Nutrient removal in a slow-flowing constructed wetland treating aquaculture effluent

Danish model trout farms (MTFs) use stream-like constructed wetlands for effluent polishing, and the industry is keen to improve wetland removal efficiency. To facilitate this, we examined longitudinal and seasonal nutrient removals in an MTF wetland with a hydraulic retention time (HRT) of 1.7 d, a free water surface (FWS) area of 7510 m2, and a volume of 6008 m3. Biweekly, 24-h composite water samples were obtained for 1 yr at 6 sampling stations along the wetland. Assuming plug flow conditions, reductions in particulate and dissolved nutrient concentrations were modelled as first-order removal processes, and removal rate constants (k1,A, m d−1) were plotted to reveal seasonal fluctuations. Particulate phosphorus and organic matter k1,A fluctuated more or less randomly through the year, reflecting that particulate nutrient removal predominantly takes place by sedimentation. In contrast, dissolved nitrogen, phosphorus, and organic matter k1,A fluctuated seasonally, demonstrating that dissolved nutrient removal relies on biologically mediated processes. Temperature oscillations probably governed the observed seasonal fluctuations in nitrate-N k1,A and could be approximated with an Arrhenius temperature coefficient of 1.07. Furthermore, denitrification appeared to be carbon-limited. Incoming dissolved phosphorous and ammonia became incorporated in the natural wetland growth cycle that in cluded periods of net removal and release, resulting in minimal annual net removal. In summary, this study shows that improving nitrate removal in a slow-flowing MTF wetland would require some kind of carbon dosing, while further improving ammonia and phosphorus removal would require a reduction of the amounts of ammonia and dissolved phosphorus entering the wetland.


INTRODUCTION
Within the last 10−15 yr, many Danish fish farmers have reconstructed their land-based flow-through systems into model trout farms (MTFs;Jokumsen & Svendsen 2010), and approximately half of the Danish rainbow trout Oncorhynchus mykiss production (15 070 t yr −1 ) currently takes place in these systems (Statistics Denmark 2020). MTFs are semi-intensive recirculation systems that, depending on type, typically apply between 2−4 m 3 (type III MTFs) and 5−24 m 3 (type I MTFs) make-up water (MUW, i.e. water exchange) per kg feed (Svendsen et al. 2008a, Danish Ministry of the Environment 2012, Dalsgaard et al. 2018). Type III MTFs are the most common and most intensive MTFs, producing approximately 10 100 t of fish per year (Statistics Denmark 2020). They use ground or surface water as MUW and they are equipped with inline sludge cones, drum filters, and biofilters. Less intensive MTFs (type I) use stream water as MUW and generally do not apply biofiltration (Pedersen et al. 2003). To comply with the EU Water Framework Directive water quality standards (WFD 2000), all Danish MTFs are required to treat system effluent in a free water surface (FWS) constructed wetland or ensure equivalent treatment by other methods (Danish Ministry of the Environment 2016). An MTF wetland is a passively operated treatment unit that is relatively inexpensive to build, often constructed by interconnecting abolished earthen fish ponds in a stream-like meandering fashion (DCE 2012). System reject water (corresponding in volume to system MUW) and sludge basin runoff run to the wetland, which removes excess nutrients by natural processes prior to final discharge.
An MTF wetland differs from most other constructed wetlands applied for wastewater treatment in that the hydraulic loading rate (HLR) is more or less constant throughout the year, the waste nutrient composition is well known, and nutrient loadings within a MTF are within the same magnitude throughout the year, reflecting that fish are produced year round (Svendsen et al. 2008a, Kadlec & Wallace 2009, Jokumsen & Svendsen 2010. To document their effect, Svendsen et al. (2008a) measured inlet and outlet concentrations and flows for 2 yr in 8 of the firstever constructed MTF wetlands. FWS areas in the 8 wetlands ranged from 1375 to 14 800 m 2 and the hydraulic retention time (HRT) ranged from 0.9 to 2.0 d. Mass-balance calculations showed that total nitrogen (TN), total phosphorus (TP), total biochemical oxygen demand (BOD 5-TOT ), and total chemical oxygen demand (COD TOT ) were reduced by 50, 76, 93, and 87%, respectively, corresponding to mass removal rates of 2.7 g TN, 0.18 g TP, 4.4 g BOD 5-TOT , and 13.1 g COD TOT m −2 d −1 . These rates have since been applied for sizing other aquaculture wetlands including faster-flowing wetlands with shorter HRT associated with type I MTFs.
All Danish MTFs are subjected to discharge control. This means that there is no upper limit to how many fish MTFs may produce as long as wetland effluent total ammonia N (TAN) and BOD 5-TOT concentrations, and discharged TN and TP masses, stay below certain farm-specific regulatory limits. The industry is therefore keen to improve and ensure constant wetland treatment performance year round. Due to a lack of empirical data, it is assumed that nutrients are removed at a constant rate throughout the year, and a better understanding of internal removal dynamics at the flows and loading conditions characteristic for type III MTF wetlands is needed. To facilitate this, the present study for the first time examined the longitudinal and seasonal removal efficiency of N, P, and organic matter (OM) in a slowflowing type III MTF wetland for a full year. Furthermore, rather than examining total nutrients as done in most wetland studies (see Kadlec & Wallace 2009), particulate and dissolved nutrient fractions were analysed separately to account for different removal mechanisms. A set of simple first-order removal rate constants were derived for all particulate and dissolved nutrient fractions, and the rate constants were plotted against time to disclose seasonal fluctuations in nutrient removal efficiency. Furthermore, rate constants were applied to assess the specific wetland surface area needed to remove a given fraction of certain nutrients in order to keep effluent nutrient concentrations within given limits.

