SEARCH

SEARCH BY CITATION

Keywords:

  • conservative transport;
  • constructed wetlands;
  • multitracer;
  • nonconservative transport;
  • solute transport;
  • transient storage model

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

[1] Because of emerging vegetation and sedimentation processes, the succession of wetlands is a dynamic process. Hence, a noticeable impact on the functioning and the efficiency of constructed treatment wetlands regarding solute retention can be expected. Within 5 months a reduction of active wetland volume, a decrease of light decay, and an increase of sorption capacity were observed using four multitracer experiments in a newly established constructed wetland. Tracer breakthrough curves of conservative and nonconservative tracers were analyzed with the help of a transient storage model. The model characterized the impact of vegetation development and sediment accumulation on solute transport properties. Three different tracers allowed an assessment of wetland hydraulics, sorption processes, and light impact on photodegradable solutes. Finally, the exemplary transport prediction of a fourth, independent tracer that was both photodegradable and sorptive demonstrated a cost-efficient technique to determine the influence of succession processes on treatment efficiency.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

[2] Recent publications show that there is an increasing interest in understanding transport processes within aquatic ecosystems [e.g., Cheng et al., 2011; Deng et al., 2010; Herrman et al., 2010; Hu et al., 2010; Kadlec, 2009; Schmid et al., 2010; Su et al., 2009; Wahl et al., 2010]. This interest is partly caused by the increasing recognition of diffuse micropollutants as a threat to ecosystems [Brown et al., 2010]. Also, there is a need for a better understanding of processes in nonpoint source pollution mitigation systems such as constructed wetlands, which are already widely in use [Rygaard et al., 2011]. Two types of constructed wetlands are commonly used to treat wastewaters: wetlands with free water surfaces and horizontal subsurface flow wetlands [Kadlec, 2009]. This study focuses on the first type.

[3] After surface flow wetlands have been designed and constructed, they alter because of the establishment of vegetation or sediment accumulation [Keefe et al., 2010]. Numerous studies have demonstrated that in addition to geomorphological aspects like shape [Cardenas, 2009; Persson, 2000; Stern et al., 2001; Thackston et al., 1987] and depth distribution [Carleton and Montas, 2007; Holland et al., 2004; Lightbody et al., 2007; Somes et al., 1999], also vegetation influences flow and solute transport in wetlands [e.g., Kjellin et al., 2007; Lightbody et al., 2008; Nepf et al., 1997, 2007; Wörman and Kronnäs, 2005] and streams [e.g., Salehin et al., 2003; Vionnet et al., 2004]. Wetland vegetation affects hydraulic roughness and eddy viscosity [Somes et al., 1999; Vionnet et al., 2004] and may generate areas of preferential flow [Carleton and Montas, 2007; Kjellin et al., 2007; Larsen and Harvey, 2010; Somes et al., 1999; Stern et al., 2001; Vionnet et al., 2004; Wilson, 2007]. Seasonality of vegetation communities induces additional variations in wetland hydraulics [Keefe et al., 2010]. Vegetation development also affects chemical processes. Uptake and sorption on plants as well as degradation by microbial biofilms were reported for solutes such as nutrients, heavy metals or pesticides [Alvord and Kadlec, 1996; Dobberteen and Nickerson, 1991; Huang et al., 2008; Moore et al., 2000, 2001, 2002]. Photolysis of organic compounds can be reduced by shading [Dierberg and DeBusk, 2005; Keefe et al., 2004].

[4] Considering the dual impact of emerging vegetation on wetland hydraulics and on solute retention, a strong interaction between these two aspects can be expected, which has not been studied yet. When constructed wetlands are used as treatment systems, simple methods to assess the interaction between hydraulic performance and retention processes could also improve management practices and structural design.

[5] Hydrological tracers are common tools to study hydraulic behavior and solute transport in wetlands and streams [e.g., Gooseff et al., 2008; Harvey et al., 1996; Kadlec, 1990, 1994; Wagner and Bencala, 1996; Werner and Kadlec, 2000]. Applying the residence time distribution (RTD) approach [Levenspiel, 1972], the temporal moments of the RTD are commonly used to describe wetland hydraulics by tracer mass recovery, mean residence time and by the variance of the residence time, which expresses the spread of the tracer response [e.g., Kadlec, 1994; Thackston et al., 1987]. Physically based solute transport modeling, like the application of two-region transport models, can increase the understanding of dynamic processes inside wetland flow systems, such as the exchange between flow dominated and stagnant areas [Kadlec, 1994]. The physical interpretation of model parameters is commonly used to characterize streams and wetlands [e.g., D'Angelo et al., 1993; Keefe et al., 2004; Schmidt et al., 2010]. Thus, a physical interpretation of changing model parameters could help to infer successional changes of wetland transport characteristics without repeated bathymetric surveys.

