Persistent Trace Organic Contaminants Are Transformed Rapidly under Sulfate- and Fe(III)-Reducing Conditions in a Nature-Based Subsurface Water Treatment System

Subsurface treatment systems, such as constructed wetlands, riverbank filtration systems, and managed aquifer recharge systems, offer a low-cost means of removing trace organic contaminants from treated municipal wastewater. To assess the processes through which trace organic contaminants are removed in subsurface treatment systems, pharmaceuticals and several major metabolites were measured in porewater, sediment, and plants within a horizontal levee (i.e., a subsurface flow wetland that receives treated municipal wastewater). Concentrations of trace organic contaminants in each wetland compartment rapidly declined along the flow path. Mass balance calculations, analysis of transformation products, microcosm experiments, and one-dimensional transport modeling demonstrated that more than 60% of the contaminant removal could be attributed to transformation. Monitoring of the system with and without nitrate in the wetland inflow indicated that relatively biodegradable trace organic contaminants, such as acyclovir and metoprolol, were rapidly transformed under both operating conditions. Trace organic contaminants that are normally persistent in biological treatment systems (e.g., sulfamethoxazole and carbamazepine) were removed only when Fe(III)- and sulfate-reducing conditions were observed. Minor structural modifications to trace organic contaminants (e.g., hydroxylation) altered the pathways and extents of trace organic contaminant transformation under different redox conditions. These findings indicate that subsurface treatment systems can be designed to remove both labile and persistent trace organic contaminants via transformation if they are designed and operated in a manner that results in sulfate-and Fe(III)-reducing conditions.


List of Figures and Tables
Summary of K D values derived from isotherms, microcosms, and field results..........S15 Table S5 Experimental treatments and corresponding shorthand used for labels in Table S6 ...S20 Table S6 Sampling

S-1.1 Additional Analytical Details
An analyte list and spike-recoveries for soil extractions can be found in this section.

S-1.2 Sediment Extraction Methods and Recovery Experiments
Trace organic contaminants were extracted from 5 g of homogenized wet sediment using 12 mL of a 3:5 methanol:Milli-Q water solution via sonication and shaking for 30 minutes each.The process was repeated with 7.5 mL of a 1:1 methanol:Milli-Q water solution and combined extracts were filtered through a 0.2-µm polyethersulfone membrane prior to analysis via LCMS.
Recovery experiments were used to validate the extraction method.Briefly, an aqueous mixture containing approximately 5.8 ng of each trace organic contaminant was spiked into 5 g of moist soil in a 15 mL centrifuge tube.The spiked soil sample was mixed and allowed to equilibrate for approximately 1 hour prior to extraction.Spiked and unspiked samples were extracted to calculate recoveries.All recoveries were calculated using concentrations on a dry weight basis to account for differences in moisture content among soil samples.Results of the recovery experiments can be found in Figure S1.Carbamazepine metabolites (Acridine-COOH, 10-OH-CBZ, and DiOH CBZ) may have had elevated recoveries relative to the other trace organic contaminants due to the use of carbamazepine-d10 as an internal standard for the quantification of those compounds.All other compounds had a corresponding stable-isotope labeled version that was used as the internal standard.

S-1.3 Isotherm Methods
Gravel larger than 0.5 cm was removed from sediment collected for isotherm experiments and the remaining sediment was allowed to air-dry at room temperature for at least 48 hours.After homogenization, 5 g of air-dried sediment was sub-sampled into 50-mL centrifuge tubes and 25 mL of an aqueous solution containing 10 mM CaCl 2 and increasing concentrations of trace organic contaminants was added to each tube.Trace organic contaminants were spiked into the liquid phase by adding them from an aqueous solution containing all the contaminants in a mixture.The centrifuge tubes containing sediment and water were placed on a rotary shaker and allowed to equilibrate for 24 hours prior to sample collection.Isotherms, which were run in triplicate, were conducted at concentrations that extended beyond the concentrations entering the wetland for most compounds.Distribution coefficients were compared to field-derived and microcosm-derived distribution coefficients calculated by pairing porewater concentrations with extractable sediment concentrations.

