Stimulating In Situ Hydrogenotrophic Denitrification with Membrane-Delivered Hydrogen under Passive and Pumped Groundwater Conditions

The field site was located in Becker, Minnesota, approximately 100 km northwest of Minneapolis, Minn. The aquifer was a surficial sand and gravel aquifer confined at the base by a dense glacial till approximately 18–46 m below land surface (Xcel Energy 2000; Northern States Power (NSP) 1991). The primary groundwater flow direction was from the northeast to southwest and the main discharge area was the Mississippi River (located approximately 200 m downgradient of the field site) (Xcel Energy 2000). The depth to groundwater at the site was approximately 15 m below the land surface.

The well placement at the field site is shown in Fig. 1. This well placement was chosen based on the general groundwater flow direction being perpendicular to the Mississippi River (Northern States Power (NSP) 1991). The well casings consisted of 5.1-cm outside diameter (OD) polyvinyl chloride (PVC) pipes and were screened over a length of 4 m, 14–18 m below the land surface. The aquifer material consisted of poorly graded sand and gravel with a mean grain diameter of 0.4 mm and porosity of 0.3. The groundwater was aerobic ( $7\pm 2\text{mg}$ dissolved oxygen (DO)/L, average  ± 95% confidence interval) and contained $21.2\pm 8.1\text{mg}$ ${\text{NO}}_3^{-}\text{-N}/\text{L}$ .

A bromide tracer test was conducted to determine the direction and velocity of the groundwater flow and the local longitudinal dispersion coefficient in the aquifer. Water was pumped from Well M2 into two 20-L carboys and mixed with NaBr to create a stock solution of 2,500 mg/L bromide $\left({\text{Br}}^{-}\right)$ . This solution was immediately pumped into Well H6 (approximately 1 m below the water table) at a rate of 1 L/min using a peristaltic pump. A ${\text{Br}}^{-}$ selective electrode (Cole Parmer, model #27502-04, Vernon Hills, IL) equipped with a data logger (WTW, pH340I, Germany) was positioned downgradient in Wells H2, H7, and M7 (approximately 1 m below the water table) to record hourly changes in the ${\text{Br}}^{-}$ concentration of the water.

Two membrane module designs were tested at the field site (Fig. 2). Both types of modules used 1-mm model # $\times 1.6\text{-mm}$ OD nonporous silicone-coated reinforced fiberglass membrane fibers (Varflex Corporation, #20RC1, Rome, N.Y.).

The first membrane module was designed to be operated in series [Fig. 2(a)] and each module consisted of 20 membrane fibers (3.0-m length). The top and bottom of the fibers were sealed (3M, Scotch Weld DP-125, St. Paul, Minn.) into 5.1-cm lengths of 2.5-cm OD stainless steel pipe. When operated in series, the exhaust gas from the one module was connected to the influent gas line of another module via Tygon tubing. Gas from the last module was vented to the atmosphere. This membrane module design was tested in the field under constant ${\text{H}}_2$ flow but was plagued by water condensation within the membranes, which prevented gas flow (results not shown). To overcome these limitations, a second membrane design was developed.

The second type of membrane module was designed with a sealed end [Fig. 2(b)]. A single membrane fiber, 24.0 m in length, was coiled to form a 3-m-long module (approximately 1 loop/cm). The influent gas line was connected to one end of the membrane and the other end was attached to a moisture escape membrane apparatus (MEMA). The MEMA was made with six polyethylene microporus membranes (5.1 cm long, 0.30-mm OD, $0.26\text{-}\mu \text{m}$ ID) (Mittsubishi Rayon, MHF 300, New York, N.Y.). The polyethylene membranes were sealed at the terminal end with DP-125 epoxy and wetted with methanol immediately before they were installed in the well. When the coiled membrane was pressurized, any condensed water inside the membrane was transported to the surrounding groundwater. A 4.4-cm solid stainless steel weight was also attached to the bottom of the membrane to compensate for the buoyancy of the membrane module and to keep the membrane fully extended $\left(\sim 3\text{m}\right)$ within the well.

