface biological processes disrupts the continuity of water-filled pore spaces (Bouwer and Rice, 1989). During infiltration, particulate matter in water can be entrapped in the pore spaces (Konikow et al., 2001; Pavelic et al., 2007). In addition, the density difference between recharged water and native groundwater may generate a density-induced flow (Ward et al., 2007, 2008, 2009) in an aquifer with high hydraulic conductivity so that some portion of the recharged water may never be recovered. Chemical reactions related to the introduction of dissolved oxygen in an aquifer during infiltration or injection may result in the precipitation of insoluble products such as manganese oxides and iron hydroxides (Moorman et al., 2002). Anoxic to anaerobic conditions in the aquifer can mobilize manganese, iron, and other inorganic trace elements. Biomass growth, known as bioclogging or biofouling, can occur through introduction of elevated levels of substrates and nutrients, which can plug the pore spaces of the infiltration zone (Albrechtsen et al., 1998; Baveye et al., 1998).

To reduce the degree of clogging in the infiltration zone in practice, the near-subsurface area is agitated mechanically by scrapers, flushed by sand-washing devices, or maintained by applying wetting and drying cycles as a self-cleaning mechanism (National Research Council, 2008; Grützmacher and Reuleaux, 2011; Sharma and Amy, 2011). To maintain infiltration rates in injection wells, periodic backflushing and well rehabilitation methods such as shock disinfection by chlorination (Houben and Treskatis, 2007) need to be performed to avoid solids buildup and biofouling (Pyne, 1995). For injection wells not being in use, providing a steady slow flow of chlorine to maintain a chlorine residual is also critical. Reduction of the concentration of organic and colloidal material in pretreatment systems before recharge may prevent or minimize clogging from the outset.

Water quality issues of current ARR installations

Percolation and infiltration of reclaimed water through the vadose zone takes advantage of the natural attenuation capability of the subsurface. However, open recharge basins can experience excessive algae blooms when spreading nutrient-rich water, water losses due to evaporation, and occasionally an increase in mosquitos (Grützmacher and Reuleaux, 2011). After infiltration, the water quality in some ARR systems has been shown to be degraded by enhanced mixing (Eastwood and Stanfield, 2001; Lowry and Anderson, 2006), limited mass transfer (Culkin et al., 2008; Lu et al., 2011), and metal leaching/acid rock drainage (Mirecki, 2004; Arthur, 2005).

ARR systems are highly effective for the removal of turbidity, nutrients, organic matter, pathogens, and regulated and unregulated trace organic chemicals (Maeng et al., 2010; Bekele et al., 2011; Fox and Makam, 2011; Hogg et al., 2011; Zhang et al., 2011). Less efficient removal during ARR has been reported for specific pathogens and recalcitrant trace organic chemicals present in reclaimed water applied to recharge basins (Drewes et al., 2003; Cordy et al., 2004; Heberer et al., 2004). The detection of a variety of trace organic contaminants in municipal wastewater effluent (Richardson and Ternes, 2011) has raised concerns about the potential presence of wastewater-derived contaminants in water produced by ARR systems (Díaz-Cruz and Barceló, 2008; Drewes et al., 2008). The state of science for managing these chemicals of emerging concern and pathogens is addressed in the following section.

Factors influencing removal of pathogens and trace organic contaminants

Biotransformation of organic contaminants in ARR systems is dependent on multiple factors (Díaz-Cruz and Barceló, 2008). Previous studies have pointed to the redox environment of ARR systems as a primary driver for mobility, dissolution, degradation/transformation, and toxicity of organic as well as inorganic chemicals present in infiltrating water (McMahon and Chapelle, 2008; Farnsworth and Hering, 2011; Grützmacher and Reuleaux, 2011; Stuyfzand, 2011; Wiese et al., 2011). The impact of other physicochemical parameters like pH, ionic strength, and temperature on attenuation processes are estimated to play a significant though smaller role (Stuyfzand, 2011). Recent research has revealed that the removal of trace organic contaminants is also dependent upon the availability of biodegradable dissolved organic carbon (Rauch-Williams et al., 2010; Li et al., 2013), the concentration of trace organic chemicals in the applied water (Hoppe-Jones, 2012), and the degree of adaptation of the microbial community to the presence of trace organic contaminants (Rauch-Williams et al., 2010). As many subsurface reactions are time dependent, the residence time of groundwater is generally one of the most important controlling factors related to contaminant attenuation (Wiese et al., 2011). Rhine et al. (2003) reported that repeated or continuous exposure to trace organic contaminants can enhance biotransformation due to the selection of more adapted microorganisms. For hydrophobic trace organic contaminants, sorption onto organic matter and metal oxides can also play an important role in attenuation in subsurface systems as the retardation allows more time for biotransformation (Tolls, 2001; Higgins and Luthy, 2006). However, if sorption sites are limited, this does not provide a sustainable removal pathway.

