Tidal controls on riverbed denitrification along a tidal freshwater zone

Abstract

In coastal rivers, tidal pumping enhances the exchange of oxygen-rich river water across the sediment-water interface, controlling nitrogen cycling in riverbed sediment. We developed a one-dimensional, fluid flow and solute transport model that quantifies the influence of tidal pumping on nitrate removal and applied it to the tidal freshwater zone (TFZ) of White Clay Creek (Delaware, USA). In field observations and models, both oxygenated river water and anoxic groundwater deliver nitrate to carbon-rich riverbed sediment. A zone of nitrate removal forms beneath the aerobic interval, which expands and contracts over daily timescales due to tidal pumping. At high tide when oxygen-rich river water infiltrates into the bed, denitrification rates decrease by 25% relative to low tide. In the absence of tidal pumping, our model predicts that the aerobic zone would be thinner, and denitrification rates would increase by 10%. As tidal amplitude increases toward the coast, nitrate removal rates should decrease due to enhanced oxygen exchange across the sediment-water interface, based on sensitivity analysis. Denitrification hot spots in TFZs are more likely to occur in less permeable sediment under lower tidal ranges and higher rates of ambient groundwater discharge. Our models suggest that tidal pumping is not efficient at removing surface water nitrate but can remove up to 81% of nitrate from discharging groundwater in the TFZ of White Clay Creek. Given the high population densities of coastal watersheds, the reactive riverbeds of TFZs play a critical role in mitigating new nitrogen loads to coasts.

Key Points

• Denitrification rates decrease along tidal rivers as oxygen exchange increases
• Denitrification rates fluctuate with tides in coastal riverbeds
• Coastal riverbeds can remove a large portion of groundwater-borne nitrate

1 Introduction

Aquatic ecosystem productivity requires biologically available forms of nitrogen (N), such as nitrate ( ) and ammonium ( ). Historically, reactive N limits primary productivity within estuarine and marine ecosystems [Elser et al., 1990]. In recent decades, increases in the supply of to the coast, due to anthropogenic activities such as agriculture and the burning of fossil fuels [Howarth, 1998; Howarth and Marino, 2006; Elser et al., 2007], have led to declines in coastal water quality. Estuaries are especially sensitive to loading, with excess tied to reductions in biodiversity [Nixon, 1995; Howarth, 2007], hypoxia [Smith, 2003] and changes to local vegetation that increase vulnerability to storm erosion [Turner and Rabalais, 1994; Edmonds and Slingerland, 2010].

Denitrification in river networks mitigates loads to coasts [Seitzinger et al., 2002; Mulholland et al., 2008]. Shallow riverbed sediments are often high in organic carbon and low in dissolved oxygen (DO) and, therefore, act as preferential zones for denitrification [Duff and Triska, 1990]. The removed can be sourced from infiltrating surface water, discharging groundwater, or both [Hester et al., 2014]. In many lowland rivers, both surface water and groundwater are elevated in . Near the coast, tides drive alternating infiltration of river water and exfiltration of groundwater (tidal pumping), which may influence removal from either source. During rising tide, river water infiltrates and delivers to microbially active sediments, where it may be removed before subsequent discharge during falling tide (Figure 1). Flow reversals may also provide extended opportunities to remove from upwelling groundwater [Rocha et al., 2009]. Alternatively, tidal pumping, may limit denitrification in upwelling groundwater by enhancing the supply of dissolved oxygen from surface water [Rocha et al., 2009]. If surface water is particularly rich in DOC and sediments are particularly depleted in organic matter, tidal pumping could enhance aerobic respiration and denitrification simultaneously. These complex interactions highlight uncertainties in nitrogen removal within tidally influenced riverbeds.

Figure 1

Conceptual model for aerobic respiration and denitrification in the riverbed mixing zone under (1) rising tide, when river water infiltrates, and (2) falling tide, when groundwater discharges. Arrows indicate flow direction. Groundwater is a source of . River water is a source of DO, , and DOC. Organic matter (OM) in streambed is an additional source of DOC.

Tidal influence often extends inland for tens to hundreds of kilometers in unregulated rivers. TFZs are fresh river reaches influenced by a substantial tidal regime [Odum, 1988; Muylaert et al., 1997]. Found worldwide, TFZs are usually associated with larger rivers such as the Delaware, Hudson, and Thames [Odum, 1988]. They are often unmonitored for water and solute fluxes because of their complex hydrodynamics [Odum, 1988; Muylaert et al., 1997; Destouni et al., 2008]. Few studies have examined surface water-groundwater interaction and contaminant transport within TFZs. Bianchin et al. [2010] showed that tidal pumping in the TFZ of the Fraser River (British Columbia, Canada) increased the depth of hyporheic exchange up to 1 m. Schuchardt et al. [1993] found that hyporheic zone residence times were prolonged in the TFZ of the Weser Estuary (Germany). They showed that longer residence times facilitated greater nitrification over shorter reaches compared to non-tidal reaches. Hensley et al. [2015] observed N removal dynamics within a clear, spring-fed TFZ in Florida and invoked changes in residence times to explain the dynamics. Ensign et al. [2008] observed oxidation-reduction (redox) potential and developed an empirical model of denitrification in the floodplain of the Newport River, North Carolina. They suggested that denitrification in TFZ floodplains promotes attenuation and reduces delivery to the coast. Past work at our site in White Clay Creek (Delaware) showed that at least 11% of river water exchanged through riverbed and bank storage zones [Musial et al., 2016]. These studies illustrate that TFZs are regions of enhanced surface water-groundwater interaction and N processing. Although TFZs may play a critical role in modulating nutrient loads to the coast, N transport dynamics and removal efficiencies remain poorly known.

In this study, we quantified the effects of tidal pumping on removal in the hyporheic zone of a TFZ. We hypothesized that tidal pumping creates a deep zone of aerobic respiration that shifts over tidal time-scales and influences removal rates. We tested our hypothesis using a one-dimensional transient groundwater flow and reactive transport model developed from observations of fluid fluxes and pore water chemistry in the bed of White Clay Creek Delaware, USA (Figure 2). The site is representative of other populated lowland settings where is elevated in both river water and groundwater. We first present observed and modeled fluid flow and dynamics at the field site. We then use sensitivity studies to explore spatial and temporal variations in removal along a generalized TFZ. We show that removal is highly dynamic, and significantly more is removed from discharging groundwater than infiltrating river water. Tidal pumping increases the supply of oxygen to riverbed sediments, which may inhibit removal of groundwater-borne .