S-2.1 Mass balance calculations
The following equation was used to describe the mass balance in the wetland over the course of the wetland operation.
Where m is the mass in each wetland compartment over the full operation of the wetland.and are both storage terms that do not change during the monitoring period     because porewater concentrations remained constant over the monitoring period.Therefore, they could be excluded from the mass balance during the monitoring period (i.e., the contaminants did not accumulate in the sediment or porewater over the monitoring period).However, to provide a conservative estimate of the fraction of contaminants removed via other mechanisms, which make up the remainder of the mass that entered the wetland, we assumed that all of the mass of contaminants quantified in each of these compartments ( and ) entered the     wetland during the monitoring period, in the absence of sufficient inflow concentration data prior to the monitoring period.The mass that entered the wetland was quantified using flowrates averaged over entire the monitoring period and inflow concentrations averaged over each observation interval:   = ∫   ()  () = ( ∑  , ∆t ∆ )

S.2
Where m in is the mass of a trace organic contaminant entering the wetland over the entire monitoring period.The average inflow concentration over each monitoring date time interval time interval ∆t, was , and the average inflow flowrate was .The mass that exited the wetland  , ∆t   was quantified by averaging outlet concentrations over the entire monitoring period due to minimal variability in outlet concentrations.

S.3
Where m out is the mass of a trace organic contaminants exiting the wetland over the time interval ∆t at an average outflow concentration, , and outflow flowrate .was approximated       by accounting for evapotranspiration through increases in the concentration of the conservative ion, Cl -. Therefore: The mass of contaminants sorbed to soil particles, , was calculated using isotherm K D   values and porewater concentrations, .Because porewater concentrations varied spatially   () across the wetland, concentrations were integrated over the length of the wetland: A uniform bulk density, , of 1.5 g cm -3 , and cross sectional area, A, equal to the width of each   wetland cell (10 4 cm) times the total depth (10 2 cm), was used to provide a maximum estimate of sorption in the subsurface.
The mass of contaminants in porewater, , was approximated similarly, assuming a   porosity, , of 0.4.

S.7
The mass of contaminants in plants, , taken up during the monitoring period, , was   ∆ approximated using the following equation because plant biomass and concentrations in plants varied over time and space.
Where , is the width of each wetland cell (10 m), , is the total concentration of    () contaminants and their plant-derived metabolites in plants across the wetland per dry mass of plants, is the annual above-ground biomass produced in the wetland per unit area after the  ℎ system had been operating for approximately 2 years and plant communities had matured.Biomass produced during the monitoring period was assumed to be equal to that produced during one year. S.9 In order to generate a range of possible plant uptake scenarios (Table S3), low and high ,  ℎ (the fraction of the total mass of a contaminant taken up into a plant that was eventually   metabolized), and scenarios published in the literature 6 were used and the average of these   scenarios formed the basis of the estimates reported herein.Concentrations measured in meadow plants in the summer of 2019 were estimated to represent half of the annual uptake of carbamazepine because samples were taken at approximately the midpoint of the season and concentrations increase in plants over the growing season. 6e contribution of plant roots to sorption was likely small because the mass of plant roots added to the subsurface per year (<700 kg) 7 is much lower than the mass of sediment in the subsurface (>6 x 10 5 kg).For example: 13 Where BCF root is the bioconcentration factor in roots given by the ratio of the concentration of a trace organic contaminant in roots to the concentration in soil per gram of dry soil and B roots is the annual below-ground biomass produced in the wetland per unit area.BCF root values are much less than 10 2 for hydrophilic compounds such as those monitored in this study. 8

S-2.1.1 Plant Uptake Scenarios
Table S3 Plant uptake scenarios used to estimate the contribution of plant uptake and translocation to the removal of trace organic contaminants in the horizontal levee.

S-2.2 Estimates of HRT
In January of 2020, one of the wetland cells was converted to operation with RO concentrate.During the transition to RO concentrate, Cl -concentrations were monitored in the effluent of the RO concentrate cell periodically (Figure S2) .Monitoring of the increase in Cl - concentrations in the effluent of the RO concentrate cell allowed us to validate the hydraulic retention time estimates made using lithium bromide tracer injections into smaller portions of the wetland previously. 11Using the following set of equations, we estimated that the hydraulic retention time is approximately 19 days in the RO concentrate cell.This is within our prior estimates of HRT of about 12-20 days. 11 []   *   S.16

S-2.3 Contaminant Half-Lives
To calculate contaminant half-lives, we fit first-order removal rate models to porewater data.Examples of rate fitting can be found in Figure S4 & Figure S5.This analysis assumes that the system is operated after contaminant breakthrough and, therefore sorption does not contribute to contaminant removal.In other words, contaminants loaded into each section of the wetland are removed via biotransformation (e.g., the contaminant removal profiles are no longer migrating across the wetland as they would if sorption was the primary mechanism of removal).A lack of seasonal removal and the low contribution of sorption to the mass balance suggests that the contaminant removal profiles were relatively stable in 2019 when rate constants were derived (Section S-4.2 of the SI).