Wells were sampled approximately every one to two weeks for ${\text{NO}}_3^{-}$ , ${\text{NO}}_2^{-}$ , DO, pH, temperature, total organic carbon (TOC), alkalinity, and water level. Samples were collected using a submersible 4.6-cm diameter pump (Grundfos, Redi-flow2 pump, Fresno, Calif.) into 250-mL Teflon-lined polyurethane bottles (Nalgene). Dissolved ${\text{H}}_2$ was not quantified at the site because it was determined that the submersible Grundfos pump produced ${\text{H}}_2$ by electrolysis during sampling; this has been observed by others using similar submersible pumps (Chapelle et al. 1997).

Microcosm experiments were conducted in the laboratory to determine kinetic parameters for the reduction of ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ for use in the model. Sediment was obtained from the upper 2 m of the saturated zone at the Becker site and was stored under aerobic conditions at $10°\text{C}$ prior to use.

Microcosms were prepared in triplicate in 715-mL glass bottles (Wheaton, Millville, N.J.) containing 30 g wet sediment, 297-mL synthetic groundwater [per 1 L Milli-Q (Millipore, Bedford, Mass.) water]: 23 mg ${\text{MgSO}}_4$ , 167-mg ${\text{MgCl}}_2\cdot 6{\text{H}}_2\text{O}$ , 247-mg ${\text{NaHCO}}_3$ , 13-mg ${\text{KHCO}}_3$ ), 3 mL of a trace element stock solution (per 1 L Milli-Q water: 8.5-mg ${\text{K}}_2{\text{HPO}}_4\cdot 3{\text{H}}_2\text{O}$ , 13.4-mg ${\text{MnCl}}_2\cdot 6{\text{H}}_2\text{O}$ , and 5-mg ${\text{FeCl}}_2\cdot 4{\text{H}}_2\text{O}$ ), and varying amounts of ${\text{NaNO}}_3$ or ${\text{NaNO}}_2$ (1–18 mg/L as N). The liquid in all microcosms was sparged with ultrahigh purity ${\text{N}}_2$ for approximately 2 min to remove DO. The headspace was flushed with 10% ${\text{H}}_2/90\%$ ${\text{N}}_2$ gas for approximately 30 min and the electron donor-free biological controls were flushed with 100% ${\text{N}}_2$ . The bottles were sealed with screw-on caps fitted with butyl rubber septa for sample collection (Wheaton, Millville, N.J.). Sterile controls were prepared by autoclaving microcosms containing a 10% ${\text{H}}_2/90\%$ ${\text{N}}_2$ headspace for 30 min. The microcosms were placed on their sides on a shaker table (New Brunswick Scientific, Model R-2, Edison, N.J.) and shaken at 250 rpm at $10°\text{C}$ for the duration of the experiment. A 3.5-mL liquid sample was withdrawn to quantify ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ concentrations every 1–3 days.

The DO concentrations were measured in the field using 1–12 mg/L Chemet ampules (Chemetrics Inc., Calverton, Va.). During the groundwater recirculation experiment, DO was measured with a DO probe (MI-730 Dip-type ${\text{O}}_2$ microelectrode) and meter (OM-4, oxygen meter) from Microelectrodes Inc. (Bedford, N.H.). Additionally, a YSI Model 58 (YSI Inc., Yellow Springs, Ohio) DO meter and probe with a 30.5-m lead were used to obtain in situ DO concentration profiles. Temperature and pH were measured in the field using a portable pH meter (Oakton, model-pH 6) and electrode (Oakton, model-WD-35801-00, Vernon Hills, Ill.). Water levels were measured using an electronic water level finder (Slope Indicator Co., Model 51453, Seattle, Wash.) before pumping the groundwater. Quantification of alkalinity (mg/L as ${\text{CaCO}}_3$ ) was determined by Standard Method 2320 B (American Public Health Association (APHA) et al. 1998). Nitrate and nitrite concentrations were determined by ion chromatography [Metrohm 761 Compact Ion Chromatograph (Herisau, Switzerland); Metrosep A Supp 5 column; 3.2-mM ${\text{Na}}_2{\text{CO}}_3$ and 1.0-mM ${\text{NaHCO}}_3$ eluent; 0.7 ml/min eluent flow rate; 100 mM ${\text{H}}_2{\text{SO}}_4$ regenerant]. The method detection limits (MDLs) for ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ were 0.035 mg/L ${\text{NO}}_3\text{-N}$ and 0.033 mg/L ${\text{NO}}_2\text{-N}$ , respectively, as determined by Standard Method 1030 C (American Public Health Association (APHA) et al. 1998). The TOC was analyzed in triplicate using a Dohrmann Phoenix 8000 TOC analyzer equipped with a STS 8000 autosampler (Dohrmann, Mason, Ohio). The MDL for TOC was 0.12 mg/L as carbon.