Figure 2

(a) Field site (green dot) lies 17 km from the Delaware River on White Clay Creek. Stage at the field site was monitored during geochemical sampling in 2014. Additional monitoring sites (orange and grey dots) were added in 2015. Stars show USGS stations 1480065 (blue) and 01480120 (black). (b) Example of variations in stage along the TFZ. Colors correspond to locations in (a). Stage is stable during low tide near the upstream limit of the TFZ (green line). (c) At the upper limit of the TFZ of White Clay Creek, tidal range increases rapidly toward the coast and then plateaus.

2 Materials and Methods

2.1 Study Area

The study site (39.701172°, −75.649987°) is located on White Clay Creek, a fifth-order river within the Christina River Basin Critical Zone Observatory (CRB-CZO) Delaware (Figure 2). Land use in the watershed is predominantly agricultural, forested, and suburban [Newbold et al., 1997]. The study site is located within the TFZ, 17 km upstream of the Delaware Bay (Figure 2). The semidiurnal tidal range is approximately 0.9 m. Like many coastal rivers, White Clay Creek is deeply incised [Dellapenna et al., 1997; Woolfe et al., 1998] and, therefore, the floodplain is not inundated at high tide (Figure 3). Sandy point bars are common along the channel.

Figure 3

Stratigraphic cross section of study site on White Clay Creek (green dot in Figure 2). River flow direction is out from page. Vertical Exaggeration = 1.67. Modified from Musial et al. [2016]. Location B, the focus of this study, is relatively representative of riverbed sediment in the area.

Riverbed deposits at the study site generally consist of a lower sequence of alternating silty sand and sandy silt overlain by a sand unit approximately 25 cm thick (Figure 3) [Musial et al., 2016]. Macroscopic organic matter occurs sporadically throughout the lower silty sand and sandy silt unit. Porosity is approximately 0.45, and the vertical effective hydraulic conductivity is 4.0 × 10⁻⁴ m s⁻¹ [Musial et al., 2016]. As part of a broader study to characterize surface water-groundwater exchange, a series of wells and piezometer nests were installed in June 2014 along a channel-perpendicular transect on the eastern bank of the river and within the riverbed [Musial et al., 2016]. For the current study, we focus on location B (Figure 3), which was inundated throughout most of the tidal cycle. Location B includes five sample ports at depths of 12, 25, 50, and 100 cm below the riverbed (Figure 3). Sample ports were constructed out of 0.635 cm outer diameter polypropylene tubes perforated over a 2-cm interval and attached to a PVC riser. The screen length for the sampling ports was 2 cm.

2.2 Field Measurements

Surface water and pore water chemistry were monitored over two consecutive days. On 16 June 2014, specific conductivity, temperature, pH, oxidation reduction potential, and dissolved oxygen (DO) were measured in surface water and all ports over low and high tides using a Thermo Orion Star A3229 multiparameter probe. On 17 June 2014, pore water samples were collected for anions, nutrients, and dissolved organic carbon (DOC) at approximately 2 h intervals over one tidal cycle using a peristaltic pump at a rate of ∼50 mL min⁻¹. We did not collect nutrient samples at the shallowest port (12 cm) to minimize disturbance of steep vertical gradients in pore water chemistry. One tubing volume (∼60 m L⁻¹) was discarded before sample collection. Samples were filtered (0.45 µm), immediately placed on ice, and frozen until analysis. Samples were analyzed for DOC using high temperature combustion with a Shimadzu Total Organic Carbon analyzer. Duplicate samples, instrument replication, reference standards and blanks were used to ensure analysis quality. and were measured using a Skalar flow-injection nutrient analyzer. All and values are reported as -N and -N.

To measure surface water-groundwater interaction, vertical head gradients (VHG) between the deepest port (1 m below ground) and the river were measured every 15 min for the duration of one tidal period on 18 June 2014 using a manometer board. VHG cannot be measured while pore water samples are collected because pore water extraction would disturb VHG. However, river stage data collected from a local stilling well between 16 June 2014 and 19 June 2014 verify that tidal dynamics were consistent over the 3 days of pore water sampling and VHG measurements (see Figure 3 in Musial et al. [2016]). VHG data were used to calculate the exchange rate across the riverbed using Darcy's law, assuming a vertical hydraulic conductivity of 1.09 × 10⁻³ m s⁻¹ [Musial et al., 2016]. Manometer board measurements were accurate to ±2 mm, resulting in a minimum seepage rate error of ±0.16 m d⁻¹. In practice, the seepage rate error is probably greater due to additional uncertainties associated with hydraulic conductivity.

We measured river discharge rates and collected surface water samples in order to evaluate downstream fluxes in the channel for comparison with fluxes across the riverbed. River discharge was measured every 60 min with an acoustic Doppler current profiler (Teledyne RDI StreamPro) on 18 June 2014, and river water samples were collected simultaneously. Instantaneous riverine loads were calculated as the product of river discharge and river concentration.

To understand longitudinal variations in tidal amplitude along the TFZ, pressure sensors were installed in a stilling well at the study site and at two locations downstream the following field season in 2015. Tidal fluctuations at all stations were recorded every 10 min between 18 August 2015 and 21 August 2015. Note that river stage was also continuously monitored at field site during VHG measurements and water sample collection in 2014.