[6] By combining conservative with nonconservative or reactive tracers, the overall impact of chemical processes on solute transport can be estimated [Kadlec, 1994; Keefe et al., 2004; Lange et al., 2011]. Wagner and Harvey [1999] showed that initial calibration of transport models to conservative transport could improve parameter estimation for chemical processes. First-order degradation can easily be implemented into solute transport models, since only one parameter is needed to estimate time-dependent degradation rates. Modeling was sufficient for several compounds and processes, e.g., for nutrient uptake [McKnight et al., 2004] or pesticide degradation by microorganisms [Dungan and Yates, 2003]. Sorption by first-order kinetic mass transfer is a second chemical process that was incorporated into solute transport models. A linear isotherm described reversible sorption on streambed sediments for Sr [Bencala, 1983; Gooseff et al., 2005], 51Cr (III) [Jonsson et al., 2003] and Rhodamine WT [Keefe et al., 2004]. To the best of our knowledge, transport predictions for compounds which underlie multiple retention processes were not undertaken so far. Those would result in large parameter uncertainties resulting from parallel process calibrations. The present study uses a different approach. It applies specific tracers to separately mimic each retention process. Individually estimated process parameters are then combined in a single model run to describe transport and retention processes within a wetland characterized by emerging vegetation (Phragmitis australis, Thypha latifolia, Juncus conglomeratus) and sediment. We use multiple tracers with different physicochemical properties and the OTIS model [Runkel, 1998] to study wetland succession during 5 months: We choose one conservative tracer to characterize hydraulic changes of the system, induced by altered discharge, sedimentation and vegetation patterns and two nonconservative tracers to study photolysis and sorption inside the wetland. The calibrated transport model is finally used to predict a fourth sorptive and photosensitive tracer.

2. Study Site

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

[7] The study was performed in a 258 m2 constructed free water surface wetland located in the wine-growing area of the Kaiserstuhl, southwest Germany (Figure 1). Designed as a bypass to the mainstream, it is located within a 2.2 ha flood detention pond at the outlet of the 1.8 km2 Löchernbach catchment. The high-flow bypass is constructed to decrease sediment accumulation inside the wetland. Discharges exceeding 15 L s−1 coincide with water levels in the inlet ditch of the wetland which cause an overflow of the bypass weir. Thus, increasing discharges lead to an increasing fraction of bypassing water fluxes. No additional measures prevented sediment deposition inside the wetland. The wetland is 30.5 m long, has a mean width of 8.3 m and a mean water depth of approximately 0.4 m. At a water level corresponding to a discharge of 10 L s−1 (time of the first experiment), the volume of the wetland is 100.1 m3 (±3.4 m3). A V notch weir was installed at the wetland outlet to control and measure water levels and to take water samples. The wetland is sealed by an impermeable clay layer. Prior to the first experiment a sediment layer of 0.05 to 0.15 m depth consisting of eroded material of the typical calcaric regosol of the Löchernbach catchment with 10% sand, 80% silt and 10% clay had accumulated. This layer contained 5.3% organic carbon, as determined by calcination. The survey of the sediment layer was used to estimate the initial potential volume of the hyporheic zone (mean depth of the sediment layer times wetland area). We defined the combination of observable water volume (VWATER), including initial vegetation mass and potential volume of the hyporheic zone (VHYP) as the total observable wetland volume VOBS (Figure 2). For a densely vegetated wetland the volume of biomass VBIO (L3) must be considered as well. During construction in the preceding year, no lateral inflow had been observed in an empty pond for two weeks. The tracer experiments were performed during constant base flow ranging from a maximum of 10 L s−1 in spring to minimum of 3.2 L s−1 in summer.

image

Figure 1. Schematic map of the study site.

Download figure to PowerPoint

image

Figure 2. (top) Scheme of conceptual model volume fractions and observable volume fractions in an initial and a densely vegetated wetland in spring and summer, respectively. (bottom) Schematic relation of observable and computed wetland volumes.

Download figure to PowerPoint

[8] Initial vegetation was planted with a density of 1 plant per m2 in spring (Figure 3a). Planting consisted of approximately 60% Phragmitis australis, 20% Typha latifolia and 20% Juncus conglomeratus. In addition to planted species (Typha latifolia was dominating after 5 months), a wide variety of other aquatic plants (e.g., Cardamine amara) colonized the wetland. To prevent disturbances of vegetation coverage, detailed surveys inside the wetland were avoided in summer. Nevertheless, a distinct sediment accumulation was observed (but not quantified) and a vegetation coverage of 70% to 80% of the wetland area was estimated (Figure 3b). Distinct water pathways evolved through vegetation and sediment. Considering these succession processes, significant changes of the total active wetland volume, of the distribution of flow paths and of the volume of the hyporheic zone were expected.

image

Figure 3. (a) Constructed wetland with initial vegetation (≈60% Phragmitis australis, 20% Typha latifolia, and 20% Juncus conglomerates) in spring; the photograph has been taken in the flow direction. (b) Constructed wetland with dense vegetation cover in summer; the photograph has been taken against the flow direction.

Download figure to PowerPoint

3. Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

3.1. Tracer Experiments

[9] Tracer experiments were conducted during spring (2 and 3 March) and summer (9 and 19 August) 2010. Slug injection (SI) and constant rate injection (CRI) were applied to account for differences in solute exchange with stagnant areas or the hyporheic zone [Gooseff et al., 2008]. For each experiment four different tracers with specific physicochemical properties were used: sodium bromide (BR) and the three fluorescent dyes sulforhodamine B (SRB), uranine (UR) and eosin (EOS).