Microbiologically driven transformation of contaminants in the subsurface occurs naturally and, therefore, is not site-specific or unique to certain ARR locations. Rather, microbial processes are determined by geochemical and hydrological subsurface conditions (Haack and Bekins, 2000). A recent study supports the presence of similar microbial communities across both laboratory and field scales in geographically distinct, but geochemically analogous settings (Li et al., 2012). Despite advances in understanding microbes and enzymes involved in biotransformation of trace organic chemicals at the single-strain, laboratory scale (Sharp et al., 2007, 2010), the composition, interaction, and function of microbial communities linked to effective attenuation is still an evolving field of research (Haack and Bekins, 2000; Lovley, 2003; Shade et al., 2009). In addition to molecular and microbial inquiries, automated sensing technologies can be helpful to direct the response of the microbial community to key boundary conditions (e.g., temperature, pH, dissolved oxygen, nutrient availability) in aquatic ecosystems (Shade et al., 2009).

Capturing the microbiological function through removal rate constants is important for the development of appropriate contaminant transport models that can assist in better design and operation of ARR facilities. While removal kinetics of trace organic chemicals can be determined using laboratory-scale one-dimensional (1D) soil column experiments, multiple mechanisms (e.g., hydrolysis, sorption, and biotransformation) occurring in these systems can contribute to contaminant attenuation, which limits the transferability of rate constants (Wiese et al., 2011). In addition, some laboratory-scale studies have employed influent trace organic chemical concentrations that are orders of magnitude higher than environmentally relevant concentrations (Patterson et al., 2011) resulting in degradation rate constants that are not always representative of ARR field conditions.

Based on previous biotransformation models, column studies, and field data, several authors have hypothesized that a threshold concentration for specific trace organic contaminants exists below which transformation of these compounds does not take place (Bouwer and McCarty, 1984; Grützmacher and Reuleaux, 2011; Wiese et al., 2011). However, recent results presented by Baumgarten et al. (2011) for sulfamethoxazole attenuation during ARR imply that proposed threshold values might be a matter of insufficient adaptation time of the microbial community in geomedia rather than the existence of a true threshold concentration.

Pathogen removal and deactivation by ARR occurs primarily by filtration/adsorption and die-off within the soil matrix (Gerba et al., 1991). Pathogen removal rates are specific to the physical and chemical properties of the microbes, subsurface media, solution chemistry, transport scale, type of source water (i.e., wastewater vs. surface water), and duration of contamination (years of operation) (Pang, 2009). Temperature appears to be a key parameter for inactivation in these systems (Gerba et al., 1991). Due to their small size (i.e., ability for longer distance subsurface travel), their low infectious dose (i.e., smaller numbers are of greater concern), and their potential for long-term persistence in groundwater, viruses are the pathogens of major concern during an ARR operation. Hepatitis A virus, adenoviruses, and parvoviruses appear to be the most thermally stable waterborne pathogens, although bacteria as well as protozoa may persist for long periods of time if environmental conditions are favorable (Gerba, 2007). In attending to model pathogen transport in soil-aquifer systems, the controlling factors are die-off, retardation, velocity, hydrodynamic dispersion, filtration, adsorption, and soil saturation. Furthermore, die-off of organisms is not always linear (i.e., rapid inactivation in the early beginning followed by a much slower inactivation rate), especially at low concentrations and this must be taken into consideration in any attempts to estimate inactivation rates (Gerba et al., 1991; Pang, 2009). Pang (2009) proposed a modified equation for the estimation of pathogen removal at ARR sites: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}N = H_v \lambda_v + L_a \lambda_a \tag{2}\end{align*} \end{document}

where N is the total log reduction of the pathogen between the source and extraction, H is the thickness of the vertical travel distance in meters, L is the horizontal travel distance in meters, and λ is the removal rate in log/m. The subscripts represent the removal in the subsurface, v and a represent the differences in removal between vertical (vadose zone) and horizontal transport.

Modeling approach for engineered ARR systems

In ARR systems, groundwater flow and storage is mostly affected by recharge and recovery regimes as well as existing regional flow conditions in the vadose and saturated zones. In general, depending on the degree of heterogeneity, the flow in a variably saturated aquifer is 3D. If a vertical path with no significant horizontal flow is assumed, it may be described by the 1D Richards equation (Richards, 1931) as follows: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}\left( C ( \psi ) + \frac { \theta } { \theta_s } S \right) \frac { \partial \psi } { \partial t } = \frac { \partial } { \partial z } \left( K ( \theta ) \left( \frac { \partial \psi } { \partial z } + 1 \right) \right) + q \tag { 3 } \end{align*} \end{document}

where ψ is the pressure head [L], C(ψ)=∂θ/∂ψ is the specific moisture capacity [1/L], θ is the soil water content [-], θ_s is the saturated water content [-], S is the specific storativity [1/L], K is the hydraulic conductivity [L/T], z is the vertical coordinate [L], and q is the recharge/discharge term [L³/T].