2.3 Fluid Flow and Reactive Transport Modeling

Unsteady fluid flow and reactive transport were simulated in one dimension within a vertical sediment column (Figure 1). The modeled reactions include aerobic respiration, nitrification, denitrification, and NH₄⁺ and DOC production from sediment. Subsurface flow was first solved using the transient groundwater flow equation with a source term to represent the effects of an oscillating tidal load, or change in applied stress, on the compressible sediment column [van der Kamp and Gale, 1983; Wang and Davis, 1996; Wang, 2000; Wilson and Gardner, 2006]:

(1)

where S_s is specific storage, h is hydraulic head, t is time, K is hydraulic conductivity, x is depth below the sediment-water interface, and γ is the tidal loading efficiency in the source term (Table 1). The storage term allows for vertical variations in Darcy velocity along the sediment column, and S_s was chosen to represent a typical value for unconsolidated, fully saturated sediments of 2.45 × 10⁻⁵ m⁻¹ [Freeze and Cherry, 1979]. A constant upward flux was designated at the lower boundary of the flow domain on the basis of observed VHG and net groundwater discharge rates from Musial et al. [2016], representing net groundwater discharge. A fluctuating hydraulic head signal, h(0,t) was specified at the upper boundary to represent semidiurnal tides. The lower boundary condition sets the time-average discharge rate, while the upper boundary condition influences the range of discharges rates across the bed from rising tide to falling tide. For non-tidal conditions, the fluctuating hydraulic head condition at the sediment-water interface was replaced with a constant hydraulic head condition to yield steady, uniform groundwater discharge along the sediment column. We also simulated a 1-cm thick zone of surface water with high K (10⁻³ m s⁻¹) above the sediment-water interface so that we would not need to constrain the solute concentration or dispersive solute flux immediately at the sediment-water interface, where the flow direction reverses.

Table 1. White Cay Creek Model Parameters Where a, b, and c Represent the C:N:P Ratio

Parameter	Value	Reference
Physical Parameters
n (porosity)	0.45	Musial et al. [ 2016]
qgw (ambient groundwater flux at lower boundary)	4.27 (cm d^−1)	Observed
A (tidal amplitude)	0.45 (m)	Observed
K (hydraulic conductivity)	4.0 × 10^−4 (m s^−1)b	Observed
Ss (specific storage)	2.45 × 10^−5 (m^−1)	Freeze and Cherry [ 1979]
γ (loading efficiency source term)	0.90	Wang and Davis [ 1996]
λ (tidal period)	12 (h)	Observed
Reaction Parameters
C:N:P (redfield ratio)	112:20:1	Canavan et al. [ 2006]
αL (dispersivity)	0.01 (m)	Freeze and Cherry [ 1979]; Hester [ 2013]; Hester et al. [ 2014]
OM0 (sediment organic matter)	0.2	Marmonier et al. [1995]
Dm molecular diffusion	5 × 10^−11 (m^2 s^−1)	Ingebritsen and Sanford [ 1999]
X (microbial biomass)	0.000142 (kg m^−3)	Gu et al. [ 2007]
Var (maximum specific O2 reaction rate)	1.97 (h^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]
Vnit(maximum specific reaction rate)	1.08 (h^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]
Vden(maximum specific reaction rate)	3.98 (h^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]
βar,O2 (stoichiometric coefficient for aerobic respiration)	2.67
βnit,O2 (stoichiometric coefficient: nitrification)	4.57
βden,NO3 (stoichiometric coefficient: denitrification)	0.933
βar,NH4 (stoichiometric coefficient: aerobic ammonification)	0.21
βden,NH4 (stoichiometric coefficient: anaerobic ammonification)	0.21
KDOC (half saturation constant for DOC)	8.68 (mg L^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]; Sawyer [ 2015]
KO2 (half saturation constant for DO)	5.28 (mg L^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]; Sawyer [ 2015]
(half saturation constant for )	1.65 (mg L^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]; Sawyer [ 2015]
KNH4 (half saturation constant for )	0.43 (mg L^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]; Sawyer [ 2015]
KI (oxygen inhibition)	0.24 (mg L^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Hester et al. [ 2014]; Sawyer [ 2015]
α (mass transfer coefficient for OM to DOC)	5 × 10^−5 (h^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Sawyer [ 2015]
Kd (distribution coefficient)	50 (m^3 kg^−1)	Gu et al. [ 2007]; Zarnetske et al. [ 2012]; Sawyer [ 2015]
Groundwater Solute Concentration
DO	0.1 (mg L^−1)	Observed
	4 (mg L^−1)	Observed
	0.1 (mg L^−1)	Observed
DOC	0 (mg L^−1)	Observed
Surface Water Solute Concentration
DO	9 (mg L^−1)	Observed
	3.5 (mg L^−1)	Observed
	0.007 (mg L^−1)	Observed
DOC	4 (mg L^−1)	Observed

• ^a Parameters were guided by literature or chosen by fits with data. Stoichiometric coefficients were constrained by reaction equations and stoichiometry.
• ^b Musial et al. [2016] reports a K of 1.09 × 10⁻³ m s⁻¹ for the riverbed sequence while we use a slightly lower value 4.0 × 10⁻⁴ m s⁻¹ as it is more representative of conditions right at sampling point B (Figure 2).

Resulting pore water velocities were used as inputs to the advection-dispersion-reaction equations for dissolved oxygen (DO), nitrate ( ), ammonium ( ), and dissolved organic carbon (DOC):

(2)

(3)

(4)

(5)

where v is the pore water velocity, V_k is the maximum microbial process reaction rate for the kth reaction and X is the biomass of the functional microbial group. β_ar,O2, β_nit,O2, β_den,NO3, β_ar,NH4 and β_den,NH4 are stoichiometric coefficients [Doussan et al., 1997; Canavan et al., 2006; Gu et al., 2007]. K_j is the half saturation constant for reactant j. OM is organic matter content by weight percent. DOC derived from OM (third term on the right side of equation 5) is represented as a first-order reversible kinetic reaction with mass transfer coefficient α that is intended to represent the combined effects of dissolution and desorption [Jardine et al., 1992; Gu et al., 2007; Zarnetske et al., 2012]. Additional reactions include aerobic respiration, nitrification, and denitrification. Denitrification (last term in equations 3 and 4) occurs under anoxic conditions. We, therefore, include I in equations 3 and 4 to represent non-competitive inhibition of denitrification by DO [Widdowson et al., 1988]:

(6)

where K_I is the inhibition constant.

The hydrodynamic dispersion coefficient (D) is:

(7)

where D_m is the molecular diffusion coefficient in porous media, and α_l is dispersivity, which represents pore scale mechanical dispersion (Table 1).