[10] BR is an often used tracer for solute transport studies in wetlands [e.g., Kadlec, 1994; Keefe et al., 2004]. Sorption or degradation of bromide is reported to be negligible. Because of its low background concentration (<0.1 mg L−1 in natural waters), smaller amounts, compared to chloride, can be applied. Consequently, density effects of traced water are reduced. SRB (Acid Red 52, C27H29N2NaO7S2) has successfully been applied in wetlands [Lange et al., 2011; Passeport et al., 2010; Torres et al., 1997], although it is reported to be sorptive with different materials [Käss and Behrens, 1998; Leibundgut et al., 2009; Sabatini, 2000; Smart and Laidlaw, 1977]. Leibundgut et al. [2009] reported half-life times under sunlight of up to 850 h. UR, also known as fluorescein (Acid Yellow 73, C20H10O5Na2), is almost nonsorptive and considered as a conservative tracer for groundwater studies [Leibundgut et al., 2009]. Yet it is highly photosensitive with a photolytic half-life time of 11 h under sunlight [Leibundgut et al., 2009]. EOS (Acid Red 87, C20H6Br4Na2O5) is characterized as moderately sorptive and highly photosensitive (half-life time of 6.5 h under sunlight [Leibundgut et al., 2009]). In our study, BR was analyzed with ion chromatography (Dionex-DX 500) and the fluorescent dyes with a spectral fluorometer (LS-50B, Perkin-Elmer). The separation of UR and EOS was achieved by alkalizing (pH 9) and acidifying (pH of 4). Baseline samples were taken prior to the experiments to account for background fluorescence. Detection limits were 0.1 mg L−1 for BR, 0.1 μg L−1 for SRB, 0.05 μg L−1 for UR and 0.1 μg L−1 for EOS.

[11] For the SI experiments, all tracers were mixed in a 8 L solution and simultaneously injected into the inlet ditch. The shallow depth of the ditch (0.15 m) guaranteed complete mixing. For the CRI experiments, tracer solutions were constantly pumped from a 25 L container into the inlet ditch. Flow rates in the tube were measured and logged to keep a constant pumping rate. Pumping lasted for 0.5 and 2.78 h at rates of 2.3 and 1.6 mL s−1 in spring and summer, respectively. Inlet samples were taken at a frequency of 5 min in spring and 30 min in summer. Total durations of the spring experiments were 3.5 h (SI) and 6 h (CRI) and for the summer experiments 20 h (SI) and 25 h (CRI).

[12] Water samples were taken at the outlet using 100 mL brown glass bottles. Sampling started at the time of injection with a frequency of 5–10 min. After 3–5.5 h, autosamplers were initiated with a frequency of 0.5 h. Unfortunately, no autosamples were taken during the spring experiments. Autosamplers were immediately emptied and samples were stored in a refrigerator. Analyses of the fluorescent dyes were completed within 2 weeks following the experiments. Discharge was measured by direct volumetric measurements at the outlet weir with 9–10 repetitions for each experiment (standard deviation 0.3–0.7 L s−1). Absolute water table elevations were observed during each discharge measurement. Wavelets and discharge fluctuations caused water table oscillations of 1.3–2.7 cm during the experiments. The vertical banks of the wetland allowed estimates of wetland volumes (VOBS) for each experiment (initial VOBS minus (absolute change of water level times wetland area)). Comparable concentrations of chloride, nitrate and sulfate were observed in the stream water during all experiments. Water temperatures at the wetland outlet ranged from 5°C to 8°C in spring and from 15°C to 20°C in summer. An overview of the different tracer experiments is given in Table 1.

Table 1. Summary of the Tracer Experimentsa
Season of ExperimentInjection MethodDischarge (L s−1)Injection Time (h)BR Injection Mass (g)SRB Injection Mass (g)UR Injection Mass (g)EOS Injection Mass (g)
  • a

    BR, sodium bromide; SRB, sulphorhodamine B; UR, uranine; EOS, eosin; SI, slug injection; CRI, constant rate injection.

  • b

    Tracers were injected instantaneously. Endurances are the integration time steps used for modeling.

SpringSI100.01b38810.20.8
 CRI8.40.53602.20.431.75
SummerSI3.20.01b3881.30.20.8
 CRI6.92.787301.61.11.1

[13] During all experiments global radiation was recorded at 10 min resolution using a THIES CM-3 pyranometer. During the summer experiments, light decay of the fluorescent tracers was studied by hourly samples from three buckets with different tracer concentrations (factors between starting concentrations 1:2:5). To account for diurnal variation of global radiation, light decay was evaluated in relation to cumulated global radiation. Exponential decay was described by:

  • display math

where Ci is the concentration of tracer i as a function of the cumulated global radiation δ (Wh m−2) and the rate constant k. For the adjustment of UR modeling parameters (light decay) to EOS characteristics, the factor

  • display math

was determined.

[14] Distribution coefficients (Kd, mL g−1) for the three dyes were determined in batch tests with composite samples of the wetland sediments. Batch tests were conducted following recommendations of the laboratory standards of the U.S. Environmental Protection Agency [1999], taking duplicate samples with a 1:5 ratio for sediment and water and varying tracer concentrations. Tracer concentrations were chosen according to the relative fluorescence intensity of each tracer [Leibundgut et al., 2009] and with reference to expected tracer concentrations in the wetland (SRB, 1/10 μg L−1; UR, 0.2/2 μg L−1; EOS, 0.4/4 μg L−1).