The mathematical model used in this study was an extension of an existing two-dimensional (2D) model that incorporated advective-dispersive transport and biodegradation of one electron donor (formate) and two electron acceptors ( ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ ) coupled with the growth and decay of a single microbial population (Killingstad et al. 2002). The Killingstad et al. (2002) model was modified to simulate constituent concentrations in a real, 2D domain in which concentrations were assumed to be uniform with depth, and to account for the transfer of ${\text{H}}_2$ through hollow-fiber membranes, according to the gas transfer correlation developed by Fang et al. (2002). The model was also modified to model three electron acceptors ( ${\text{O}}_2$ , ${\text{NO}}_3^{-}$ , and ${\text{NO}}_2^{-}$ ). The microbial population was a single facultative population attached to sediment particles. The population occupied no physical space within the model domain.

The general form of the 2D equations of mass balance for electron donor and acceptor transport and biodegradation arerespectively, where $H=\text{aqueous}$ phase concentration of hydrogen $\left(\text{M}/{\text{L}}^3\right)$ ; $E=\text{electron}$ acceptor concentration $\left(\text{M}/{\text{L}}^3\right)$ ; $x$ and $y$ are distances along and across the direction of flow, respectively (L); $v_x=\text{average}$ groundwater velocity (L/T); $D_x$ and $D_y$ are the dispersion coefficients in the $x$ - and $y$ -directions, respectively $\left({\text{L}}^2/\text{T}\right)$ ; $R_{\text{bio},H}^{\text{sin}k}$ represents the mass loss of ${\text{H}}_2$ due to biological utilization $\left(\text{M}/{\text{L}}^3/\text{T}\right)$ ; $R_{\text{flux},H}^{\text{source}}$ accounts for the mass flux of ${\text{H}}_2$ from the membranes $\left(\text{M}/{\text{L}}^3/\text{T}\right)$ ; $R_{\text{bio},E}^{\text{sin}k}$ represents the mass loss of electron acceptor due to biological utilization $\left(\text{M}/{\text{L}}^3/\text{T}\right)$ ; and $R_{\text{bio},E}^{\text{source}}$ accounts for the production of ${\text{NO}}_2^{-}$ , according to the stoichiometry in Eq. (9).

The source term for the mass transfer of ${\text{H}}_2$ to the aqueous phase $\left(R_{\text{flux},H}^{\text{source}}\right)$ is expressed as the mass transfer rate per unit volume of liquidwhere $M_H=\text{mass}$ transfer rate of ${\text{H}}_2$ out of the membrane (M/T); ${\theta }_w=\text{porosity}$ of the mass transfer zone $\left({\text{L}}^3/{\text{L}}^3\right)$ ; and $V_T=\text{total}$ volume of the mass transfer zone $\left({\text{L}}^3\right)$ (i.e., one model cell). The ${\text{H}}_2$ mass transfer rate under steady state conditions is expressed aswhere $K=\text{overall}$ ${\text{H}}_2$ mass transfer coefficient (L/T); $A=\text{surface}$ area of the membrane $\left({\text{L}}^2\right)$ ; $\omega =\text{molecular}$ weight of ${\text{H}}_2$ (M/mole); and $H_L^{\ast }={\text{H}}_2$ liquid saturation concentration $\left(\text{mol}/{\text{L}}^3\right)$ , which is related to the ${\text{H}}_2$ partial pressure by the Henry’s Law constant. The value for $K$ can be predicted using the following dimensionless relationship (Fang et al. 2002):where $D=\text{diffusivity}$ of ${\text{H}}_2$ in water $\left({\text{L}}^2/\text{T}\right)$ ; $d=\text{outside}$ membrane diameter (L); ${\upsilon }_w=\text{water}$ velocity across the membranes (L/T); and $\nu =\text{kinematic}$ viscosity of water $\left({\text{L}}^2/\text{T}\right)$ .