S-2.4 Methods Used to Calculate Distribution Coefficients
All isotherms were linear apart from sulfamethoxazole (Figure S3).Therefore, distribution coefficients were used to estimate sorption given as the ratio between the concentration of contaminants on soil (C s µg/g-dry soil) to the concentration in the aqueous phase at equilibrium (C aq µg/L).

K D =
S.17 Isotherm K D 's were derived by measuring the mass added into each microcosm, , and the   mass measured in the aqueous phase, after equilibration.The amount sorbed to the solid,   , , was calculated as the difference between the mass added and the mass measured after   equilibration.
=   +   S.18 19 Field-and microcosm-derived K D 's were calculated by measuring concentrations in porewater and on sediments through solid-liquid extractions.The mass of contaminants in the porewater of wet sediments was accounted for by assuming porewater concentrations were uniform in sediments using the following equation Where f w is the moisture content of the wet soil.

S-2.5 One-Dimensional Fixed Bed Sorption Modeling and Breakthrough Times
The following equation 12 describes the fate and transport of contaminants undergoing sorption in a fixed-bed column assuming empty bed conditions.Using a numerical solution to this equation, we modeled the anticipated breakthrough curve in the horizontal levee if sorption alone controlled the fate of trace organic contaminants and assumed that sorption is instantaneous (e.g., ignoring sorption kinetics).This assumption provides a conservative estimate of breakthrough time because if sorption kinetics are slower than anticipated, we would expect trace organic contaminants to exit the wetland prior to their sorption breakthrough times.

S.24
Where L wetland and HRT av are the wetland length and hydraulic retention time respectively.
The accuracy of the results of the numerical solution were validated by comparing the time at which 50% of the influent concentration was expected to appear at the outlet (e.g.breakthrough times, ) to those generated by calculating retardation factors, R.  50% ℎℎ  50% ℎℎ =  *   S.25

S-3 Redox Zonation
Redox zones were delineated in individual wetland cells to determine the fraction of contaminant removal occurring in each zone.The nitrate and oxygen reducing zone was defined as the distance from the start of the wetland to the first porewater sample where concentrations of nitrate declined to below 1mg-N/L.We defined the start of the sulfate reducing zone as the first porewater sample that had sulfate concentrations more than 10% less than the influent value and continued until concentrations declined to below 1mg/L.Contaminant removal in each zone relative to the total amount removed was calculated.Wetland cells with overlapping nitrate and sulfate reducing zones were excluded from the analysis because we could not isolate the redox communities in these sections of the wetland.

S-4.2 Impact of Wetland Cell Type and Temperature on Performance
Because the horizontal levee removed trace organic contaminants via biotransformation, water temperature and wetland cell design (e.g., hydraulic residence time) could be important to system performance.For example, the temperature of the wetland porewater varied by about 7 degrees between winter and summer during the monitoring period (i.e., from approximately 9 to 16˚C).Microbial activity typically decreases by about 40% over this temperature change 13 though this estimate may vary significantly between sediment microbial communities. 14rewater concentrations from the summer of 2019 and winter of 2019 showed that all trace organic contaminants were removed at similar rates in the summer and winter (Figure S4).Seasonal shifts in biogeochemistry may have contributed to enhanced winter removal of trace organic contaminants because sulfate-reducing conditions were important for their transformation.For example, during winter, the zone in which nitrate was reduced was not significantly different than the zone in the summer. 15As the activity of heterotrophic organisms declined, autotrophic nitrate-reducing organisms that employ sulfide minerals as an energy source contributed more to nitrate reduction in winter.The increased activity of microorganisms performing sulfide-driven denitrification also supplied additional sulfate to the subsurface.This additional supply of sulfate may have allowed for more sulfate-reducing activity in the winter month.

Compound
Field A lack of seasonal contaminant removal trends suggests that other seasonal factors such as plant senescence or changes in rhizosphere-associated microbial communities were not important for the removal of trace organic contaminants.However, there is a need to study subsurface systems like horizontal levees under conditions that exhibit greater seasonal variations in temperature and microbial activity.