Concentrations of , , DO, and DOC in surface water and groundwater end-members were assigned to the upper and lower boundaries based on field observations (Table 1). OM content was estimated by fitting the modeled and measured DOC profiles. Reaction kinetics were constrained from the literature [Molz et al., 1986; Gu et al., 2007; Zarnetske et al., 2011a, 2011b; Marzadri et al., 2012; Sawyer, 2015] and adjusted by fitting measured concentration data (Table 1). Models were solved in COMSOL, a generic finite-element solver, and run until fluid fluxes and solute concentrations approached quasi-steady state (∼130 days). The maximum element size was 1.5 cm. The groundwater flow equation was solved for a larger domain (100 m), while the advection dispersion reaction equation was solved for the upper 2 m only, similar to the approach of Gu et al. [2008]. Reported modeled reaction rates were integrated over the upper 1 m of riverbed pertaining to the interval where we have geochemical observations.

2.4 Sensitivity Study

To explore potential downstream trends in N transformation along a generalized TFZ, we varied tidal amplitude from 0.1 to 1 m (Table 2), well within the range of tidal fluctuations seen in White Clay Creek. We also varied net groundwater discharge rates from 0.43 to 8.6 cm d⁻¹ in order to explore the sensitivity of the model to changes in base flow. Net groundwater discharge at the study site was measured as 4.27 cm d⁻¹. Groundwater discharge rates for the sensitivity analysis were chosen to range over an order of magnitude and encompass the measured value. We tested four water quality end-members by varying nitrate ( -N): (1) high in groundwater (10 mg L⁻¹, denoted HG), (2) high in surface water (8 mg L⁻¹, denoted HS), (3) low in groundwater (2 mg L⁻¹, denoted LG) and (4) low in surface water (1 mg L⁻¹, denoted LS) (Table 3). We also examined the influence of sediment properties by varying K from 4.0 × 10⁻⁶ (silty sand) to 9.0 × 10⁻⁴ m s⁻¹ (clean sand) [Freeze and Cherry, 1979]. Although K can vary between 1.0 × 10⁻⁷ and 1.0 × 10⁻³ m s⁻¹ in natural riverbed sediment [Calver, 2001], we constrained our range of K to approximately two orders of magnitude encompassing our measured K of 4 ×1 0⁻⁴ m s⁻¹. For this sensitivity study, we assumed an exponential OM profile:

(8)

where, OM₀ is 0.2 and b is 9 (Marmonier et al., 1995). Modeled organic matter concentrations thus range from 20% at the sediment water interface to 0.0025% at 1 m depth with an average concentration of 4.1%.

Table 2. Parameters Varied in Sensitivity Analysis

Parameter	Range
OM (%)	4.1–5.1
SW (mg L^−1)	1–10
GW (mg L^−1)	1–10
Tidal amplitude (m)	0–1
Hydraulic conductivity (m s^−1)	4 × 10^−6−9.0 × 10^−4
Ambient GW discharge (cm d^−1)	0.43–8.6
Ki (mg/L)	0.05–2

Table 3. Concentrations for Surface Water and Groundwater End Members in Sensitivity Analysis

Parameter	Value
HG (High groundwater)	10 mg L^−1
LG (Low groundwater)	2 mg L^−1
HS (High surface water)	8 mg L^−1
LS (Low surface water)	1 mg L^−1

2.5 Long-Term Simulations

To capture long-term dynamics in nitrogen cycling, tidal pumping was simulated using river stage data from USGS monitoring station 1480065 at Newport, DE between 14 July 2015 and 17 August 2015. The station lies 8 km downstream from the study site. The time range was chosen to include both spring and neap tides and encompasses several storm events. Stage data were used as the boundary condition for hydraulic head at the sediment-water interface, while all other parameters remained unchanged from the model scenario for White Clay Creek at location B (Table 1).

3 Results

3.1 Fluid Flow and Nitrogen Dynamics at White Clay Creek

Tides drove reversals in fluid flow across the sediment-water interface. River water infiltrated into the bed during rising tide and discharged during falling tide (Figure 4). Measured vertical seepage rates fluctuated between −0.66 ± 0.16 m d⁻¹ (infiltration) and 0.38 ± 0.16 m d⁻¹ (discharge) over the tidal cycle [Musial et al., 2016]. Net flux was calculated as −0.09 ± 0.16 m d⁻¹ (near neutral given the uncertainty) at location B. Overall net flux was 0.87 m d⁻¹ along the entire transect (Figure 3), indicating net discharge of groundwater to the channel [Musial et al., 2016]. Simulated vertical seepage rates fluctuated between −0.15 m d⁻¹ (infiltration) and 0.19 m d⁻¹ (discharge) over the tidal cycle (Figure 4a). The modeled VHG between the sediment-water interface and a depth of 1 m generally agrees with measured VHG but has a slightly more positive range that reflects our choice of net groundwater discharge at the model base (Figure 4b).

Figure 4

(a) River stage (black) and modeled Darcy flux across the bed (grey). (b) Measured (black) and modeled (grey) vertical head gradient. Black and grey squares in (a) correspond to sampling times for profiles in Figure 5. Both VHG and stage were recorded simultaneously on 18 June 2014. At low tide the river stage sensor was still submerged by approximately 20 cm.

DOC concentrations in surface water and pore water were relatively stable over the tidal cycle and ranged from 2.38 to 4.42 mg L⁻¹ (Figure 5a). concentrations were also stable (Figure 5b). Pore water concentration was highest (4.22 mg L⁻¹) at 1 m depth but negligible between 0.75 and 0.25 m depths (Figure 5b). Surface water concentrations were consistent (∼4 mg L⁻¹) over the tidal cycle (0 m depth, Figure 5b). Groundwater concentrations increased steadily from 0 to ∼1 mg L⁻¹ between depths of 1 and 0.25 m (Figure 5c). in surface water was negligible. DO concentrations were low and mostly negligible at all sampling depths in the subsurface (Figure 5d). The only exception was at 0.5 m depth, where DO concentrations were ∼1.5 and ∼0.5 mg L⁻¹ at high and low tide, respectively. Surface water DO concentrations were consistently high (∼8 mg L⁻¹) over the tidal cycle.

