# Journal of Environmental Quality - Article

1. Vol. 41 No. 3, p. 724-731
OPEN ACCESS

* Corresponding author(s):
View
Permissions
Share

doi:10.2134/jeq2011.0331

# Evaluation of a Denitrification Wall to Reduce Surface Water Nitrogen Loads

1. Casey A. Schmidt * and
2. Mark W. Clark
1. Dep. of Soil and Water Science, Univ. of Florida, 106 Newell Hall, Gainesville, FL 32611. Assigned to Associate Editor Christopher Green

## Abstract

### Abbreviations

DO, dissolved oxygen; DOC, dissolved organic carbon; TKN, total Kjeldahl nitrogen

In 1909, with the successful demonstration of the Haber–Bosch process, the triple-bond of elemental N was broken by human action, and today this process has now surpassed worldwide bacterial N fixation (Galloway et al., 2003). This ushered in an increase in food production to such an extent that it is estimated 40% of the world's population owes their lives to the Haber–Bosch process (Smil, 2001). In the last few decades, U.S. fertilizer consumption has increased 20-fold, and demands for food, biofuels, and other crops will ensure that N demand will continue to increase (FAO, 2008; Puckett, 1995). This widespread fertilizer application has led to ubiquitous groundwater contamination, as nitrate has been found to be the most common groundwater contaminant in the United States (Nolan and Stoner, 2000).

Because N is a biologically essential element, its transport, transformations, and storage from groundwater to aquatic ecosystems is partially mediated by biological activity. Depending on the biogeochemistry of the watershed, a single N-containing molecule applied to the landscape can interact with many different biological components, sometimes in close proximity or separated by great distances in time and space. This type of cycling has been termed the N cascade, relating the cascade of impacts to the ecosystem that can occur with a single application of reactive N to the landscape (Galloway et al., 2003). One of the most dramatic examples of the impacts of the N cascade is the annual occurrence of hypoxic dead zones in the Gulf of Mexico resulting from contaminated groundwater sources hundreds of miles away (Goolsby and Battaglin, 2000).

Biological denitrification is a process that removes nitrate from aquatic ecosystems, thus terminating the N cascade. The transport of shallow groundwater through riparian soils creates highly suitable conditions for denitrification due to the close interaction between groundwater and saturated, C-rich riparian areas (Cooper, 1990; Hill, 1996). Despite the suitable conditions for denitrification in riparian zones, the effect on overall nitrate concentration is often diluted when the majority of the groundwater does not contact high C riparian zones but is instead transported through sandy vadose zones or less C-rich riparian areas (Cooper, 1990; Hill, 1996; Schipper et al., 1993). Moreover, riparian subsoils may not have enough C and/or may not be waterlogged frequently enough to support continuously active denitrifying microbial communities (Lowrance, 1992). Still others have found that the distribution of C in subsoils is so patchy that denitrification only occurs in sparse hotspots of buried C (Addy et al., 1999; Groffman and Tiedje, 1989; Jacinthe et al., 1998; Parkin, 1987). In areas where the water table is relatively close to the surface, there is an opportunity for an enhancement of denitrification by increasing the contact between shallow groundwater and high-C areas that can support denitrification.

In many studies, denitrification had been enhanced by adding woodchips or sawdust in contact with agricultural effluent, in what are termed bioreactors (Schipper et al., 2010). Many different techniques have been utilized, including containerized treatment systems of woodchips to treat concentrated discharges (denitrification beds) and traditional permeable reactive barriers, (denitrification walls) where sawdust is usually mixed within the soil structure to treat diffuse groundwater flowing perpendicularly through the wall (Schipper et al., 2010). Denitrification walls have been proven to be sustainable, with nitrate reductions from 60 to 90% for at least 15 yr with no maintenance (Long et al., 2011; Moorman et al., 2010; Robertson et al., 2008).