3.2. Transport Modeling and Hydraulic Characterization

[15] The one-dimensional numerical model OTIS [Runkel, 1998] was chosen to model flow and transport processes. The OTIS modeling framework has successfully been applied for conservative and nonconservative solute transport in wetlands [Harvey et al., 2005; Keefe et al., 2004, 2010; Martinez and Wise, 2003; Wang and Jawitz, 2006] and streams [e.g., Bencala, 1983; Bencala and Walters, 1983; Gooseff et al., 2003; Gooseff et al., 2005; Wagner and Harvey, 1997]. The numerical finite differences approach describes flow systems using two conceptional zones. In a main channel (MC) the convection-dispersion equation describes longitudinal transport, while a continuous stirred tank reactor acts as a lateral storage zone (SZ). Exchange between the two zones is controlled by a first-order exchange term. In our study we assumed that the MC cross-sectional area summarizes all areas where advection is the dominating flow process. The SZ was assumed to represent all retaining areas, i.e., stagnant zones with low flow velocities, areas with diverging flow and the hyporheic zone in the sediment layer.

[16] The main equation of OTIS [Runkel, 1998] for conservative and nonconservative transport processes in the MC is

  • display math

and for the SZ

  • display math

where C is the solute concentration (M L3), inline image the sorbate concentration on the streambed sediment (M M−1), Q the discharge (L3 T−1), t the time step (T), x the system length (L) and A are the cross-sectional areas (L2). System characteristics and process rates are the longitudinal dispersion DL (L2 T−1), the exchange rate α (T−1) between MC and SZ, the available sediment concentration inline image (M L3), the distribution coefficient for the stream sediments Kd (L3 M−1), the first-order degradation rate λ (T−1) and the sorption rate λsorp (T−1). The subscripts MC and SZ indicate the two model areas.

[17] Parameters were estimated using an IDL routine for Monte Carlo simulations with 30,000 model runs for each data set. A multicriteria optimization (absolute error, RMSE, correlation coefficient, cumulated efficiency, Nash-Sutcliffe efficiency (NSE), log NSE, and mean NSE) was applied. Best model fits were chosen using NSE as objective function. A positive NSE indicates that the explanatory power of the model is larger than the mean of the observed data. For parameter interpretation the best 0.2% (n = 60) parameter sets of the Monte Carlo simulations were selected. To allow a direct comparison between the derived parameters of the different experiments, we maintained identical calibration ranges for all parameters and experiments. Therefore, initial parameter ranges were too large to apply iterative modeling routines, such as the commonly used OTIS-P approach [Runkel, 1998]. Wagener et al. [2002] showed that OTIS transport parameters can hardly be identified for wide posterior parameter ranges. Therefore, we compared the resulting parameter ranges by the uncertainties of their best model calibrations. To show whether resulting parameter ranges differed significantly between our tracer experiments, we applied multihypothesis testing using the nonparametric Mann-Whitney U rank sum test statistic [Mann and Whitney, 1947] with a Bonferroni corrected level of significance (α) of 0.05.

3.2.1. Conservative Transport

[18] Bromide tracer breakthrough curves (TBC) were used for hydraulic characterization. Parameters describing conservative transport were the longitudinal dispersion DL in the MC, cross-sectional areas in the MC (AMC), and SZ (ASZ) and the first–order exchange rate α. AMC and ASZ are defined by

  • display math
  • display math

where AACT (L2) is the total active wetland cross section and SZfrac (dimensionless) is the relative fraction of storage compared to the total active wetland cross section. During calibration AACT and SZfrac were varied instead of the original model parameters AMC and ASZ. This change allowed us to link the observable wetland volume VOBS (L3) to AACT by

  • display math

where VACT (L3) is the total simulated wetland volume and VDEAD (L3) is the volume of all areas without any contact to the tracer [Martinez and Wise, 2003]. Thus, the change of active wetland volume through developing vegetation and sediment accumulation could be analyzed by comparing VACT (L3) estimated by OTIS with VOBS (Figure 2).

[19] Inside a wetland, the RTD of a conservative tracer describes the probability density function of the residence time of all tracer molecules under the boundary conditions of the experiment [Kadlec, 1994]. In order to compare RTDs of experiments with different discharges and water volumes, Werner and Kadlec [1996] proposed the dimensionless RTD, which is calculated using the dimensionless flow- and volume-weighted time inline image as follows:

  • display math
  • display math

where VWETLAND (L3) is the observed or estimated wetland volume, Qconst the constant discharge (L3 T−1) during the experiments and Tn the nominal residence time of each experiment. The dimensionless RTD C′( inline image) is calculated as

  • display math

where MOut is the recovered tracer mass. The resulting distributions characterize the dominating flow and transport processes inside the wetland, including preferential flow and the degree of mixing [Holland et al., 2004; Werner and Kadlec, 1996], which are a measure of retention efficiency. The first moment of the dimensionless RTD C′( inline image) is equivalent to the effective volume ratio e [Thackston et al., 1987], which is defined as the ratio of mean tracer retention time to Tn.

3.2.2. Nonconservative Transport

[20] Model parameter sets for conservative transport obtained for BR were the basis to analyze the chemical processes which reduced the tracer recovery of SRB, UR and EOS. Sorption was considered to mainly influence the recovery of SRB. Hence, OTIS was calibrated using equation (3) and (4) with zero degradation and a tracer specific Kd value from the batch experiments. Light decay was assumed to reduce UR recovery. Hence, OTIS was calibrated using equation (3) and (4) with zero sorption. Thereafter, the transport of EOS was simulated by OTIS accounting for both sorption and light decay. Calibrated sorption parameters λsorp and inline image were complemented with the Kd value obtained for EOS. Light decay rates λMC and λSZ were corrected by the factor κ, determined from the light decay experiments (equation (1) and (2)).