Figure 5

Concentration profiles of (a) DOC, (b) , (c) , and (d) DO from measurement and models at location B. Black and grey dots on Figure 4a correspond with sample times at high and low tide. (e) Vertical variation in Darcy flux associated with tidal pumping at high (black) and low (grey) tide. Concentration profiles of (f) DOC, (g) , (h) and (i) DO for non-tidal scenario. (j) Darcy flux is uniform and steady for non-tidal scenario.

The modeled concentration profiles for DOC, , , and DO generally agreed with the measured data (Figure 5). The model accurately predicted a zone of negligible between 0.25 and 0.75 m and higher at 1 m (Figure 5b). The model predicted elevated concentrations in the zone of depleted (0.25 to 0.75 m) but overestimated concentrations at ∼0.75 m (Figure 5c). These overestimations may have been due to additional processes occurring at depth that were not included in the model, including heterotrophic microbial and autotrophic uptake and perhaps anaerobic ammonium oxidation (anammox) [Burgin and Hamilton, 2007; Lansdown et al., 2012]. The simulated DO profile indicated an aerobic zone that extends from the sediment-water interface to 0.08 m at low tide and 0.12 m at high tide (Figure 5d).

Modeled solute concentration profiles were remarkably similar in the absence of tidal pumping (compare Figures 5f–5j with 5a–5e). Under steady groundwater discharge, was still depleted over much of the profile. A subtle but important difference is that the simulated aerobic zone only extended 0.03 m below the sediment-water interface (Figure 5i) instead of fluctuating between 0.08 and 0.12 m (Figure 5d). Modeled removal rate (aggregate effect of denitrification and nitrification) was also 10% less under tidal than non-tidal conditions. In the tidal scenario, the time-average rate was 122 mg m⁻² d⁻¹.

Removal rates fluctuated over tidal timescales (Figure 6a). During low tide, rates of denitrification and removal steadily increased while the nitrification rate (difference between removal and denitrification) decreased. Just prior to rising tide, denitrification and removal rates reached a maximum of 160 mg m⁻² d⁻¹ and 150 mg m⁻² d⁻¹, respectively, and the nitrification rate reached a minimum of 6.82 mg m⁻² d⁻¹ (Figure 6a). During rising tide, denitrification and removal rates then decreased by ∼25% and 40% to 120 mg m⁻² d⁻¹ and 90 mg m⁻² d⁻¹, respectively (Figure 6a). Concurrently, nitrification increased to a maximum of 30.04 mg m⁻² d⁻¹ (difference between black and grey lines in Figure 6a). The removal rate was consistently lower than the denitrification rate, indicating coupled nitrification-denitrification throughout the tidal cycle. For simplicity, the model assumed removal is due to denitrification exclusively; however, there are several other processes that may have contributed to removal. Autotrophic uptake, dissimilatory nitrate reduction to ammonium (DNRA), and coupled reduction and iron oxidation could have all played a role in removing from the system [Tiedje, 1988; Giblin et al., 2013].

Figure 6

(a) Modeled removal rate (black) and denitrification rate (grey) are at their lowest when stage (shown for reference as a dotted line) is high. (b) Modeled (black) and DO flux (grey) are downward (negative) when flow is into the bed and upward when flow is out of the bed. (c) River discharge (black) and riverine load (grey) are low when stage is high and increase with falling tide.

Simulated DO and fluxes across the bed were positive at low tide, indicating transport from sediments to surface water (Figure 6b). As tide rose, fluxes became negative, indicating transport from the surface water to sediments. flux across the sediment-water interface ranged from −1300 to 750 mg m⁻² d⁻¹, and DO flux ranged from −520 to 690 mg m⁻² d⁻¹, respectively (Figure 6b). The timing of downstream transport was similar to flux across the bed. The measured riverine load was lowest (523 kg d⁻¹) approximately 2 h before high tide, at a period when river water infiltrated the bed. The measured riverine load was highest (2200 kg d⁻¹) 1 h after high tide (Figure 6c). As tide fell, hyporheic water exited storage, and the downstream transport rate increased. The riverine load was proportional to river discharge (Figure 6c).

3.2 Sensitivity Study

As simulated tidal amplitude increases from 0 to 1 m along an idealized TFZ, tidal pumping exchanges more DO across the sediment-water interface, causing a steady increase in the aerobic respiration rate (Figure 7a). Given that oxygen increases nitrification and inhibits denitrification, the removal rate decreases (Figure 7b). For the HG/HS scenario (representative of low water quality in surface and groundwater), removal is 165 mg m d⁻¹ under low tidal amplitude (amplitude of 0.1 m). However, removal decreases to 46 mg m d⁻¹ at a tidal amplitude of 1 m (Figure 7b). This shift is primarily due to a decrease in denitrification (Figure 7c) and, to a lesser extent, an increase in nitrification (Figure 7d). The nitrification rate increases steadily from 3.1 mg m⁻² d⁻¹ at 0.1 m tidal amplitude to 36 mg m⁻² d⁻¹ at 1 m.

Figure 7

(a) Rates of aerobic respiration, (b) removal, (c) denitrification, and (d) nitrification along a continuum from low to high tidal influence (inland to coast) under conditions of high and low nitrate in groundwater (HG and LG, respectively) and high and low nitrate in surface water (HS and LS, respectively). Rates are less sensitive to surface water than groundwater concentration, which indicates that denitrification mostly removes groundwater-sourced

The other three water quality conditions (HG/LS, LG/HS, and LG/LS) show similar trends in aerobic respiration and N transformations (Figure 7). Denitrification rates are nearly identical for both high groundwater conditions (HG/HS and HG/LS) but differ from low groundwater conditions (LG/HS and LG/LS). In other words, denitrification rates are much more sensitive to concentrations in groundwater than surface water (Figure 7c). removal rates scale nearly linearly with both groundwater and surface water concentration (Figure 8a). We also tested the influence of surface water DOC concentration by varying it from 6 to 10 mg/L. The influence on denitrification rates was negligible (not shown).