Figure S4
Concentrations of trace organic contaminants versus distance in horizontal levee inflow and throughout wetland cells.Seasonal data were collected from fine cells in winter and summer of 2019.Rate constants did not display statistically significant differences in winter and summer, except for acyclovir, but this is due to low concentrations in the inflow during the winter of 2019.
Different design parameters in each wetland cell type may have influenced horizontal levee performance by changing subsurface mass loading and hydraulic retention times.For example, despite having the same hydraulic grade line, willow cells with coarse sediments exhibited higher subsurface flows (3.6 m 3 /d) than coarse meadow (2.3 m 3 /d) or fine meadow (1.4 m 3 /d) cells during the period when flows were adjusted so that overland flow was eliminated from the wetland cells. 11onitoring of concentrations of trace organic contaminants at the outlet of the wetland cells from the second year of operation showed that the percentage removal of trace organic contaminants in each wetland cell type is similar (Figure 2).However, because loading into each cell was different, the willow and coarse cells removed a significantly higher mass of contaminants than fine cells.
In addition, porewater samples revealed modest differences among cell types.For example, compounds that were removed under sulfate reducing conditions (e.g., carbamazepine) required longer distances for removal in wetland cell types that had sulfate reduction over larger areas (e.g., willow cells) due to higher flow rates and presumably shorter hydraulic retention times in those cells.In contrast, compounds that were removed under nitrate reducing conditions (e.g.metoprolol) were removed over approximately the same distance in all wetland cell types (Figure S5).

Figure S5
Concentrations of carbamazepine and metoprolol versus distance in horizontal levee influent and throughout wetland cells.Data from all three wetland cell types collected in the Summer of 2019 were plotted along with pseudo first order rate fits.Carbamazepine removal rate constants were statistically significantly different, but the magnitude was modest.Metoprolol removal rate constants were not significantly different.

S-4.3 Mass Balance Results
Analysis of data from sorption isotherm experiments (Figure S3) suggested that trace organic contaminant removal could not be explained by partitioning to sediments.For example, assuming that sediments were at equilibrium with measured porewater concentrations, predictions made with isotherm data from the treatment layer (i.e., K D values, Table S4) indicated that sorption accounted for less than 12% of the mass of all compounds that entered the wetland between August 2018 and July 2019 other than trimethoprim and metoprolol (Figure S6).For metoprolol and trimethoprim, less then 45% of the inflow mass could have been removed by sorption.(SI S-2.1 for calculations) However, this analysis only serves to demonstrate that sorption cannot explain the observed removal, not as a quantification of sorption that did occur during the monitoring period.This approach inherently overestimates the contribution of sorption during the monitoring period because contaminant concentrations were relatively stable in wetland porewater during this period.

Figure S6
Mass of monitored trace organic contaminants as a fraction of that which entered the wetland between August of 2018 and July of 2019.The mass fraction in porewater, effluent, plant uptake and sorption wetland compartments were determined based on measurements whereas removal via transformation was estimated by subtraction.
To predict contaminant breakthrough, estimates of hydraulic retention time made with a lithium bromide tracer 11 were validated over the full length of the wetland by monitoring chloride concentrations when the water source of one wetland cell was converted to reverse osmosis (RO) concentrate generated from municipal wastewater reuse (Figure S2).Using this information along with sorption data, we predicted that breakthrough (i.e., the occurrence of effluent contaminant concentrations that were at least 50% of the inflow concentrations) should have occurred for sulfamethoxazole, acyclovir, and carbamazepine within 8 to 12 months of introducing municipal wastewater effluent into the subsurface wetland, which was 4-8 months prior to the date when the first samples were collected.The two compounds with the highest affinities for the sediments in the sand and gravel layer, trimethoprim and metoprolol, were estimated to have travelled approximately 6 to 9 meters and 20 to 30 meters, respectively, by the time that the porewater samples were collected if sorption controlled their removal.Although complete breakthrough was not expected for these compounds, we conclude that sorption was only of modest significance for metoprolol and trimethoprim because their concentrations decreased by over 90% within the first 5 meters of the wetland cells.(Figure 3 See SI S-2.5 for calculations) Additional evidence for the limited role of sorption was obtained by measuring trace organic contaminants in sediments collected approximately 5 meters and 20 meters from the inlet of the wetland in March 2021 (i.e., 47 months after the system started receiving wastewater effluent) and in microcosm experiments.Field-derived and microcosm-derived K D values were either similar to or considerably lower than isotherm-derived K D values, suggesting that the mass balance estimates based upon laboratory isotherms provided a conservative estimate of the contribution of sorption to the removal of trace organic contaminants (Table S4).
Analysis of concentrations of trace organic contaminants in wetland vegetation indicated that plant uptake, translocation and in-planta transformation also did not account for a significant fraction of the observed removal of trace organic contaminants.Among the compounds present in the wetland porewater, those that are uncharged and of intermediate hydrophobicity (i.e., log K ow ranging from 2 to 4) are most susceptible to plant uptake and translocation via the xylem. 16,17ue to their physicochemical characteristics, only a small fraction of the trace organic contaminants present in the porewater entered the plants because they were partially or completely excluded from the transpiration stream due to difficulty bypassing the casparian strip to enter the xylem.Among the contaminants monitored in this study, carbamazepine was expected to have the highest levels of uptake due to its physicochemical characteristics (Table S2) and persistence in the horizontal levee porewater (Figure 3).Also, carbamazepine and its metabolites were the only monitored contaminants that were consistently detected in wetland plant samples. 6Therefore, we assumed that plant uptake and translocation was negligible for all contaminants apart from carbamazepine.
Using carbamazepine plant concentrations measured in various plant species reported elsewhere, 6 estimates of in-planta metabolism 9,10 and biomass measurements, 7 we estimated that plant uptake and translocation removed less than 12% of the carbamazepine that entered the horizontal levee subsurface (Figure S6, SI S-2.1 for calculations).This estimate is similar to estimates of plant uptake and translocation of carbamazepine made in other soil-plant systems. 18