The reaction term for ${\text{H}}_2$ utilization $\left(R_{\text{bio},H}^{\text{sin}k}\right)$ , expressed as the sum of utilization as a result of DO, ${\text{NO}}_3^{-}$ , and ${\text{NO}}_2^{-}$ reduction, iswhere $M=\text{autotrophic}$ biomass per volume of porous medium $\left(\text{M}/{\text{L}}^3\right)$ ; $\theta =\text{aquifer}$ porosity; and ${\nu }_{H,O}$ , ${\nu }_{H,N_1}$ , and ${\nu }_{H,N_2}$ are the ${\text{H}}_2$ utilization rates by the autotrophic biomass for DO, ${\text{NO}}_3^{-}$ , and ${\text{NO}}_2^{-}$ reduction, respectively (M/M per T). The DO, ${\text{NO}}_3^{-}$ , and ${\text{NO}}_2^{-}$ mass loss rates $R_{\text{bio},E}^{\text{sin}k}$ , resulting from their utilization as an electron acceptor, are linked to ${\text{H}}_2$ utilization by the stoichiometry of the biological reductionwhere ${\gamma }_E$ represents the mass of electron acceptor reduced per unit mass of hydrogen (M/M) and $E$ represents either DO, ${\text{NO}}_3^{-}$ , or ${\text{NO}}_2^{-}$ . Utilization rates in Eqs. (5) - (6) are represented by dual Monod kinetics of the electron donor and acceptor. Switching functions prevented the uptake of ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ until DO was below a set level ( ${\kappa }_{O,N_1}$ and ${\kappa }_{O,N_2}$ , respectively), and prevented the degradation of ${\text{NO}}_2^{-}$ until ${\text{NO}}_3^{-}$ was below a set level $\left({\kappa }_{N_1,N_2}\right)$ (Killingstad et al. 2002).

Changes in the microbial densities are simulated by a mass balance equation that accounts for biomass growth and decay termswhere $Y_O$ , $Y_{N_1}$ , and $Y_{N_2}$ are the yield coefficients (M/M), the ratio of microbial biomass produced per mass of electron donor consumed for DO, ${\text{NO}}_3^{-}$ , and ${\text{NO}}_2^{-}$ respiration, respectively, and $k_d$ is the biomass decay coefficient (M/M per T). Eq. (7) is solved directly and is updated at each time step.

A 2D finite-difference grid was used with a 0.0508-m node spacing. Initial conditions for concentrations of ${\text{NO}}_3^{-}$ , ${\text{NO}}_2^{-}$ , and DO in the model were determined by sampling data collected just prior to ${\text{H}}_2$ addition and were assumed to be uniform throughout the model domain. The initial condition for ${\text{H}}_2$ concentration was assumed to be zero throughout the model domain.

Concentrations of ${\text{NO}}_3^{-}$ , ${\text{NO}}_2^{-}$ , DO, and ${\text{H}}_2$ were assumed to be constant with time and uniform over the width of domain at the inflow boundary. The lateral, upper, and lower boundaries were represented as no-flow boundaries, and the downgradient boundary was set as a zero gradient condition. Hydrogen source nodes were added at the well locations to represent ${\text{H}}_2$ addition via membrane modules.