Study site
The 1 yr monitoring study was conducted in a 10 yr old, slow-flowing Danish type III MTF wetland with a mean ± SD HLR of 0.49 ± 0.10 m d −1 and an average width and depth of approximately 8.5 and 0.8 m, respectively ( Fig. 1a,b). The former flow-through farm was rebuilt in 2007 and produced ap proximately 525 t market-size rainbow trout Oncorhynchus mykiss per year applying 500 t feed. The fish farm used groundwater as MUW. System reject water (averaging 42 ± 9 l s −1 during monitoring) flowed continuously into the wetland, which was constructed by interconnecting abandoned earthen ponds no longer used for fish production. Cutting off 2 minor wetland sub-sections prior to the study (red dotted lines in Fig. 1b) ensured that total farm effluent volume entered a single well-defined wetland string during monitoring. In addition to system reject water, a sludge basin overflow entered the monitored wetland section immediately upstream of sampling Stn 1 (Fig. 1b). The monitored wetland had an FWS area (A) of 7510 m 2 , a volume (V ) of 6008 m 3 , and a total HRT of 1.7 ± 0.4 d (Table 1). Wetland vegetation was dominated by Glyceria spp. growing along the banks of the open-water, stream-like wetland, while duckweed (Lemna spp.) covered parts of the surface area in the main growing season (Fig. 1).

Sampling
Six sampling stations were positioned at regular distances along the wetland (Fig. 1b, Table 1), assuming that water moved as a plug flow, i.e. at constant velocity and flow direction throughout. Given the constant daily flow, biweekly, 24 h composite water samples were obtained at each station for a year (21 December 2017−05 December 2018), corresponding to 26 sampling days (n = 26). Water samples were obtained using refrigerated automatic samplers (Glacier® Portable, Teledyne ISCO) programmed to sample 300 ml from the middle of the water column every hour for 24 h. Temperature (T, °C), dissolved oxygen (mg l −1 ), and pH were obtained by separate measurements in the water column next to sampler inlet pipes at each sampling station upon sample collection using Hach Lange HQ40D multimeters. Flow rates in and out of the wetland were derived by measuring water heights (cm) across well-defined inlet and outflow weirs (Fig. 1c,d) and applying the fundamental weir equation (Winther et al. 2011).