Recharge or recovery components are suitably included in q as boundary conditions of the Richards equation for ARR systems (Kuehn and Mueller, 2000; Ray, 2002). To simulate horizontal flow, the above equation can be expanded into multiple dimensions. Eq. (3) is highly nonlinear due to the dependence of both the hydraulic conductivity and the pressure head on the water content that varies with time and is typically solved numerically with a suitable numerical scheme (Feddes et al., 1988; van Dam and Feddes, 2000). Temporal infiltration flux, total infiltrated water volume to the underlying aquifer, and the leakage of recharged water can be computed from the numerically derived pressure head distribution. The Richards equation is an approximate model for a complex process. The predictive capability of this model is limited by the presence of heterogeneity that may cause complex 3D flow patterns. Characterization of this heterogeneity remains a challenge.

Besides modeling reliable groundwater resource storage, the prediction of water quality is another important element of actively managing ARR systems. A solute transport model for ARR should be capable of simulating the movement, mixing, and reaction of dissolved chemicals (Kirkner and Reeves, 1988) in the aquifer when linked with the groundwater flow model. In its simplest form, the transport of a solute under isothermal and 1D isodensity flow includes the processes of advection, dispersion, and retardation. The following equation captures these processes: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}R \frac { \partial C } { \partial t } = D \frac { \partial } { \partial x } \left( \frac { \partial C } { \partial x } \right) - \frac { \partial vC } { \partial x } + \frac { q_s } { \eta } C_s + r \tag { 4 } \end{align*} \end{document}

where R is the retardation factor [-], C is the concentration [M/L³], q is the water flux field [L/T] obtained from the groundwater flow model Eq. (3), D is the dispersion coefficient (tensor in 2D or 3D space) [L²/T], θ is porosity [-], q_s is the flux of the source/sink of concentration C_s [M/L³T], η is porosity [-], and r is a reaction or biotransformation term [M/L³T], which usually is assumed to follow first-order kinetics.

Multicomponent transport models that solve mixed equilibrium and kinetic geochemical reactions such as mineral precipitation/dissolution and ion exchanges (Hunter et al., 1998; Parkhurst and Appelo, 1999; Brun and Engesgaard, 2002; Prommer et al., 2003) can be expanded from Eq. (4). For rate-limited mass transfer, that is, the solute exchanges between mobile and immobile domains due to aquifer heterogeneity and preferential flow, a dual-domain ARR simulation needs to be employed (Culkin et al., 2008; Lu et al., 2011).

While mathematical equations of hydrogeochemical processes such as flow or equilibration of different element species are relatively simple, modeling the activity and metabolism of microorganisms under changing environmental conditions is more complex. At present, inclusion of microbially facilitated reactions relies on biotransformation rate constants. These kinetic constants are generally fixed values calculated from laboratory-scale studies as described above or estimated for the environment based on available biotransformation data (Lovley, 2003). In addition to biochemical limitations, diffusion controlled mass transfer can be a rate-limiting process. As a consequence, selected modeling efforts may not predict biogeochemical reaction rates correctly without field-specific calibration, or at least careful consideration of integrated biotransformation, transport, and mixing rates.

Clogging in recharge and recovery paths in an aquifer is another operational challenge to be incorporated into numerical modeling for better understanding and prediction of spatial and temporal evolution of an ARR system. Clogging can be modeled by porosity changes as a result of the attachment of particles (Herzig et al., 1970; McDowell-Boyer et al., 1986), precipitation of minerals, and biomass growth (Molz et al., 1986; Taylor et al., 1990; Clement et al., 1996; Thullner et al., 2004). Integration of these physical, chemical, and biological clogging processes into multiphase flow transport has also been developed and applied (Pérez-Paricio and Carrera, 1998; 2000). However, the resultant codes are complex and require a large amount of data and long simulation times; thus, their applicability has been limited (National Research Council, 2008). Instead, an empirical approach to approximate physical, chemical, and microbial clogging can be obtained by using laboratory and field measurements (Pavelic et al., 2007), which is more frequently used in practice. A combined water flow and contaminant transport model that can predict accurate water supply and quality changes within an ARR system has not yet been successfully developed. It is widely recognized that flow and reactive transport models are not able to predict physical, chemical, and biological processes accurately, although they can be useful for evaluating existing or new hypotheses on those interactions with laboratory- and field-scale experiments. Furthermore, while flow and transport through saturated and unsaturated laboratory-scale column setups using porous media have been well studied, experimentally validated models at the laboratory scale may not be applicable to problems at local or regional scales, where ARR is implemented in practice. Therefore, uncertainty in the numerical model should be considered (Kitanidis, 2007) with a need for recalibration and refinement as data become available during operation. To this end, the design of sampling networks, including monitoring wells or implementing real-time (Cheng et al., 2011), nonintrusive geophysical methods (Haines et al., 2009; Maliva et al., 2009; Nenna et al., 2011) can be helpful for improvement and optimization of process model outputs.