[21] In the OTIS model framework sorption processes are considered to be different in the MC and the SZ [Bencala, 1983]. In the MC reversible sorption follows a linear isotherm (equation 3), whereas sorption processes in the SZ are approximated by irreversible first-order kinetic mass transfer (equation (4)). Our study used a homogeneous sorption rate constant (λsorp (T−1)) for the total wetland. To account for the differences between MC and SZ, inline image was interpreted as a scaling factor (see equation (3)), which described the relation of the sorption rates in the different model areas. To account for the intensity of light decay, different rate constants were calibrated for MC and SZ.

4. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

4.1. Experimental Data

4.1.1. Tracer Data

[22] Figure 4 illustrates the four observed TBCs in the four different experiments in spring with initial and in summer with dense vegetation cover. The SI experiment in spring had comparable flat TBCs with small secondary peaks prior to the main peak. In contrast, the TBCs of the SI experiment in summer showed pronounced peaks, approximately 5 times higher than in spring. The TBCs of the spring CRI experiment showed comparably small main peaks and accentuated tailings but plateaus with a continuous decrease in summer. Hence, the different tracer injection modes (Table 1) produced different TBCs that were difficult to compare without using a transport model.

image

Figure 4. Overview of all experimental tracer breakthrough data. Tracer concentrations are normalized by injected tracer masses. Gray shaded areas depict injection intensities and duration for the constant rate injection (CRI) tracer experiments. SI, slug injection; BR, sodium bromide; SRB, sulforhodamine B; UR, uranine; EOS, eosin.

Download figure to PowerPoint

[23] The recovery rates for all tracers and experiments are summarized in Table 2. Since both spring experiments could not be observed during night, the relative recovery rates were incomplete (Figure 4). Nevertheless, recovery rates of the fluorescent dyes were obviously below those of BR for all experiments. SRB showed highest recovery rates of the dyes except for the spring SI experiment, where almost no reduction of UR occurred. EOS generally showed lowest recovery.

Table 2. Overview of Relative Tracer Recovery
TracerSpringSummer
SIaCRIaSICRI
  • a

    The first number is the relative tracer recovery obtained by incomplete tracer breakthrough curves. The second number is the relative tracer recovery resulting from transport simulations.

BR0.41/1.000.75/1.000.971.01
SRB0.39/0.710.60/0.770.820.85
UR0.39/0.550.21/0.240.380.69
EOS0.33/0.440.19/0.200.270.55
4.1.2. Batch Tests and On-Site Experiments

[24] During the batch tests Kd of UR was found to be 0 (mL cm−3) (<0.05 ± 0.2), while SRB had a Kd of 2.9 ± 0.6 (mL cm−3) and EOS 2.0 ± 1.6 (mL cm−3). Hence, we did not consider sorption for UR in OTIS.

[25] For the summer experiments equation (1) was fitted to tracer concentrations measured in the buckets, yielding coefficients of determination (R2) in between 0.998 and 0.999. During the summer SI experiment δ was 37,600 Wh m−2, resulting in a κ of 1.2 (0.0003 (Wh m−2)−1 for kEOS and 0.00025 (Wh m−2) −1 for kUR, respectively). For the summer CRI experiment δ was 43,800 Wh m−2 and κ was 1.14 (0.0004 (Wh m−2)−1 for kEOS and 0.00035 (Wh m−2)−1 for kUR, respectively). In spring we observed considerably lower radiation input (1558 Wh m−2 and 12,278 Wh m−2 for the SI experiment and the CRI experiment, respectively). The minimum δ during the spring SI experiment was caused by a dense cloud cover. For the transport prediction of EOS we used an average κ of 1.17.

4.2. Solute Transport Modeling

4.2.1. Conservative Transport

[26] With OTIS BR recoveries were simulated with a mean deviation of less than 1% (e.g., Figure 5a and Table 3). Figure 6 depicts the transport parameters of all experiments showing the distribution of parameter values for the 0.2% best model fits. For the CRI and SI experiments NSEs vary within 3% and 5% in summer, and within 15% and 18% in spring, respectively. AACT showed the most obvious variation with high values in spring and low values in summer, whereas the fractions of SZ were relatively similar among the experiments. The dispersion coefficient DL showed high variability and poor identifiability. Parameter values of the summer experiments were significantly higher than those of the spring experiments. The exchange rate α depended on the injection mode: a large difference among the SI experiments was found, but not for the CRI experiments. Considering the best fit parameters of DL and α, we found a decrease of lateral/transversal exchange processes and an increase of longitudinal dispersion processes in summer.

image

Figure 5. Depiction of stepwise model calibration for conservative and nonconservative transport simulations during the summer CRI experiment. (a) BR, (b) SRB, (c) UR, and (d) EOS.

Download figure to PowerPoint

image

Figure 6. Calibration results of conservative transport parameters for all four experiments. The box plots represent the best model fit (points), the 25%, 50%, and 75% intervals (boxes), the 5%–95% interval (bars), and minimum and maximum values (crosses) of the 0.2% best model fits of 30,000 Monte Carlo runs. The axes show the calibration range of each parameter, which was constant for all experiments. DL (m2 s−1) and α (s−1) were logarithmic calibrated and have logarithmic axes. Gray shading depicts parameter ranges of the summer experiments. Black double arrows depict which experiments revealed no significant differences (p = 0.95) between the parameter ranges.