Chemical analysis
Upon return (within 2 h) to DTU Aqua's Section for Aquaculture in Hirthshals, Denmark, sampling containers were gently shaken to obtain homogeneous subsamples for total (TOT) and dissolved (DISS) nutrient analysis, while particulate (PART) nutrient concentrations were determined as the difference between the 2 measurements. Subsamples were filtered through 0.2 μm sterile syringe filters (Filtropur S 0.2, Sarstedt) prior to analysis of N DISS compounds, while other subsamples were filtered through 1.6 μm glass microfiber filters (Whatman® GF/A, GE Healthcare) prior to analysis of dissolved organic compounds and P DISS compounds. TN was analysed according to International Organization for Standardization (1986, 1997, and nitrite (NO 2 -N) were analysed following Danish Standards Foundation (1975), International Organization for Standardization (1986), and Danish Standards Foundation (1991; detection limit 1 μg l −1 ), respectively. For total and dissolved OM, carbonaceous BOD 5 was analysed as described in International Organization for Standardization (2003b) modified by adding allylthiourea to inhibit oxygen consumption due to nitrification (International Organization for Standardization 2003a, Tchobanoglous et al. 2003), while COD was analysed as described in International Organization for Standardization (1989) using LCK 114 digestion vials (Hach Lange). TP and P DISS (PO 4 -P) were analysed as described in International Organization for Standardization (2004; detection limit 0.01 mg l −1 ). All analyses were performed in duplicate.

Modelling and statistics
Many biologically mediated wetland processes exhibit Monod kinetics, i.e. first-order reactions at limiting concentrations of a key substrate and zero-order reactions at non-limiting concentrations (Mitchell & McNevin 2001). Consistent with this, many wetland re moval processes, including sedimentation and sorption, behave as first-order processes within observed concentration ranges (Kadlec & Wallace 2009). Given measured inflow concentrations in the present study, and assuming plug flow conditions (long and relatively narrow channel), nutrient removal along the wetland was therefore modelled as a first-order process applying the equation (Kadlec & Knight 1996): where C is the nutrient concentration (mg l −1 ) at a specific sampling station along the wetland, C 0 is the nutrient concentration (mg l −1 ) at inflow Stn 1 (see Fig. 1b), C* is the nutrient background concentration (mg l −1 ), k 1,A is the area-based first-order rate constant (m d −1 ), and q is the hydraulic loading rate (HLR; m d −1 ) at a specific wetland station relative to Stn 1. According to this model, nutrient concentrations decline asymptotically towards a residual background concentration (C*) characterized by no net nutrient uptake or conversion. The same background concentrations as applied by  for a faster-flowing type I MTF wetland (HLR of 2.23 versus 0.49 m d −1 in the present study) were used (Table 2). Values for k 1,A were derived for each sampling day by fitting Eq. (1) to measured concentrations at each station along the wetland that day, and minimizing the sum of squared residuals between observed and predicted nutrient concentrations using the Solver GRG non-linear function in Microsoft Excel®. Rate constants were subsequently plotted against sampling date to disclose seasonal fluctuations. In case of apparently non-limiting nutrient concentrations, nutrient removal was modelled as a zeroorder process applying the equation (Mitchell & McNevin 2001): where C out is the nutrient concentration (mg l −1 ) at outflow Stn 6, k 0,V is the zero-order volumetric rate constant (g m −3 d −1 ), and τ is hydraulic retention time (d).
In addition to nutrient removal due to biologically mediated turnover and physical sedimentation, removal of dissolved nutrients due to downward infiltration/ leakage and transpiration was estimated as: ( 3) where r is the nutrient removal rate (kg d −1 ) due to downward infiltration and transpiration, C i is the nutrient concentration at a given wetland station, i (mg l −1 ), and Q gw,i is the downward infiltration and transpiration rate (m 3 d −1 ) between the inlet station and any given downstream wetland station, calculated as: where Q in is the wetland inlet flow rate (m 3 d −1 ), Q out is the wetland outlet flow rate (m 3 d −1 ), A is total wetland surface area (m 2 ), and A i is measured wetland surface area between inlet Stn 1 and a given downstream wetland station (m 2 ). Correlations were examined using Pearson product moment correlation tests, and a modified Arrhenius temperature coefficient was derived using Eq. (5) to resolve the impact of temperature on NO 3 -N removal (Kadlec & Wallace 2009): where k T is the rate constant at a given temperature, k 20 is the rate constant at 20°C, and θ is the dimensionless temperature correction coefficient. Statistical analyses were performed using Graph-Pad Prism 8.4.1 (© 2018 GraphPad Software) considering p ≤ 0.05 as statistically significant. Results are shown as means ± SD, and concentration time series are shown as 3-point moving averages with missing data points replaced by arithmetic means of neighbouring data points.