The earliest research on denitrification walls focused on the feasibility of this technology to reduce nitrate in localized groundwater plumes by determining nitrate reductions in contiguous well transects (Robertson and Cherry, 1995; Schipper and Vojvodic-Vukovic, 1998; Schmidt and Clark, 2012). The long-term success of these walls has shifted the focus toward determining the viability of this technology at an appropriate scale and treatment volume to reduce nitrate loads in surface waters draining agricultural properties. The efficacy of this technology to reduce stream N loading will depend on maximizing the volume of water that can be routed through the C substrate with a sufficient detention time to be treated. In this study, to maximize the volume of water treated and measurably impact receiving waters, the wall was located in rapidly flowing groundwater adjacent to significant seepage headwaters that intercepted an area in which a large portion of the groundwater was transported. Similarly to other denitrification walls (Robertson and Cherry, 1995; Schipper and Vojvodic-Vukovic, 2000) the efficacy of this denitrification wall to remove N in localized groundwater was previously determined from well transects (Schmidt and Clark, 2012). The Schmidt and Clark (2012) study determined that all nitrate entering the denitrification wall was removed even though porewater velocities (1.1 m d−1) were one to three orders of magnitude higher and the detention time (1.7–1.9 d) was generally lower than other denitrification walls (0.007–0.47 m d−1) (Robertson and Cherry, 1995; Schipper and Vojvodic-Vukovic, 2000; Schipper et al., 2005; Schmidt and Clark, 2012. These rapid groundwater velocities indicated the potential to markedly reduce N load not only in localized groundwater but also in receiving surface waters. To determine N load reductions at the stream scale, the affected tributary and a control watershed were monitored before and after denitrification wall installation. This tributary monitoring allowed for an assessment of the efficacy of the denitrification wall in the larger watershed context to meet surface water N load regulations.

## Materials and Methods

### Site Location and Construction

The headwaters of the two catchments monitored for this study drain almost solely from a 65-ha container nursery located in Alachua, Florida (82°29′55′′ W, 29°55′7′′ N) within the Santa Fe watershed, which is a part of the greater Suwannee River Basin. Both the Santa Fe River and the Suwannee River have total maximum daily load restrictions for nitrate (Hallas and Magley, 2008).

Soils in the groundwatershed of the denitrification wall generally consist of surface soils containing >93% well-drained sands down to 2 m (USDA, 1985). A clay aquitard that consists of translocated clays (Bt horizon) and Miocene age clay deposits of the Hawthorne formation are present 2 to 2.4 m below the surface where the denitrification wall was installed and at greater depths in the groundwatershed upgradient of the wall. The denitrification wall constructed in one of the two catchments was 55 m long, 1.7 m wide, and 1.8 m deep. The wall was constructed on 30 Sept. 2009 by excavating a trench and backfilling with a 1:1 mixture of pine sawdust and a commercially sold sieved and washed sand (Edgar Minerals Inc.). The wall was installed 14 m from the headwaters of a seepage stream (Fig. 1). The lowest depth of the wall was keyed into a clay-rich confining layer to prevent groundwater bypass and was installed just below the elevation of the main surface water seep. The shallowest depth was 1.8 m above the bottom of the wall. The headwaters of the stream begin as a well-defined seep/spring only approximately 0.3 to 0.9 m wide that occurs due to a break in elevation where the shallow groundwater penetrates the surface. Although two other headwater tributaries contributed to this stream, the targeted seep was visibly the largest single contributor to the discharge of this small tributary, although there are numerous small seepages along the flow path. The long easternmost tributary shown in Fig. 1c, had no visible surface water discharge throughout the duration of the study until the tributary encountered the same break in elevation that caused seepage discharges in the targeted stream located due west. The headwaters of all three tributaries therefore begin at the same break in elevation that occurred at the edge of the field.

Fig. 1.

(A) A side-shot diagram of the denitrification wall indicating the dimensions. (B) An image drawn to scale, delineating the denitrification wall location and the receiving seepage headwaters stream. (C) Large-scale map of the denitrification wall, including the contributing edge of field perimeter, elevation, and sampling stations where discharge, N concentration, and N load were measured within the treatment and control streams. Map created by author, using aerial imagery from Alachua County Property Appraisers (unpublished data, 2006).