Download figure to PowerPoint

Table 3. Nash-Sutcliffe Efficiencies for All Model Applications
TracerSpringSummer
SICRISICRI
  • a

    Nash-Sutcliffe efficiencies resulting from transport prediction.

BR0.720.950.990.98
SRB0.870.950.990.98
UR0.880.950.990.98
EOS0.73a0.77a0.98a0.96a

[27] Dimensionless RTD's (C′( inline image)) were simulated assuming a virtual unit pulse injection and using the best fit model parameters of each experiment within two different wetland volumes (VOBS/VACT) (Figure 7). Spring RTDs derived by VOBS and VACT did not differ much, while the summer RTDs showed a clear reduction and retardation of the peak for VACT compared to VOBS. Overall, the RTDs of the summer experiments (VACT) showed higher and earlier peaks than the RTDs of the spring experiments. The first moments of the RTDs obtained by VACT were reduced from inline image ≈ 0.6 in spring to inline image ≈ 0.44 for the summer experiments. In spring the difference between VOBS and VACT was marginal: 10% (SI) and 2% (CRI), while in summer VACT was much smaller than VOBS (Figure 8). This suggested a substantial loss of wetland volume (49% (SI), 45% (CRI)).

image

Figure 7. Dimensionless residence time distributions for all experiments. The distributions were derived with a virtual unit pulse injection and the best fit model parameters of each experiment and either the observed wetland volume VOBS or the estimated volume VACT, yielded by OTIS transport simulations.

Download figure to PowerPoint

image

Figure 8. Comparison of observed (water table corrected) wetland volumes VOBS and the estimated volume VACT yielded by OTIS transport simulations. Errors of VOBS were estimated with observed water table oscillations during each experiment. Errors of VACT were estimated with the parameter range of AACT of the best 0.2% best model fits of 30,000 Monte Carlo runs for each experiment.

Download figure to PowerPoint

4.2.2. Nonconservative Transport

[28] Model simulations of the summer CRI experiment for BR, SRB, UR and EOS illustrated the transformation of the TBC´s from conservative (BR, Figure 5a) to nonconservative transport for SRB (Figure 5b) and UR (Figure 5c). Figure 5d presents transport predictions of EOS applying previously obtained conservative, sorption and light decay parameters without any recalibration. NSEs were generally high but were lower during spring because of the missing information of TBC tailings (Table 3).

[29] Sorption rate constants (λsorp) varied in a wider range during spring than during summer, even though the spring SI experiment and the summer CRI experiment did not differ significantly (Figure 9). Also the resulting sediment concentrations ( inline image) showed a higher variation during spring. The variances of λsorp and inline image were an order of magnitude larger in spring than in summer.

image

Figure 9. Results of sorption parameter calibration (SRB) for all four experiments. The box plots represent the best model fit (points), the 25%, 50%, and 75% intervals (boxes), the 5%–95% interval (bars), and the minimum and maximum values (crosses) of the 0.2% best model fits of 30,000 Monte Carlo runs. The axes show the calibration range of each parameter, which was constant for all experiments. Both parameters were logarithmic calibrated and have logarithmic axes. Gray shading depicts parameter ranges of the summer experiments. Black double arrows depict which experiments revealed no significant differences (p = 0.95) between the parameter ranges.

Download figure to PowerPoint

[30] The narrow ranges of light decay rate constants (λMC, λSZ) suggested a good identifiability for the MC model area (Figure 10). Decay rates in the MC were generally higher than those in the SZ, especially in summer.

image

Figure 10. Results of light decay parameter calibration (UR) for all four experiments. The box plots represent the best model fit (points), the 25%, 50%, and 75% intervals (boxes), the 5% to 95% interval (bars), and the minimum and maximum values (crosses) of the 0.2% best model fits of 30,000 Monte Carlo runs. The axes show the calibration range of each parameter, which was constant for all experiments. Both parameters were logarithmic calibrated and have logarithmic axes. Gray shading depicts parameter ranges of the summer experiments. Black double arrows depict which experiments revealed no significant differences (p = 0.95) between the parameter ranges. The observed relative cumulated global radiation (gray line and axis) of all experiments is depicted relative to cumulated global radiation observed during the summer CRI experiment.

Download figure to PowerPoint

5. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

5.1. Conservative Transport and Hydraulic Characterization

[31] One motivation of this study was to analyze parameter ranges obtained by multicriteria analysis of Monte Carlo simulation runs, since poor identifiability of the parameters of the OTIS model for large initial parameter ranges had been demonstrated before [Wagener et al., 2002]. Because of the same uniform posterior setting of the parameter ranges for all experiments, a direct comparison of the resulting best model parameters and its uncertainty was possible. The starting conditions of all Monte Carlo simulations differed according to (1) discharge during each experiment, (2) injected tracer mass, and (3) tracer injection times. Even if the identifiability of the model parameters was not acceptable in all cases, the resulting parameter sets were significantly different between the experiments: 80% of all significance tests revealed significant differences (p = 0.95) between the analyzed parameter ranges (see Figures 6, 9, and 10). Hence, Monte Carlo parameter sets were more robust against equifinality effects than single best fit parameters obtained by iterative parameter estimation. Moreover, they allowed a physical interpretation of conceptual model parameters in relation to observable system changes (Figure 2).