Water measurements and flow rates
Wetland water temperatures oscillated in a sinusoidal manner (Fig. 2a), as also seen in other temperate  . Drought, rather than altered operation routines, therefore likely explains the fact that inlet flow rates dropped from an average of 50 l s −1 in May to 31 l s −1 in August, while outlet flow rates dropped from 44 to 25 l s −1 in February− April. The changes corresponded to an average water loss of 12% in December− February, increasing to above 50% in May, and stabilizing at approximately 22% in August− November (Fig. 2c).
Oxygen concentrations (measured during daytime) were highest in February−March (> 8.6 mg l −1 at all stations) and lowest in August (<1 mg l −1 downstream of Stn 2), and were negatively correlated with wetland temperatures (r = −0.601, p < 0.0001, n = 149). In addition, oxygen concentrations increased in the flow direction during spring and decreased in the flow direction in autumn (Fig. 2b). The wetland was slightly alkaline, with pH values (measured during daytime) averaging 7.2 ± 0.3 (n = 149; Fig. 2d), and there was a strong correlation be tween pH and oxygen (r = 0.749, p < 0.0001, n = 148).

Nitrogen
Inflowing TN concentrations were relatively constant through the year (17.2 ± 2.2 mg TN l −1 ) including 97% dissolved N compounds. N PART was not detected in many in stances and results on this fraction are therefore not included. NO 3 -N constituted approximately 80% of TN DISS , decreasing from 13.2 ± 1.4 mg l −1 at Stn 1 to 11.1 ± 2.3 mg l −1 at Stn 6 ( Fig. 3c). TN DISS k 1,A fluctuated more or less randomly throughout the year (0.084 ± 0.043 m d −1 ; Fig. 3b), whereas NO 3 -N k 1,A oscillated between relatively low k 1,A values in autumn and winter (0.063 m d −1 in October− February) and high k 1,A values in spring and summer (0.183 m d −1 in May−August; Fig. 3d). Furthermore, the k 1,A values correlated significantly with average wetland temperatures (r = 0.648, p < 0.001, n = 24), and the oscillations corresponded to a temperature correction coefficient (θ) of 1.07 (p = 0.007, R 2 = 0.311, SE = 0.024) and a k 20 of 0.20 m d −1 .
There was a relatively large input of TAN from the sludge basin, with concentrations increasing from 1.33 ± 0.56 mg l −1 prior to the sludge basin inflow (data not shown) to 2.27 ± 1.23 mg l −1 at Stn 1 (Fig. 3e). Ammonia removal oscillated in a seasonal manner interchanging between net removal in spring and late autumn and net production in early spring and late summer. As Eq. (1) does not apply to negative removal, Fig. 3f instead shows the constant removal rate (g m −2 d −1 ) based on flow and concentration differences between Stns 1 and 6 divided by the monitored wetland area. NO 2 -N entered the wetland at 0.477 ± 0.123 mg l −1 (Fig. 3g) and like nitrate, NO 2 -N k 1,A followed a sinusoidal pattern (Fig. 3h). The k 1,A values were significantly correlated with both NO 3 -N k 1,A (r = 0.527, p = 0.008, n = 24) and average wetland temperatures (r = 0.450, p = 0.024, n = 25).

Phosphorus
P PART concentrations averaged 0.380 ± 0.475 mg P l −1 at the inlet (Fig. 4a), and there was a strong corre- lation through the year between P PART and COD PART concentrations (r = 0.929, p < 0.0001, n = 22). P PART concentration k 1,A values averaged 1.14 ± 0.61 m d −1 (Fig. 4b) and were significantly correlated with those for COD PART (r = 0.662, p = 0.001, n = 21). P DISS concentrations averaged 0.479 ± 0.237 mg l −1 at Stn 1 and, as similarly observed for TAN (Section 3.2), removal of P DISS oscillated be tween periods of net re lease and net removal (Fig. 4c). As for TAN, Fig. 4d therefore shows the constant removal rate of P DISS (g m −2 d −1 ) calculated from the differences in flows and P DISS concentrations be tween Stns 1 and 6, divided by the monitored wetland area. There was a strong correlation between TAN and P DISS removal rates (Figs. 3f & 4d; r = 0.930, p < 0.0001, n = 24).