To maximize treatment volume, the wall was centered in a groundwater cone of depression where surface water discharges had locally lowered groundwater levels (Fig. 2). The groundwater flow rate of this denitrification wall was previously determined as 1.7 m d−1 although due to curvatures in flow paths the detention time ranged from 1.7 to 1.9 d (Schmidt and Clark, 2012). These curvatures in the flow path reported in Schmidt and Clark (2012) generally converged inward toward the seepage headwaters corresponding to the locally lower water table (Fig. 2). This indicated that groundwater from a wide area was directed through the wall toward the seepage headwaters (Fig. 1). A manual watershed delineation utilizing contour lines from a submeter digital elevation model (DEM) indicated that the denitrification wall only consists of approximately 10 to 11% of the edge of field perimeter contributing to the “treatment” watershed (Fig. 1). Because this small denitrification wall was targeted adjacent to a significant surface water discharge, it was expected to have a disproportionate impact on N loading in the receiving stream.

Fig. 2.

A diagram of soil and water table elevation measured before the wall was installed and the approximate location of the wall.

### Surface Water Monitoring

To determine the influence of the denitrification wall at the watershed scale, surface water discharge, we monitored stream N concentration and N loads before and after wall installation in two catchments, which drain almost entirely from the property. Surface water monitoring was conducted within a “treatment” tributary that was affected by denitrification wall installation and a “control” watershed with similar land-use, climate, hydrology, fertilizer applications, and N concentration that should not be affected by the wall (Fig. 1).

To measure discharge within the treatment and control catchments, we controlled stream flows by installing weirs spanning the entire stream bank width. The weirs were a compound v-notch design, where baseflow discharged through a v-notch and short-duration high-flow events discharged through the v-notch and a larger compound rectangular weir. The water head above the bottom of the stream bank was measured every second and reported as a 5-min average with pressure transducers (Instrumentation Northwest Inc.) and recorded in dataloggers (Campbell Scientific). Discharge was calculated from head measurements with equations programmed into the datalogger. The discharge equations and weir installation design proceeded following standard protocols outlined in USBR (2001).

The relationship between stage and discharge for the v-notch portion of these weirs was determined using the Kindsvater–Shen equation as described in USBR (2001):where Qv is the discharge through the v-notch weir [L3 T−1], θ is the v-notch angle (90°), g is the gravitational acceleration constant [L T2], Ce1 is the effective coefficient of discharge reported in Kulin and Compton (1975), which is a function of v-notch angle only (θ), and he1 is the effective head [L] calculated from the following equation:where h1 is the head above the bottom of the v-notch [L] and Kh is a constant reported in Kulin and Compton (1975), which is a function of v-notch angle only (θ). Flows through the rectangular portion of the weir were calculated using the Kindsvater–Carter equation (Kindsvater and Carter, 1959; USBR, 2001):where Qr is the discharge through the rectangular portion of the weir only excluding the v-notch flows [L3 T−1], Ce2 is the effective coefficient of discharge, which is a function of constant weir geometry and the measured head above the rectangular notch (h2), Le is the effective weir length, and he2 is the effective head. The effective coefficient of discharge was calculated using the following equation from USBR (2001):where h2 is the head value above the bottom of the rectangular notch as measured by the transducer, p is the distance from the stream bottom to the bottom of the rectangular notch, C1 is an equation coefficient, and C2 is an equation constant. The equation coefficient (C1) and equation constant (C2) are based on empirical relationships as a function of weir crest length (L) divided by the stream bank width (USBR, 2001). The effective weir length was calculated with the following equation:where L is the weir crest length [L] and kb is a correction factor reported in USBR (2001). The effective head was quantified with the following equation:where kh2 is a correction factor reported in USBR (2001). When water was discharging through the larger rectangular portion of the weir, flows through the v-notch were calculated as a fully contracted orifice flow using a standard orifice flow equation (USBR, 2001):where Ce3 is the effective coefficient of discharge determined empirically from site calibration as recommended in USBR (2001), A is the surface area of the v-notch orifice, and h3 is the head of water above the midpoint of the v-notch orifice. When water was flowing through the v-notch portion alone, the discharge was calculated as Qv only. When the stream was discharging through the v-notch and rectangular portion of the weir, the discharge was calculated as Qr + Qo.