S-4.4.1 Microcosm Methods
Microcosm experiments were performed using fresh sediments collected from the subsurface treatment layer.Sediments from the treatment zone of the nitrate-reducing region and the sulfate-reducing region were placed in plastic containers without headspace, stored on ice and transported back to the lab.Porewater that would later be spiked with trace organic contaminants and terminal electron acceptors for use in the microcosms was collected from approximately 3 meters from the outlet of the wetland.Porewater was collected at each sampling location and paired with sediment extractions prior to the microcosm experiments to determine the initial background mass of trace organic contaminants in the microcosms.
In the lab, approximately 30 g of well-homogenized sediments of each type were subsampled into pre-weighed 20-mL glass scintillation vials, after removing gravel larger than 0.5 cm from the sediment.Approximately 2 mL of a spiked-wetland porewater solution containing the corresponding terminal electron acceptor and a mixture of the trace organic contaminants (approximately 50 µg/L of each) without any residual solvent was added to each vial.Experimental treatments are outlined in Table S5.The time between sediment collection and initiation of the experiment was approximately 10 hours.
Microcosms were incubated in the dark at room temperature (23ºC +/-0.7ºC).Porewater was collected in triplicate by sacrificially sampling microcosms.Incubations lasted 2 days in the nitrate-reducing microcosms and over 28 days in the sulfate-reducing microcosms.Trace organic contaminants were extracted from microcosm sediments collected before and during incubation.A detailed sampling and experimental plan for the microcosm experiments are presented in Table S6.

S-4.4.2 Microcosm Results
We used microcosm experiments to test the removal of trace organic contaminants under different redox conditions because some compounds (e.g., acyclovir, 10-OH carbamazepine and trimethoprim) were fully removed before they reached the sulfate-reducing zone in the full-scale wetland (Figure 3).Analysis of the total mass of trace organic contaminants in microcosm porewater and soil extracts allowed us to attribute removal in the microcosms to transformation.Acyclovir, trimethoprim, and 10-OH carbamazepine, three compounds that were depleted within the first five meters of the inlet in the wetland, were removed in microcosms that contained sediment from both the nitrate-reducing zone and the sulfate-reducing zone (Figure S8a, Figure S9a).These results are consistent with previous reports in which acyclovir was removed under anaerobic conditions in membrane bioreactors 19 and trimethoprim was removed under anoxic and anaerobic conditions in lab-scale experiments. 20However, 9-acridine carboxylic acid, a metabolite of 10-OH carbamazepine, increased and was quickly removed in only the microcosms containing sediment from the nitrate-reducing zone suggesting that the 10-OH carbamazepine degradation pathway differed under nitrate-reducing and sulfate-reducing conditions in the microcosms.
The removal of trace organic contaminants in the microcosms containing sediments collected from the nitrate-reducing zone was more modest than the removal observed in the field likely because only 2.5 mg-N/L of nitrate was available to nitrate-reducing microbial communities in the microcosms.This concentration of nitrate was rapidly removed within 20 hours (i.e.. the first trace organic contaminant sampling point in Figure S8a) indicating that nitrate-reducing conditions were not predominant for the period between 20 and 48 hours.Persistent trace organic contaminants (e.g., carbamazepine and sulfamethoxazole) were also removed in microcosms that contained sediments collected from the nitrate-reducing and the sulfate-reducing zones.][23] Carbamazepine was only removed in these microcosms after nitrate decreased to less than 0.5mg-N/L in the microcosms (e.g., after 20 hours).After this period, sulfate and iron reduction may have occurred in these microcosms (Figure S8b).These results were also consistent with our field results.