Organic matter
As for P DISS and TAN, a large fraction of the OM entering the wetland came with the sludge basin influx, leading to an increase in COD PART concentrations from 4.4 ± 4.2 mg l −1 prior to the sludge basin inflow to 10.6 ± 5.3 mg l −1 at Stn 1. In comparison, COD DISS in creased from 14.9 ± 1.5 to 18.2 ± 1.9 mg l −1 . COD PART concentrations de creased to 2.6 ± 2.3 mg l −1 at the outflow station (Fig. 5a), and k 1,A oscillated in a seemingly nonsystematic manner, averaging 1.08 ± 0.66 m d −1 (Fig. 5b). Decreases in COD DISS concentrations were less pronounced (Fig. 5c) and k 1,A values were lower (0.12 ± 0.07 m d −1 ). Furthermore, the k 1,A values appeared to decline slightly from December onwards (Fig. 5d).
BOD 5-TOT doubled following the sludge basin inflow (from 5.7 ± 2.5 to 11.2 ± 5.5 mg l −1 at Stn 1), and approximately 55% of BOD 5-TOT was in particulate form, while 45% was in dissolved form (Fig. 5e,g). Removal rate coefficients of BOD 5-PART averaged 1.03 ± 0.57 m d −1 and were significantly correlated with those for COD-PART (r = 0.609, p = 0.004, n = 20). Similarly, the k 1,A values for BOD 5-DISS were positively correlated with those of COD DISS (r = 0.768, p < 0.0001, n = 24),  Table 3 summarizes the estimated mass removal of nutrients (kg d −1 ) by internal wetland processes (see Eq. 1) and downward infiltration and transpiration (see Eqs. 3 & 4), and compares the sum to the measured mass removal. There was generally a very high consistency between the estimated and measured mass removal. The table also shows that the mass removal of OM PART and P PART was larger than the corresponding mass removal of OM DISS and P DISS . Furthermore, the estimated mass removal of dissolved nutrients due to downward infiltration and transpiration was in most in stances larger than that due to internal wetland processes, while it was opposite for particulate nutrients.

Particulate matter removal
The longitudinal concentration profiles of P PART , BOD 5-PART , and COD PART and respective fits of the first-order model support the hypothesis that particulate nutrient re moval in this slowflowing MTF wetland predominantly took place by sedimentation, as also shown by Dalsgaard et al. (2018) in a faster-flowing type I MTF wetland (HLR of 2.23 versus 0.49 m d −1 in the present study). Most of the particulate nutrient re moval took place in the first part of the wetland, diminishing in an ex ponential manner with wetland area/ HRT. By inserting the average yearly inflow together with the average P PART k 1,A value and inlet concentration in Eq. (1), it can be estimated that 75% of the P PART re moval took place within the first 3600 m 2 of the wetland, while 95% was removed within 6300 m 2 . Along with the results for BOD 5-PART and COD PART , which rapidly de creased in a similar fashion, these re sults stress that particulate nutrient treatment efficiency diminishes as the wetland area increases. Hence, rather than improving net particulate matter removal, a larger area/longer HRT may augment the growth of phytoplankton, which may lead to elevated OM and P discharge concentrations, especially in spring and summer when light intensities are high (Kadlec & Wallace 2009). Adding to this, mineralisation of dead plant biomass may seasonally increase the release of OM PART and P PART into the water column (Christensen et al. 1990, Kadlec & Wallace 2009), similarly affecting wetland discharge con centrations.