[32] Hydraulic analysis and transport modeling of the BR experiments resulted in a quantitative assessment of wetland development with a clear reduction of active wetland volume VACT when vegetation had developed in summer. This might have been caused by the development of preferential flow paths and stagnant areas, by the development of aboveground and belowground vegetation biomass or by an increasing hyporheic zone (sediment mass), as has been shown in previous studies [Carleton and Montas, 2007; Kjellin et al., 2007; Somes et al., 1999; Stern et al., 2001; Vionnet et al., 2004; Wilson, 2007]. In principle, these changes are expressed by lower first moments and increasing peaks of the dimensionless RTDs. For 15 wetland cells, Martinez and Wise [2003] demonstrated that OTIS can be used to determine the fraction of VDEAD, when VACT is compared to VOBS. We applied the same approach to observe the impact of succession processes on wetland hydraulics. Differences between both volumes were found only for the densely vegetated wetland in summer. Accounting for the delaying impact of storage processes, VACT estimated by OTIS yielded smaller active wetland volumes (and therefore higher efficiencies) than the conventional comparison of tracer moment analysis to VOBS (compare to, e.g., the results of Persson [2000]). Hence, the conceptual description of wetlands by the OTIS model concept allowed us to quantify succession impacts without detailed information on the fractions of VWATER, VHYP, and VBIO. For the management of treatment wetlands this allows estimates of actually active wetland volumes for mitigation. However, the absolute loss of active wetland volume in summer did not change the relative SZ fractions (Figure 6). Instead of increasing SZ areas caused by slow flow paths, aboveground or belowground vegetation development and sediment accumulation, the absolute active wetland volume, which is in contact to the tracer was reduced. When looking at lateral exchange rates α (Figure 6) this fact becomes clearer: constant or rather decreasing exchange rates were not large enough to activate a relatively larger area for solute transport. Comparing SI with CRI in stream tracer experiments, Gooseff et al. [2008] revealed faster rate parameters for lateral flushing. In our study lateral exchange rate parameters α were similar for the CRI experiments, but larger during the spring and smaller during the summer SI experiment. A negative correlation was found correlating DL to discharge. This contradicts the results of D'Angelo et al. [1993], who observed a positive correlation. In our study, gauged discharge was not a measure of flow velocity inside the wetland. During both summer experiments discharges were lower than in spring. Because of the reduction of active wetland volume by succession processes, nominal residence times Tn were shorter in summer. Therefore, mean flow velocities were larger than in spring. This explains the opposite findings compared to the stream tracer experiments of D'Angelo et al. [1993].

[33] A detailed description of interior wetland flow processes is often completed with 2-D model approaches [e.g., Conn and Fiedler, 2006; Koskiaho, 2003]. Those require detailed observations of water depth distributions and vegetation patterns. Ecological succession changes these patterns. For this reason the precision of morphological information decreases with time. Repeated bathymetric measurements would result in a severe disturbance of natural wetland morphology. The lack of spatially explicit information caused by changes due to vegetation development and sediment accumulation results in a clear need to describe successional systems with conceptual models. Conceptual approaches treat wetlands as quasi-two-dimensional system. They describe the exchange to an adjacent storage and quantify those fractions of wetlands which contribute to solute retention or to downstream transport. This differentiation is basically included in all processes that are implemented in the model concept. It must be admitted that lumped model parameters of conceptual transport models may be unable to predict transport under varying hydraulic conditions [e.g., Kadlec, 2000]. The limitations of these concepts to reproduce differences in storage characteristics and multiple tracer peaks, and therefore different flow paths observable in wetlands, have been discussed before [e.g., Choi et al., 2000; Martinez and Wise, 2003; Keefe et al., 2010]. Following the law of large numbers, an increasing number of different flow paths will result in a mean RTD representing all flow paths. This concept is equivalent to porous media solute transport studies which apply the two-region model approach [Coats and Smith, 1964; van Genuchten and Wierenga, 1976].

5.2. Nonconservative Transport

[34] The spring experiments show equal log distributed linear relations of sediment concentrations and rate constants when the parameter interaction between available sediment concentration inline image and sorption rate constant λsorp were compared for the best 60 simulations (Figure 11). Hence, no differences between the model compartments could be identified. The summer experiments showed similar relations, but were less clearly defined and covered smaller ranges. Low values of inline image, which affected sorption only in the MC model area, suggested that sorption during the summer experiments was stronger in the SZ model compartments. Other studies showed that sorption of SRB is strong on mineral surfaces, but comparably low on organic matter [Sabatini, 2000; Smart and Laidlaw, 1977]. Evidence of SRB uptake on biofilms in natural systems could not be found in literature. However, because of steady protein binding, SRB is widely used as a cytotoxicity bioassay [Skehan et al., 1990]. Hence, the relative increase of SZ sorption intensity in summer can be interpreted as an effect of the increasing availability of mineral surfaces following sediment accumulation. In addition, an increase of microbial biofilms in the rizosphere of wetland vegetation may play a role [Kurtz et al., 2003; McKinlay and Kasperek, 1999]. A higher availability of tracer mass for sorption in the MC than in the SZ explains that also the sorbed tracer mass is higher. Also, Keefe et al. [2004] found higher sorption rates (available sediments times rate constant) in the MC than in SZ using OTIS to analyze rhodamine WT transport in a constructed wetland. Here it should also be noted that the MC and SZ model compartments do not represent physical regions but rather dominating process regions that cannot be exactly localized.

image

Figure 11. Comparison of the parameter ranges of the 0.2% best model fits for streambed sediment concentration (μg L−1) and sorption rate constant λsorp (s−1). The simulation results of the spring experiments show a linear relation over the calibration range, while the simulation of the summer experiments resulted in a less clearly defined relation dominated by higher sorption rates and lower sediment concentrations.