Figure S10
Redox dependent removal of carbamazepine and related hydroxylated compounds.Hydroxylation of carbamazepine alters the rates and pathways of carbamazepine transformation under diverse redox conditions.9-Acridine-COOH was only formed under aerobic and nitratereducing conditions.*only observed in microcosm experiments.

Figure S17
Results from sampling before and after treatment plant upgrades.Trimethoprim concentrations were significantly lower in the inflow to the wetland after nitrate concentrations declined, which we anticipated because trimethoprim removal is enhanced in treatment plants that employ anoxic biological nutrient removal, 24,25 but this prevented our ability to interpret the impacts of upstream treatment upgrades to its transformation from field data in the wetland.

Figure S1
Figure S1 Recoveries for monitored trace organic contaminants in spike-recovery experiments (n =3).

Figure S2
Figure S2Increases in the relative concentration of chloride during the transition from wastewater effluent to RO concentrate.The transition to RO concentrate began on January 15, 2020.Full breakthrough was first observed on February 26, 2020.

Figure S3
Figure S3Isotherm results for the monitored trace organic contaminants.

Figure S7 9 -
FigureS7 9-Acridine-COOH was formed in the microcosms that contained sediment from the nitrate-reducing zone, but not in the microcosms that contained sediment from the sulfate-reducing zone.

Figure S8
Figure S8 Total mass of trace organic contaminants remaining relative to initial total mass after the first sampling point in microcosms containing sediment from the nitrate-reducing zone (a) and concentrations of redox-active inorganic species (b).Iron concentrations are represented by the dashed symbol.

Figure S9
Figure S9 Total mass of trace organic contaminants relative to initial total mass after the first sampling point in microcosms containing sediment from the sulfate-reducing zone (a).Concentrations of redox indicator species are also plotted over time (b).Iron concentrations are represented by the dashed symbol.

Figure S11
Figure S11Inlet sulfate concentrations over the monitoring period.

Figure S12
Figure S12Sulfate concentrations measured during the monitoring period at the two water treatment plants: the Orinda WTP and the Upper San Leandro WTP, that supply drinking water to the service area that produces wastewater treated at the Oro Loma Sanitary District wastewater treatment plant.The drinking water at the two treatment facilities have significantly different concentrations of sulfate (e.g., median sulfate concentrations of 41 mg/L and 1.5 mg/L, respectively).During the monitoring period, the Upper San Leandro WTP was taken of service during periods of lower demand or to complete upgrades to the facilities.In contrast, the Orinda WTP remained in service at most times, except when taken out of service for critical maintenance.The difference in sulfate concentrations at the two drinking water treatment facilities partly explains the variation in sulfate concentrations observed in the inflow to the horizontal levee.

Figure S13
Figure S13Redox conditions as indicated by the concentrations of nitrate, dissolved iron in fieldfiltered samples, and sulfate in all three wetland cell types-redrawn from Cecchetti et al.15

Figure S14
Figure S14 Mean concentrations of carbamazepine (dots) and sulfate (triangles) in the influent and coarse, fine, and willow cells along the flow path.Error bars represent the range of measured concentrations at each sampling location.

Figure S15 Figure S16
Figure S15 Scatter plots showing trace organic contaminant concentrations and sulfate

Figure S18
Figure S18Ratio of chloride in the outflow to chloride in the inflow used to normalize concentrations for evapotranspiration.

Figure S19
Figure S19 Horizontal levee in January 2017 view from the effluent sampling wells

Figure S21
Figure S21 Horizontal levee in winter of 2022 (left) and summer of 2021 (right) at approximately the same location near the inlet

Table S2
Physicochemical properties of the monitored trace organic contaminants.All values were obtained from PubChem.Values with an * are estimates, values without an * are from laboratory measurements reported in the literature.

Table S4
Summary of K D values derived from isotherms, microcosms, and field results.

Table S5
Experimental treatments and corresponding shorthand used for labels in TableS6

Table S6
Sampling plan for the microcosm experiments including plans for periodic extraction and porewater sampling