Figure 8

(a) removal increases with groundwater concentration and is less sensitive to surface water . (b) removal rate decreases with tidal amplitude and hydraulic conductivity. (c) removal rate increases with groundwater discharge, or base flow to the stream but decreases with tidal amplitude. The star indicates White Clay Creek concentrations. Negative values indicate net production.

removal rate also varies with sediment properties. The removal rate decreases as the hydraulic conductivity of sediment increases from values typical of silt to sand (Figure 8b). removal is lowest (in fact, net production occurs) when hydraulic conductivity and tidal pumping are high, both of which promote greater oxygen exchange across the sediment-water interface. The highest modeled removal rate (110 mg m⁻² d⁻¹) occurs at a tidal amplitude of 0.1 m and hydraulic conductivity of 4.0 × 10⁻⁶ m s⁻¹, typical of silty sand [Freeze and Cherry, 1979]. Sediment organic matter content also strongly controls denitrification rates. We increased the average OM content from 4.1 to 5.3% to test the effect on DOC supply and denitrification rates (Figure 9). Under these conditions, aerobic respiration and nitrification still increase with tidal amplitude or proximity to the coast (not shown), while removal rate decreases (Figure 9b). However, the removal rate is between 1.7 and 3.5 times greater for the sediment with higher OM content across all tidal amplitudes (Figure 9). A deeper sediment organic matter profile supplies more DOC (Figure 9a) to facilitate faster respiration. The tight coupling in our model between OM in sediments and DOC availability also explains why the DOC concentration in surface water does not significantly influence N cycling. In TFZs with clean sand beds, the supply of DOC from surface water would likely exert more control on N cycling.

Figure 9

Streambed sediments are an important source of DOC in models. (a) Simulated DOC concentrations using a profile containing greater (grey line) or lesser (black line) amounts of OM. (b) removal rates is greater under a deep organic matter profile (grey) than shallow organic matter profile).

Net groundwater discharge, or base flow influence, is yet another control on the removal rate (Figure 8c). Greater net discharge supplies more from groundwater to the reactive riverbed and increases the removal rate. Greater net discharge rates are likely to occur in wetter climates and better drained soils.

3.3 Long-Term Tidal and Storm Dynamics

Four storm events increased river discharge in White Clay Creek between 14 July 2015 and 13 August 2015. During each storm event, stage rose ∼0.5 m upstream of the TFZ (Figure 10a). Within the TFZ, the effect of storms on river stage was less apparent (Figure 10b). Average stage and tidal amplitude fluctuated over the 1-month interval, but fluctuations may have been more associated with spring-neap cycles than storm dynamics. These fluctuations drove changes in modeled daily average denitrification rates and the daily range of denitrification rates (Figures 10c–10d). For example, on 24 July during base flow and neap tide, the average denitrification rate over a single tidal period was 160 mg m⁻² d⁻¹. About a week later during spring tide (greater tidal amplitude), the daily average denitrification rate was 12.5% lower (140 mg m⁻² d⁻¹). Around 8 August (a neap phase), denitrification rates fell again even though the tidal amplitude decreased. The decline in the modeled denitrification rate is likely associated with an increase in the average stage, which would tend to temporarily increase infiltration of oxic river water into the bed and reduce the rate of groundwater discharge. Both effects would tend to decrease denitrification rates.

Figure 10

Stage data for (a) USGS station 1479000 along a non-tidal reach of the Christina River at Newark (b) USGS station 1480065 (blue star in Figure 2) at Newport within the TFZ. (c) Simulated denitrification rate at tidal USGS station 1480065. (d) Daily range in denitrification rate varies with spring and neap tides.

4 Discussion

4.1 Role of Tidal Pumping in Removal

In our initial conceptual model, we suggested that tidal pumping may increase or decrease removal, depending on reactant supply from multiple sources and reaction rates along flow paths (Figure 1). Our models indicate that tidal pumping reduces removal by enhancing the supply of oxygen from surface water. As tidal amplitude increases, the additional oxygen stimulates aerobic respiration and nitrification but reduces denitrification (Figure 7). At our specific site in White Clay Creek, tidal pumping increases the aerobic zone depth from approximately 3 cm without tides to 10 cm with tides (Figure 5), which reduces net removal in the upper meter of riverbed sediment by 10%.

Most of the removed originates from groundwater instead of surface water, as evidenced by the high sensitivity of removal rates to groundwater concentrations (Figures 7 and 8a). Surface water is not removed as efficiently as groundwater because most of the oxygen from surface water is not consumed prior to discharge. and DO from surface water travel ∼0.03 m into the bed by advection over one tidal cycle before the flow reverses again towards the channel. Penetration depths of DO and from surface water are deeper than this advective exchange length (0.12 m and 0.13 m, respectively) due to enhanced dispersion over many cycles of oscillatory flow [Ataie-Ashtiani et al., 1999; Boutt and Fleming, 2009]. The net effect is that only a portion of the hyporheic zone is suitable for denitrification [Harvey et al., 2013]. Most in surface water that enters the sediment during a given tidal cycle will return to the stream without exposure to anaerobic conditions, limiting the potential for removal prior to discharge during falling tide.

Other studies have demonstrated the critical role of oxygen inhibition on coupled nitrification and denitrification under relatively simple flow regimes [Widdowson et al., 1988; Sheibley et al., 2003a, 2003b]. The propensity of a hyporheic zone to act as a sink for nitrate depends strongly on the consumption of oxygen along hyporheic flow paths [Zarnetske et al., 2012]. Under the transient flow conditions found in the riverbeds of TFZs, oxygen consumption continues to play a key role in the fate of N.

4.2 Applicability to N Transformation in Other Dynamic Flow Settings

Other studies have examined the influence of transient flows on N transformation in hyporheic zones, but these studies focused on longer term signals associated with storms or seasons. Triska et al. [1990] showed that storms can enhance nutrient uptake while post-storm discharge may increase loading to streams. Gu et al. [2012] found that storms temporarily increase denitrification within microbially active storage zones by increasing groundwater storage and prolonging residence times. In our study, tidal storage moments are shorter and more frequent. The associated flow oscillations expand the aerobic zone but do not significantly alter the amount of time that discharging groundwater spends in carbon-rich sediments. As a result, the flow oscillations in tidal riverbeds appear to inhibit denitrification. Similar behavior may occur in riverbeds downstream from dams due to hydropeaking [Boutt and Fleming, 2009; Sawyer et al., 2009].