Download figure to PowerPoint

[35] Our rates for light decay processes in the MC were 2 orders of magnitude higher than those of other studies which used Rhodamine WT as a tracer [e.g., Keefe et al., 2004]. This large difference in photosensitivity between UR and Rhodamine WT is in accordance to literature with half-life times of 11 and 1300 h for UR and Rhodamine WT, respectively [Leibundgut et al., 2009]. The difference of decay rates between MC and SZ was in line with the OTIS modeling concept, since one would expect less shaded areas in the MC than in the SZ. The conceptual idea of the SZ did not allow a distinction between shading effects caused either by an increasing vegetation cover, or by an increase of the hyporheic zone. When we compared the light decay rates of MC and SZ with the observed cumulative global radiation δ during each experiment (Figure 10), we found that during spring both decay rates increased with increasing δ. For the summer experiments decay rates did not increase further, although δ was more than double. Already Dierberg and DeBusk [2005] had observed that the temporal and spatial occurrence of light decay processes varied because of shading and depth effects.

[36] EOS transport prediction included the interaction between both retention processes: sorbed tracer mass could not be degraded by sunlight and light decay reduced potential tracer mass for sorption. This is documented by Figure 12, which shows details of the simulation results of the tailings during the summer CRI experiment. Desorption of SRB in the MC was suggested by decreasing recession (Figure 12a). UR transport estimations (Figure 12b) did not show this effect. This difference was explained by the steady decay rate constant in the model and by the fact that the tailing was sampled during night when light decay was negligible. The prediction of EOS transport (Figure 12c) showed a combination of both effects: sorption alone would have reduced EOS recovery by 14%, and light decay by 24%. The combined tracer mass reduction was less than the sum (20%).

image

Figure 12. Logarithmic depiction of nonconservative tracer tailings for the summer CRI experiments. (a) SRB, (b) UR, and (c) EOS.

Download figure to PowerPoint

[37] In general, transport and retention estimations for an independent fourth tracer (EOS) by transferring model calibration results of three other tracers (BR, SRB, UR) yielded good results. Thus, the combined application of fluorescent dyes with various physicochemical properties might serve as valuable proxy to asses transport and retention processes of micropollutants in aquatic environments. The increasing public interest to understand pollutant transport and retention processes [Brown et al., 2010] and the increasing employment of constructed wetlands to remediate waters [Rygaard et al., 2011] causes demand for methods to assess these processes. The proposed method in this study is an efficient low-cost tool to estimate effective wetland volumes, which might be relevant for an improved design of constructed wetland management. We observed a quasi-constant retention capacity for sorptive constituents although the SZ volume decreased. This was due to a change of SZ characteristics (emerging vegetation and increasing hyporheic zone), which supports the findings of other studies regarding the benefit of wetlands as biotreatments systems [Gregoire et al., 2009; Kadlec, 1999; Moore et al., 2002].

6. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

[38] The natural succession of constructed wetlands changes their hydraulic characteristics as well as processes controlling transport dynamics for nonconservative solutes. Enhanced understanding of these dynamics allows us to improve management practices and structural design of wetlands in the light of mitigating nonpoint source pollution.

[39] The assessment of wetland succession during a period of five months regarding conservative and nonconservative transport characteristics, yielded a considerable change of dominant flow characteristics and retention processes. Without bathymetric survey, the one-dimensional solute transport model OTIS revealed a 47% reduction of active wetland volume. The change of wetland volume and morphology resulted in an increase of advection and a decrease of exchange processes. Moreover, an increase of sorption and a decrease of light decay intensities could be observed in the retaining areas of the wetland. In contrast, the intensities of sorption and light decay did not change much in wetland areas dominated by advective transport. The application of multitracer experiments in combination with model estimations enabled us to determine intensity and active area of light decay and sorption processes. Moreover, this study demonstrates the potential of applying conceptual solute transport models to multitracers with diverse physicochemical characteristics. They may serve as transport proxies for solutes underlying combined retention processes in complex systems like free water surface wetlands.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

[40] This work was financed by the European Union in the framework of the LIFE-Project ArtWET (06 ENV/F/000133). We would like to thank Emil Blattmann, Klemens Rosin, and various students from the University of Freiburg who helped during the field experiments. We also thank Barbara Herbstritt for her help with tracer analysis. The editors, Graham Sander and Heidi Nepf, and three anonymous reviewers are thanked for their helpful comments and suggestions.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Site
  5. 3. Methods
  6. 4. Results
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgments
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
wrcr13345-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
wrcr13345-sup-0002-t02.txtplain text document0KTab-delimited Table 2.
wrcr13345-sup-0003-t03.txtplain text document0KTab-delimited Table 3.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.