Input parameters for the model were obtained from the field site tracer study, microcosm and laboratory experiments, stoichiometric relationships, and from the literature. No parameters were fit to the field data. Table 2 is a complete listing of input parameters used in the model. Model simulations included a site-wide analysis showing the entire field site study area and model runs focused on the apparent flow path through Wells H3 and M5.

Average values for each water quality parameter measured from all wells at the site before ${\text{H}}_2$ addition are shown in Table 3. The average ${\text{NO}}_3^{-}$ and DO concentration before ${\text{H}}_2$ addition was $21.2\pm 8.1\text{mg}/\text{L}$ as N and $7.0\pm 2.0\text{mg}/\text{L}$ , respectively (± 95% confidence interval). Nitrite was never detected in the field before the ${\text{H}}_2$ addition. The pH, temperature, TOC, and alkalinity were $7.2\pm 0.4$ , $11.4\pm 2.4°\text{C}$ , $0.6\pm 0.4\text{mg}/\text{L}$ as C, and $182\pm 14\text{mg}/\text{L}$ as ${\text{CaCO}}_3$ , respectively.

Although oxygen was observed in samples containing ${\text{NO}}_2^{-}$ , it is assumed that this was a result of oxygen mixing into the sample during sample withdrawal rather than the simultaneous reduction of oxygen and ${\text{NO}}_3^{-}$ . The sequential reduction of oxygen followed by ${\text{NO}}_3^{-}$ was confirmed in the Becker aquifer material by Schnobrich et al. (2007) and by the microcosm studies.

The greatest quantity of denitrification was observed in Well M5 during Experiment 1 in which concentrations of DO and ${\text{NO}}_2^{-}$ were 2.0 mg/L and 8.0 mg/L-N, respectively. According to the stoichiometry shown in Eqs. (8) - (10), approximately 1.74 mg/L ${\text{H}}_2$ is needed for this observed reduction. This quantity of ${\text{H}}_2$ is comparable to the solubility of ${\text{H}}_2$ at $10°\text{C}$ (1.73 mg/L) and the concentration of ${\text{H}}_2$ $\left(\sim 1.5\text{mg}/\text{L}\right)$ that is predicted by the clean water gas transfer correlation developed by Fang et al. (2002). Numerous studies have shown that biofilm growth on a membrane surface results in a gas transfer rate as high as 14 times that predicted by clean water gas transfer correlations (Semmens and Essila 2001; Haugen et al. 2002; Edstrom et al. 2005; Schnobrich et al. 2007). The ${\text{H}}_2$ needed to reduce DO and ${\text{NO}}_3^{-}$ and ${\text{NO}}_2^{-}$ below their respective MCLs at the site is approximately 6.6 mg/L ${\text{H}}_2$ , only approximately four times higher than the clean water gas transfer correlation. The residence time within the ${\text{H}}_2$ -delivery wells ( $\sim 1.0–3.0\text{h}$ , depending on the well capture zone), should be sufficient for microbial consumption to increase the overall ${\text{H}}_2$ mass transfer out of the membrane. In fact, laboratory results from Schnobrich et al. (2007), using aquifer material from the Becker site with a nearly identical flow velocity, slightly shorter residence time (1.45 h), and slightly more (3.5 to 10 times greater) membrane surface area per groundwater flow rate, demonstrated that it was possible to add sufficient ${\text{H}}_2$ to meet the electron donor demand at the Becker field site. Indeed, the flow velocity, and therefore the Reynolds number, in the well bore was expected to be approximately three times that in the laboratory experiment by Schnobrich et al. (2007). We would therefore expect enhanced ${\text{H}}_2$ transfer from the membrane to the surrounding water in the well, resulting in an increased extent of denitrification in the well itself (Fang et al. 2002). Nevertheless, the lack of more significant denitrification within the membrane wells themselves (H2, H3, H4, H7, H8, and H6), indicated that other factors at the site, such as low phosphorus concentrations (Schnobrich et al. 2007), may have contributed to incomplete denitrification.