4.3 Implications for Hot Moments in TFZs

Over a single tidal cycle at our study site, the denitrification rate decreases up to 20% as oxygen infiltrates and then increases again with exfiltration of groundwater (Figure 6). Denitrification rates also fluctuate over weekly timescales in response to storm and spring-neap events. Between 14 July and 13 August 2015 modeled denitrification rates at White Clay Creek varied by 60% (Figure 10c). Increases in average stage tend to coincide with decreases in average denitrification rates. These increases in average stage could be associated with storms, onshore winds, or even reservoir management. Note that we did not consider large changes in stage that might be associated with overbank flooding. Floods can have significant impacts on denitrification [Ensign et al., 2008] due in part to floodplain carbon inputs [Junk et al., 1989]. Increases in tidal amplitude, which can occur during spring tides, also decrease average denitrification rates. These lower frequency fluctuations in denitrification rates are analogous to observed variations in carbon cycling over spring-neap cycles in mangrove systems [Call et al., 2015].

4.4 Longitudinal Trends in TFZs

Along the continuum from rivers to estuaries, increasing tidal amplitude favors aerobic respiration and limits denitrification rates (Figure 7). Previous studies have suggested that riverine denitrification rates are generally higher than estuarine [Seitzinger, 1988; Mulholland et al., 2008], and our models provide new insights into denitrification rates along the transition zone between these ecosystems. At our study site near the upstream limit of the TFZ in White Clay Creek, the modeled average denitrification rate over a tidal cycle is 140 mg m⁻² d⁻¹. This rate is slightly greater than measurements from five locations within the Delaware River's TFZ (55–115 mg m⁻² d⁻¹) and significantly greater than denitrification rates (0–90 mg m⁻² d⁻¹) measured from N₂ production of intact sediment cores in the fully tidal Delaware Bay [Seitzinger, 1988]. If interpreted along a gradient of tidal influence, these estimates support our finding that denitrification rates decrease along TFZs. However, it is important to recognize that additional factors besides tidal pumping may contribute to longitudinal trends (for example, carbon quantity and quality in riverbed sediments).

4.5 Model Considerations

In this model, we assumed that dissolved oxygen inhibits denitrification and chose an inhibition constant similar to previous studies: denitrification is limited for DO concentrations greater than 0.24 mg L⁻¹ [Sheibley et al., 2003a, 2003b; Zarnetske et al., 2012]. However, field measurements have shown strong correlations between denitrification and high oxygen demand and the potential for high denitrification rates in shallow, aerobic riverbed sediments [Arango et al., 2007; O'Connor and Hondzo, 2008; Mulholland et al., 2008]. Therefore, an increase in tidal pumping and oxygen supply to shallow sediments may not limit denitrification as much as our models suggest. The persistence of anoxic sites in heterogeneous sediment can facilitate denitrification in bulk aerobic zones [Harvey et al., 2013; Sawyer, 2015; Briggs et al., 2015]. Heterogeneity in sediment type (both permeability and organic matter content) and microbial activity affect bulk rates of redox transformation in the hyporheic zone [Harvey et al., 2013; Sawyer, 2015].

In order to measure the sensitivity of our results to oxygen inhibition, we ran several models where the inhibition constant ranges from 0.05 to 2 mg L⁻¹. Over this range, the denitrification rate only changes by 7.8%. Changes in aerobic respiration rate and nitrification rates are negligible. For small inhibition constants (less than 0.25 mg L⁻¹), there is a nearly linear increase in denitrification rate of about 0.17 mg m⁻² d⁻¹ for every mg L⁻¹ increase in K_I. For greater inhibition constants, there is a much smaller increase in the denitrification rate. This suggests that the presence of DO may not be the only factor limiting denitrification rates in our models. DOC availability may also strongly limit removal [Heffernan and Cohen, 2010; Zarnetske et al., 2011a, 2011b], especially given the sensitivity of denitrification rates to OM content (Figure 9). Field-scale observations are needed to confirm whether aerobic sediments within TFZs are better zones of denitrification than our models suggest.

In addition to denitrification, there are several additional microbially mediated processes that could contribute to the removal of within the hyporheic zone. Examination of the carbon to nitrogen molar ratio (data not shown) illustrates significantly higher C:N ratios in pore water at 75 cm depth than at 25 or 100 cm. In this zone of relatively low inorganic nitrogen and high organic carbon, denitrification is likely to dominate [Seitzinger et al., 2006; Burgin and Hamilton, 2007]. If reduced iron (Fe) is also plentiful, then the reduction could be coupled with Fe oxidation [Burgin and Hamilton, 2007]. However, we do not have Fe data at our site. In zones with low organic carbon to inorganic nitrogen ratios and low Fe, denitrification could be coupled with anammox [Burgin and Hamilton, 2007]. Contributions of anammox to removal could help explain the overestimation of concentrations at depth (Figure 5).

Our one-dimensional models do not fully capture many hyporheic flow mechanisms that interact in TFZs. In addition to tidal pumping, mechanisms such as bedform-current interactions likely contribute to removal. Bedform-current interactions in TFZs should be highly dynamic because they are modulated by changes in currents and water depths [Hester and Doyle, 2008; Bradley et al., 2013; Boano et al., 2013]. Bianchin et al. [2010] presented a conceptual model for the influence of tides on bedform-current interactions. At high tide when river flow is low, hyporheic exchange through bedforms tends to decrease. However, tidal pumping may counteract this effect by enhancing downward flow into the bed. As tide falls and river flow increases, hyporheic exchange through bedforms tends to increase, but upward flow associated with tidal pumping again may counter this effect. Future field observations and models that include hyporheic exchange due to current-bedform interactions would more accurately constrain N removal in TFZs.

Our one-dimensional model also cannot represent horizontal flow that may occur within layers of sediment [Conant et al., 2004; Bianchin et al., 2010]. Nevertheless, model results fit VHG and geochemical data (Figures 4 and 5). Also, two-dimensional models of storm driven surface water-groundwater interaction show a predominance of vertical flow in the streambed (i.e., Figure 6 in Gu et al. [2008]). Two-dimensional models are beyond the scope of this study but would be useful for understanding the behavior of flow and N transformation in the riverbanks, which likely differs from that of the riverbed even in non-tidal rivers [Gomez-Velez et al., 2015; Harvey et al., 2013]. In TFZs, denitrification has been examined in tidally inundated floodplain soils [Ensign et al., 2008] but not bank storage zones within supratidal floodplains. Tidal water table fluctuations above bank storage zones may aerate shallow groundwater and favor net production in some settings. Future studies should incorporate higher dimensional modeling that includes reactive transport in the variably saturated floodplain aquifer. Additional field and modeling studies are still needed to refine estimates of nutrient transformation and dynamics in TFZs.