On the basis of these discharge measurements, discharge-weighted water samples were collected in an autosampler (Teledyne ISCO, Inc.) at programmed stream discharge volumes into pre-acidified bottles. Samples were removed from the autosampler weekly, filtered if applicable, refrigerated, and analyzed within 28 d. For the first 163 d of sampling, approximately 15 to 20 flow-weighted samples per week were collected depending on discharge. These high-frequency flow-weighted samples were composited in to five-sample increments for an N-load reporting frequency of three to four times per week. For the remainder of the monitoring, flow-weighted samples were collected at the same frequency and resolution but composited in to one weekly sample, which gave the exact same weekly average load as the previous method. Stream discharge, N concentration and N load were monitored for approximately 62 d before wall installation and for approximately 448 d after wall installation. In addition to the flow-weighted samples, grab samples were collected (n = 34) beginning in October 2008 for 290 d before monitoring of the discharge sampling station began (August 2009). This extended the N concentration record before the wall was installed to 352 d. Grab samples were also collected 34 d before installation (n = 5) and for 147 d after installation (n = 24) in the stream, immediately at the seepage headwaters which was located 14 m from the denitrification wall. Grab samples were carefully collected from the middle of the water column, minimizing sediment disturbance and were immediately filtered if applicable and refrigerated.

Flow-weighted and grab samples were analyzed for nitrate and total Kjeldahl N (TKN). Unfiltered samples were digested with a block digestion and subsequently analyzed colorimetrically for TKN on an autoanalyzer (Seal Analytical). Nitrate samples were prepared by filtering through a 0.45-μm membrane filter (Pall Corporation) and then analyzed colorimetrically after reduction in a cadmium column on an autoanalyzer (Seal Analytical). Nitrogen load was calculated by multiplying the sum of nitrate and TKN (total N) measured from the water sample [M L−3] by the discharge between samples [L3 T−1]. Statistically significant differences in N concentration and load in the two streams before and after denitrification wall installation were determined with a t test. A change point analysis was performed on N concentration and load to determine statistically significant changes in these values at an α level of 0.05 with the Change-Point Analyzer software (Taylor Enterprises, Inc.).

Dissolved oxygen (DO) and dissolved organic C (DOC) were measured downstream from the seep in response to a bacterial bloom immediately at the stream headwaters. Dissolved oxygen was measured using a multiprobe (556 MPS, YSI Inc.), and DOC was determined after filtering through a 0.45-μm membrane filter (Pall Corporation) as nonpurgable organic C on an infrared gas analyzer (Shimadzu Corp.).

Rainfall was measured with a tipping bucket and potential evapotranspiration was determined using the REF-ET software (Allen, 1999) as a turfgrass reference with the Penman–Monteith equation based on on-site measurements for solar radiation, air temperature and wind combined with a relative humidity probe (Campbell Scientific).

## Results and Discussion

### Stream Oxygen Depletion

Shortly after installation of the denitrification wall, filamentous white bacteria colonized the upper 10 m of the stream located immediately downstream of the seep. The bacteria covered large portions of the stream for 50 d. This was probably in response to excess C export from the wall, which stimulated bacterial colonization and possibly the activities of chemolithotrophic bacteria such as the Beggiatoa genus utilizing reduced H2S to gain energy. Beggiatoa is known to be present in sulfur-rich seeps and springs (Bonny and Jones, 2007), and the odor of H2S was quite strong during this period. One danger with creating sulfur-reducing conditions in bioreactors is the potential to produce toxic methyl mercury in the presence of H2S (Shih et al., 2011). In this study, H2S production occurred in the bioreactor when high temperatures and/or long detention times instigated a near-complete depletion of nitrate and subsequently microorganisms utilized sulfate as an alternate electron acceptor (Shih et al., 2011). Further research is needed to determine if methyl mercury is being produced in the denitrification wall described in the current study.