Because the zone of influence of each membrane-containing well was relatively narrow (Haugen et al. 2002; Clapp et al. 2004), lower levels of denitrification would be observed downgradient as treated water mixed with untreated water from the surrounding area, resulting in higher observed ${\text{NO}}_3^{-}$ concentrations. In addition, slight shifts in the direction of flow could have profound impacts on observations at downgradient wells. Therefore, modeling studies were used to help determine how to improve performance in the field.

Although the membranes installed at the site were successful in delivering ${\text{H}}_2$ to the groundwater during Experiments 1 and 2, the zone of influence was limited and the overall mass of ${\text{H}}_2$ delivered was insufficient to stimulate complete denitrification; this was confirmed by the modeling simulations. To address these problems, water recirculation was initiated to increase ${\text{H}}_2$ gas transfer by increasing the velocity of the water flowing past the membranes [see Eq. (4)].

During the 91 days of pumping operation, ${\text{NO}}_3^{-}$ , ${\text{NO}}_2^{-}$ , and DO concentrations fluctuated, as shown in Fig. 7. Total nitrogen concentrations between the upgradient Well M2 $\left(23.4\pm 0.9\text{mg-N}/\text{L}\right)$ and Well M5 $\left(23.4\pm 0.5\text{mg-N}/\text{L}\right)$ were not significantly different. Nitrite accumulation reached a maximum concentration of approximately 8.0 mg ${\text{NO}}_2^{-}\text{-N}/\text{L}$ on Day 62, four days after installing the flow distributor. Profiles of DO concentrations taken in situ are shown in Fig. 8. The low DO concentrations observed on Day 62 are indicative of microbial activity. Between Days 62–73, however, air was observed entering the manifold through one of the delivery tubes (Well H1), and as a result, pumping to Well H1 was stopped. Once pumping to H1 stopped, the DO profiles obtained for M5 on Day 78 showed that ${\text{O}}_2$ entrainment had ceased. The reduction of 4.0 mg ${\text{NO}}_3^{-}\text{-N}/\text{L}$ to ${\text{NO}}_2^{-}$ was observed in Well M5 at this time. In general, the pump installed at Well M7 appeared to draw a significant amount of water from untreated regions of the aquifer. The mixing of DO- and ${\text{NO}}_3^{-}$ -rich groundwater with treated water prevented consistent observation of ${\text{NO}}_3^{-}$ reduction at, and downstream of, membrane-containing wells H6, H3, and H4. As a result, the ${\text{H}}_2$ required for reducing ${\text{O}}_2$ and ${\text{NO}}_3^{-}$ at the site increased.

The membrane modules were removed on two occasions, Days 58 and 91. On Day 58 approximately 5.5 mg ${\text{NO}}_2^{-}\text{-N}/\text{L}$ and 1.1 mg/L DO were observed in Well H3. On Day 91, approximately 8.5 mg ${\text{NO}}_2^{-}/\text{L}$ was observed in Well H4. These results confirm that membrane delivery of ${\text{H}}_2$ was still able to reduce DO and ${\text{NO}}_3^{-}$ in these wells despite the reintroduction of ${\text{O}}_2$ and ${\text{NO}}_3^{-}$ -rich groundwater.

Agarwal et al. (2005) tested the effects of pumping on membrane gas $\left({\text{O}}_2\right)$ transfer and the zone of influence in a laboratory tank reactor filled with fine sand. Under passive conditions, with a flow velocity of 1 m/day, the distance between the 1 mg/L ${\text{O}}_2$ contour was 7.3 times the diameter of the membrane module well (Agarwal et al. 2005). At pumping rates of 1, 20, and 40 mL/min into the membrane well, however, the zone of influence was found to be approximately 2.4, 11.0, and 15.7 times greater than passive operation, respectively (Agarwal et al. 2005). These results suggest that the dispersion of gas from a hollow fiber membrane should increase with groundwater recirculation if it is performed carefully to avoid reoxygenation.