4.6 Implications for N Export to Coasts and TFZ Management

Rivers and streams are substantial sinks and act as natural buffers that limit N transport to sensitive coastal environments [Wollheim et al., 2006; Ensign and Doyle, 2006; Mulholland et al., 2008; Kiel and Cardenas, 2014]. Several field and modeling studies have quantified denitrification rates within various river networks [Seitzinger et al., 2002; Wollheim et al., 2006; Alexander et al., 2009; Helton et al., 2011], but none have considered the tidally influenced portion of the network. For example, Wollheim et al. [2008] used a modeling approach to show that ∼13–16% of N can be removed by rivers and streams on a network scale, thus mitigating N delivery to the coast. Our study illustrates that at the downstream ends of river networks, TFZs may have a reduced capacity to remove river-borne N even though tidal pumping may enhance hyporheic exchange (Figure 7b).

Based on our model results for White Clay Creek, we can evaluate the importance of riverbed denitrification on N export to the Delaware Bay through simple extrapolation. The cross-sectional width of White Clay Creek at the study site is 26 m and the length of White Clay Creek's TFZ is approximately 17 km. For simplicity, we assume a constant channel width and denitrification rate along the TFZ, though the channel width increases and the denitrification rate likely decreases in the downstream direction. The resulting removal rate for the entire riverbed area is 61 kg/d. This is only 4% of the measured riverine load near the upstream limit of the TFZ (1500 kg/d, Figure 6c). In other words, riverbed sediments within the TFZ of White Clay Creek are only able to remove 4% of the in river water that discharges to the coast. However, this removal rate is equivalent to 81% of the potential groundwater-borne load to the TFZ (∼75 kg d⁻¹ if net groundwater discharge rates and concentrations are roughly constant along the TFZ). Thus, the hyporheic zone of the TFZ plays an important role in mitigating new sources from groundwater that would otherwise discharge near the coast. This ecological service is particularly crucial, given the prevalence of sources to coastal groundwater from septic tanks and fertilizers. Groundwater-borne loads to coastal rivers can constitute a sizeable fraction of the total loads to estuaries. For example, in the Chesapeake Bay, base flow contributions of to coastal rivers represent more than half (58–73%) the total estimated load [Miller et al., 2015]. Groundwater-borne loads are only expected to increase into the future because a large component of discharging groundwater was recharged prior to changes in land use and intensive fertilizer application [Bohlke and Denver, 1995].

TFZs are rarely monitored for either river discharge or N loads due to their complex hydrodynamics [Destouni et al., 2008]. Most observations and models of N loads to coasts neglect TFZs, which may lead to underestimation of total N export. Relatively recent improvements in sensor technology have made it possible to monitor both river velocity and concentrations at sub-hourly timescales. We advocate for increased monitoring of water and solute fluxes in tidally influenced rivers across the globe in order to improve our understanding of ultimate chemical loads to coastal waters.

N fate in TFZs likely varies regionally due to both geologic and hydroclimatic controls. Denitrification rates in TFZs are influenced by hydraulic conductivity (Figure 8b), organic matter (Figure 9), and net groundwater discharge (Figure 8c). Sandier sediments typically have high hydraulic conductivity and low organic matter content, both of which correspond with lower denitrification rates, all other factors held constant. Sand-rich TFZs likely remove less groundwater-borne than silt-rich TFZs, similar to findings in beach and estuary settings [Weinstein et al., 2011; Sawyer et al., 2014]. The net rate of groundwater discharge also influences supply and thus the denitrification rate (Figure 8c). For a given lithology or hydraulic conductivity, the rate of groundwater discharge to a river increases with the recharge rate. TFZs in wetter regions may remove more groundwater-borne than TFZs in arid regions (Figure 8c). Field observations from TFZs in other geologic settings and climates are essential for verifying these predictions.

5 Conclusions

TFZs are dynamic regions of coastal rivers where N cycling in riverbed sediments varies both temporally and longitudinally. As tide rises, oxygenated river water infiltrates sediments and temporarily decreases denitrification rates. Along TFZs, removal rates decrease with increasing tidal influence due to greater oxygen exchange across the sediment-water interface. At our study site near the upstream limit of a TFZ, tides increase the depth of aerobic sediments by a factor of 6, resulting in a 10% reduction in removal rates. Nevertheless, riverbed sediments within TFZs can still be tremendously effective at mitigating loads from discharging groundwater. Riverbed denitrification removes more than 80% of the groundwater-borne load to the TFZ of White Clay Creek.

Most efforts to monitor or model N loads from rivers to coasts are concentrated upstream of TFZs. As a result, N loads from some of our most densely populated coastal catchments go unmonitored. This research provides an important step in understanding N fate in TFZs. Future research should focus on reach scale sampling and modeling to accurately predict the N transformation potential of TFZs across a range of geologic, land use, and climate settings. These observations are essential to improving predictions of N export to coasts and informing best management practices for coastal rivers. It is imperative that we understand nutrient transport in TFZs in light of their potential vulnerability to anthropogenic inputs, sea level rise, and increasing storm severity.

Acknowledgments

This research was supported by a McNair Scholar research fellowship, the National Science Foundation (Award #EAR-1446724), and the Grant Lyddon Foundation. We thank Corey Wallace, Kelsey Danner, and Megan Mave for assistance in the field. We also thank Kim and Ralph Burdick with the Hale Byrnes House for access to the field site. Anthony Aufdenkampe, Louis Kaplan, and Holly Michael provided equipment support and advice throughout the field season. We thank Sue Welch and Kathy Welch for assistance with water chemistry analyses. The manuscript benefitted from the suggestions of Associate Editor Erich Hester, Adam Ward, Robert Hensley, and two anonymous reviewers. Data used in this study would be made available upon request from the corresponding author.