As a result of this bacterial colonization, DO and DOC were measured in the stream at the immediate seep headwaters. Dissolved oxygen and DOC values were compared with concentrations measured in groundwater from wells within the wall and 3 m downgradient from the denitrification wall (Fig. 1b) as reported in Schmidt and Clark (2012). During this period, DOC in groundwater measured in wells installed 3 m downgradient of the wall regularly exceeded 70 mg L−1 (Schmidt and Clark, 2012). Concentrations of DOC in the stream 14 m from the wall declined over time from a high of 5.32 mg L−1 22 d after wall installation to 2.32 mg L−1 50 d after installation when filamentous bacteria were no longer visually detectable (Table 1). Although no DOC measurements were taken in the surface water seep before wall installation, unimpacted DOC from groundwater wells installed 3 m upgradient from the denitrification wall averaged 1.78 ± 0.29 mg L−1 (Schmidt and Clark, 2012). Similarly to DOC, DO within the stream headwaters rapidly declined 29 d after installation of the denitrification wall and rebounded to DO levels measured from seeps in the watershed (2.3–2.9 mg L−1) (Table 1). Even when DO was below the normal concentrations of seeps in the watershed, spatial sampling indicated that after approximately 20 m downstream from the headwaters, turbulence in the water column had increased the DO concentration to 3.65 mg L−1. Although DO concentrations in the stream headwaters stabilized above background concentrations (2.3–2.9 mg L−1) within 50 d, DO concentrations within groundwater around the denitrification wall still ranged from 0.6 to 0.8 mg L−1 499 d after wall installation. It appeared that as DOC leached from the wall declined or was effectively assimilated by new bacterial growth, biological oxygen demand at the seep subsequently declined and DO levels were easily elevated to background levels due to rapid aeration on atmospheric exposure. These results indicate that a denitrification wall added in close proximity to a stream may negatively affect water quality for a short time and that temporary mitigating practices should be considered in the future.

View Full Table | Close Full ViewTable 1.

Dissolved oxygen concentration (DO) and dissolved organic carbon (DOC) within receiving surface waters. Normal DO of seepage headwaters in the vicinity range from 2.3 to 2.9 mg L−1, while unimpacted groundwater DOC was 1.78 ± 0.29 mg L−1 during this period..

 Receiving stream headwaters Days since installation DOC DO mg L−1 14 4.8 2.4 22 5.3 29 4.9 1.2 36 3.6 1.6 50 2.3 2.6 499 2.9 660 0.94 2.8

In a synthesis paper on denitrification bioreactors, Schipper et al. (2010) observed that excessive DOC concentrations were found initially in many studies and that this could result in depletion of DO in receiving waters. They proposed a variety of preventative measures, including preleaching the media, installing filters downgradient, and maintaining high rates of flow during start-up to ensure nitrate is still present to assist in DOC consumption. The feasibility and cost of these options will need to be weighed against the short-term impacts on surface water quality.

The treatment stream receiving discharges from the denitrification wall and an adjacent control stream (Fig. 1) were monitored before and after wall installation to detect and quantify changes in N concentration and load due solely to the wall installation. Although no two watersheds are exactly the same in hydrology or N concentration, these two watersheds are sufficiently similar to merit comparison. The two streams discharged from immediately adjacent watersheds whose major headwaters are separated by less than 500 m As such, they both share very similar climates. Both watersheds were almost entirely under the same land use (container-plant nursery), and fertilizer was applied at the same time of year to both watersheds. Most significant, before the wall was installed, the relationship in discharge and N concentration between the two streams was strongly correlated, justifying their comparison (Fig. 3).

Fig. 3.

The correlation between discharge and N concentration between the control and treatment stream. The correlation between discharge and nitrogen concentration between the two streams was strongly significant, justifying their comparison.