Geophysical techniques provide nonintrusive methods for groundwater exploration and monitoring on scales useful for ARR placement in urban settings, design, and maintenance decisions. Geophysical exploration for possible ARR site locations is possible at a regional scale using airborne electromagnetics. Airborne electromagnetics can detect lateral changes in apparent conductivity due to changes in sediments or identify the thickness of the unsaturated zone (Paterson and Reford, 1986) and clay layers (Gamey et al., 1996; Puranen et al., 1999).

Surface geophysical methods are useful on a smaller scale for assessing a location's viability for ARR. A ground penetrating radar can be used to map stratigraphy and geological features (Davis and Annan, 1989), delineate areas that hinder water flow, and map the water table and/or perched water tables (van Overmeeren, 1998). Electrical resistivity tomography and imaging can be used to identify clays (Fukue et al., 1999), identify macro- and microporosities and permeabilities (Griffiths, 1976; Robain et al., 1996), identify preferential flow pathways (al Hagrey et al., 1999), and characterize horizontal and vertical heterogeneities (Banton et al., 1997; Tabbagh et al., 2000).

Once a location has been identified as a possible ARR site, borehole geophysics can be used to perform a more localized analysis of the subsurface allowing for the detection of finer scale heterogeneities that is not otherwise achievable via other hydrologic or geophysical methods (Maliva et al., 2009). Neutron–gamma ray methods can detect iron sulfides (pyrite) and other minerals that may have a large impact on geochemical reactions in the subsurface (Maliva et al., 2009; Maliva and Missimer, 2010). These methods can also delineate between clay-rich soils and clean soils, which allows for the detection of thin clay layers, which could impede infiltration (Serra et al., 1980). Microresistivity imaging logs can measure porosity and detect fractures 360 degrees around the borehole at a scale as small as 50 μm (Maliva et al., 2009). Nuclear magnetic resonance logs map porosity and pore-size distribution and provide information about permeability. These logs can be used to identify the preferential flow and confining zones that are important to the storage of recharged water (Maliva et al., 2009; Maliva and Missimer, 2010).

Because they are not intrusive and are capable of being automated, geophysical methods can be effective tools for monitoring ARR processes. On a regional scale, interferometric aperture radar can be used to monitor hydraulic head changes and the effects of seasonal recharge and pumping (Bell et al., 2008; Reeves et al., 2011). This allows monitoring of the extent and effect of groundwater mounding. Locally, electrical resistivity imaging is an effective method for monitoring the vadose zone and can be related to saturation. Infiltration (Daily et al., 1992; Binley et al., 2001; Mitchell et al., 2011) or diurnal fluctuations in electrical conductivity due to temperature changes can be treated as a thermal tracer to estimate infiltration (Pidlisecky and Knight, 2011). The direct push probes developed and described in Pidlisecky and Knight (2011) and Mitchell et al. (2011) can monitor saturation remotely in real time, allowing for the automated estimation of infiltration and degree of pond clogging.

Intensive computational methods are required to optimize operation decisions in the management of large-scale, complex ARR systems in the presence of uncertainty, which provide dynamic updating of aquifer parameters through recalibration and corresponding decisions in response to incoming data. However, the use of coupled large-scale numerical models for aquifer parameter characterization/model calibration, adaptive ARR operation design, and uncertainty analysis on a fine temporal and spatial resolution scale requires significant computationally intensive efforts. For this reason, model reduction techniques (Antoulas, 2005) have been highlighted in subsurface modeling to develop an approximate numerical model that reduces the cost of simulations, while not losing the accuracy of results (Vermeulen et al., 2004; McPhee and Yeh, 2008). Along with this computationally time-saving modeling approach, recent advances in high-performance computing power and efficient data assimilation techniques enable us to handle large-scale flow transport models with real-time hydrogeological monitoring and high-resolution geophysical data and to compensate for errors in the flow transport model and ARR site characterization. Integrated ARR management modeling can provide a way to understand the subsurface system, predict the ARR system performance, and evaluate design options for management.