All results are reported as total N ± 1 SD, which was the sum of measured nitrate and TKN. Total Kjeldahl N only averaged 0.7 ± 0.4 and 0.8 ± 0.4 mg L−1 in the control and treatment streams, respectively, and this concentration did not significantly change after the wall was installed. Before the denitrification wall was installed, total N concentrations were stable in both the treatment and control streams, and no significant change points occurred (Fig. 4a). After the wall was installed, the N concentration in the treatment stream immediately diverged from the control stream, and the first change point occurred in the treatment stream 2 d after the wall was installed. Since the detention time of the denitrification wall in groundwater was reported in Schmidt and Clark (2012) as 1.7 to 1.9 d, this was strong confirmation of the denitrification wall's immediate impact. Subsequent change points occurred when the concentration appeared to partially rebound higher and then stabilize at an intermediate concentration over the duration of the study. This was plausibly due to an initially high concentration of soluble and labile C sources when the wall was first installed, which instigated elevated N removal rates. After these labile C sources were depleted, the N removal rates appeared to have stabilized at a new equilibrium, utilizing consistent C sources. Long-term studies of denitrification walls have indicated that N removal rates stabilized after 1 yr of operation and were predictive of long-term rates (Robertson et al., 2000; Schipper and Vojvodic-Vukovic, 2001; Jaynes et al., 2008; Schipper et al., 2010). Removing this period of temporarily high N reductions, the total N concentration significantly declined from 6.7 ± 1.2 mg L−1 in the 352 d before wall installation to 3.9 ± 0.78 mg L−1 in the period after the last change point only. The concentrations observed in the treatment stream after wall installation had no significant overlap with concentrations measured before wall installation across the range of discharges (Fig. 5b). This indicated that the concentration reduction in the treatment stream was robust and exhibited stationarity across a variety of discharges. Additionally, the relatively even N concentration across a range of stream discharges indicated that the wall was not strongly affected by corresponding increases in groundwater discharges and subsequent decreases in detention time (Fig. 5b). This conclusion is strengthened by the fact that Schmidt and Clark, (2012) found that all nitrate traveling through the denitrification wall was removed long before discharging from the denitrification wall. No change points were detected and no subsequent decline was apparent in the control watershed, which significantly increased from 7.4 ± 0.91 mg L−1 (n = 70) before construction to 7.9 ± 0.78 mg L−1 (n = 109) after construction (Fig. 4a). The concentration measured in the control stream before and after wall installation strongly overlapped across the range of discharges measured (Fig. 5c). Lastly, the N concentration relationship between the control and treatment streams had measurably shifted, thus confirming the response in the treatment stream only (Fig. 5a).

Fig. 4.

An analysis of the impact of the denitrification wall on N concentration and load. Shown in the figures are (A) N concentrations in the control and treatment stream before (Pre) and after (Post) wall installation, with significant change points indicated; (B) N load in the treatment stream pre and post wall installation; and (C) watershed rainfall, evapotranspiration (ET), and treatment stream discharge.

Fig. 5.

A paired watershed analysis of N concentration between the Control and Treatment stream. Shown in the figures are (A) N concentration correlations between the Control and Treatment stream before (Pre) and after (Post) wall installation, and N concentrations across the range of discharges measured for the (B) treatment and (C) control stream.

One method for discerning long-term rate reductions is to compare the same seasons before and after wall installation, when hydrology is comparable. During a subsequent summer–fall period 1 yr after wall installation, when there was no significant difference in rainfall, evapotranspiration, or discharge from the previous year, the total N loading rate in the treatment watershed was 0.52 ± 0.26 kg d−1 (n = 15). Comparing this stabilized N loading rate a year after construction to the same time of year before wall construction (1.46 ± 0.32 kg d−1) indicates a significant cumulative N load reduction of 65% for an average load reduction of approximately 340 ± 130 kg N yr−1 at least during this time of year.

### Nitrogen Reductions in Groundwater vs. Surface Water

The surface water N load reductions measured in the current study can be compared to results measured in groundwater immediately around the footprint of the denitrification wall, reported in Schmidt and Clark (2012). In the latter groundwater study, site hydraulics and N load reductions were quantified from well transects installed upgradient, within, and downgradient from the denitrification wall (Fig. 1b) separated by only 3 m (Schmidt and Clark, 2012), while the receiving stream was monitored 155 m downstream. All results are reported as average ± 1 SD. The average discharge of the stream within the treatment watershed was 17.4 × 104 ± 16.4 × 104 L d−1. This high variability in daily discharges attested to the strong seasonal variability of the stream. On the basis of the findings in Schmidt and Clark, (2012), the denitrification wall treated approximately 10 × 104 ± 2.7 × 104 L of groundwater per day (∼60% of stream discharge). On the basis of rates measured 1 yr after construction, N load in the stream had been reduced by an average of 0.93 ± 0.36 kg d−1, while reductions quantified in groundwater from Schmidt and Clark (2012) were only 0.62 ± 0.42 kg d−1. These values indicate the reduction in groundwater N load resulting from water passing through the denitrification wall was on average lower than the N load reductions in the stream, although there was overlap in the variability measured. Two possible explanations for this discrepancy are measurement uncertainties or that the denitrification wall had increased in situ N reductions either in groundwater outside the footprint of the well transects or within the stream itself. The former explanation is an inevitable limitation of assessing complicated groundwater and surface water discharges, whereas the latter hypothesis has intriguing implications.

There was some evidence that at least initially, excess C spiraling within the watershed was stimulating N loss and transformations within the stream. For 34 d before wall installation (n = 5) and a duration of 147 d after wall installation (n = 24), grab samples were collected manually in the stream, right at the seepage headwaters present 14 m from the denitrification wall, and analyzed for nitrate concentration. On average, the nitrate concentration within the seepage headwaters declined by 33 ± 13% (n = 25) from concentrations measured before (7.8 ± 0.67 mg L−1) and after (5.2 ± 1.0 mg L−1) wall installation. The nitrate concentration reductions measured 155 m downstream of the seep at the surface water sampling station were greater with before wall installation concentrations of (5.4 ± 0.47 mg L−1) reduced by 57 ± 12% (n = 92) to (2.3 ± 0.64 mg L−1). Although it was reasonable to assume that some of the effluent from the denitrification wall bypassed the main seep and discharged at seepages downstream, it was plausible that the increased DOC loads resulting from the wall had stimulated further nitrate reductions. Before wall installation, significant in situ nitrate reductions as a result of denitrification had been observed in locations where stream morphology facilitated high organic C and hypoxia (Frisbee, 2007). This indicates that under normal circumstances with no wall installed, denitrification occurred within stream sediments (Frisbee, 2007).

## Conclusions

Installation of denitrification walls adjacent to streams enables reductions in groundwater N before reaching sensitive surface water bodies. Although the wall only comprised 10 to 11% of the edge of field perimeter contributing to the treatment stream, the total N load declined by 65% for a load reduction of 345 kg yr−1. Such significant surface water reductions as a result of wall installations have not been reported previously, and this work verifies that targeting walls can have a disproportionate impact on downstream N loading. The disadvantage to installing these walls near streams is the potential to detrimentally impact stream water quality from initially high C leaching.

The cost of materials and construction were approximately $20,000. Assuming a conservative 15-yr life span and stable nitrate removal rates measured 1 yr after installation, the N removal cost over the 15-yr period is$0.79 kg−1 N. Estimates of N removal costs are higher for other treatment systems, including municipal wastewater treatment ($40 kg−1 N), wetlands ($3.26–8.90 kg N−1), riparian buffers ($26 kg N−1), and even other denitrification bioreactors ($2.39–15.17 kg N−1) (CENR, 2000; Hyberg, 2007; Schipper et al., 2010). The success and cost-effectiveness of this study indicates the feasibility of utilizing denitrification walls to reduce N loading from agricultural properties.

## Acknowledgments

We would like to thank the Florida Department of Environmental Protection (FDEP) and the Florida Department of Agriculture and Consumer Services for funding and Todd Stephens for enthusiastically allowing this research on his property. We would like to thank Dr. Jim Jawitz for assistance and review with this esearch. Additionally, we would like to thank Patrick Moran for field assistance.