Nitrate removal
All MTFs with a feed loading above 0.2 kg feed m −3 MUW use nitrifying biofilters, and nitrate is therefore the main nitrogen form in associated treatment wetlands. Nitrogen removal efficiency in existing slowflowing type III MTF wetlands is typically around 50% (Svendsen et al. 2008a) and the discharge of nitrate is often the 'first limiting nutrient' with regards to a production expansion. Denitrification is  (Reddy & Patrick 1984, Christensen et al. 1990, Bachand & Horne 1999, and previous stud-ies have shown that denitrification rates in anaerobic FWS sediments rely not only on nitrate but also on readily available carbon. Because of this dependency of a second substrate, firstorder kinetics typically better describes the removal of nitrate despite seemingly non-limiting (zero-order) concentrations of nitrate (Reddy et al. 1982, Reddy & Patrick 1984, Kadlec & Wallace 2009). This also applied to the present study, where inlet and outlet NO 3 -N concentrations (13.2 ± 1.4 and 11.1 ± 2.3 mg NO 3 -N l −1 , respectively) otherwise indicated that nitrate removal should be a zero-order process. Hence, modelling nitrate re moval as a first-order process using Eq. (1) yielded a stronger correlation be tween observed and predicted outlet concentrations (r = 0.994, p < 0.0001, n = 24) than if re moval was modelled as a zero-order process using Eq.
(2) (r = 0.747, p < 0.0001, n = 25). Furthermore, a plot of estimated yearly average NO 3 -N re moval rates along the wetland against average COD PART concentrations in the water column showed an asymptotic in crease in nitrate removal rates as COD PART concentrations in creased (R 2 of semi logarithmic regression = 0.984, n = 5, plot not shown), corroborating the finding that denitrification was carbon limited. A previous study by von Ahnen et al. (2020) similarly found denitrification in a slow-flowing MTF wetland to be limited by readily available carbon. Modulating the flows in 2 parallel wetland streams, the authors showed that TN removal rates were significantly higher in the sludge-fed wetland side stream (8.4 ± 1.4 g N m −2 d −1 ) than in the parallel side stream (1.8 ± 1.0 g N m −2 d −1 ) treating the same nitrate-rich effluent but re ceiving no carbonaceous sludge. As carbon availability and oxygen conditions vary within and between MTF wetlands, and as simple NO 3 -N k 1,A values do not account for this, simple NO 3 -N k 1,A values are strictly site specific (Reddy et al. 1982), which also applies to the present study. In addition to being carbon lim-  1 and 6 shown for simplicity) and associated removal rate constants (k 1,A , m d −1 ) of (a,b) particulate (COD PART ) and (c,d) dissolved chemical oxygen (COD DISS ), and for (e,f) particulate (BOD 5-PART ) and (g,h) dissolved biochemical oxygen demand measured over 5 d  ited, denitrification in the present study also depended on wetland temperatures, corroborated by an Arrhenius temperature coefficient of 1.07. This value is similar to that observed in other FWS wetlands (e.g. Bachand & Horne 1999, Kadlec & Reddy 2001, de Klein et al. 2017) and shows that temperature also has a strong effect (directly and indirectly) on denitrification in slow-flowing MTF wetlands. A minor removal of nitrate and ammonia possibly also took place via anaerobic ammonium oxidation bacteria (anammox). This 'sensitive' and strictly anaerobic microbial process, combining ammonium and preferably nitrite (and to lesser extent nitrate) directly to form nitrogen gas, is ubiquitous in most freshwater systems where denitrification is carbon limited, and may occur in both wetlands and denitrifying bioreactor systems (Kuenen 2008, Rambags et al. 2019. Processes other than denitrification also remove nitrate. Macrophytes and microalgae take up nitrate in the growing season, but decomposition releases most of it back to the water and only minor fractions are buried as irreducible residuals in the sediment (Bachand & Horne 1999, Kadlec & Wallace 2009). Dissimilatory nitrate reduction is another removal process that converts nitrate to ammonia, but this process is usually more prevalent in carbon-rich environments (Tiedje et al. 1983). In contrast, nitrification may convert nitrite to nitrate. Nitrification is typically highest in the peak growing season, when oxygen concentrations can be high (especially dur-ing daytime) and plant litter and stems may increase biofilm areas by as much as a factor of 10 (Kadlec & Wallace 2009).

P DISS and TAN removal
A large portion of P DISS and TAN in the wetland was apparently derived from mineralisation in the associated sludge basin (based on concentration measurements before and after the sludge basin inflow at Stn 1; data not shown). Phosphorus and nitrogen otherwise removed as solids in the production unit were thereby reintroduced as dissolved nutrients into the wetland. Here, they seemingly were incorporated in the annual growth cycle involving plant and microbial uptake and release, nitrification, and mineralisation (Simmons & Cheng 1985, Christensen et al. 1990, Kadlec & Reddy 2001. This resulted in small average net annual removal rates of TAN (0.398 ± 0.766 g m 2 d −1 ) and P DISS (0.045 ± 0.129 g m 2 d −1 ) as also observed in other FWS wetlands (Kadlec & Wallace 2009). In fact, the wetland served as a net producer of TAN and P DISS at times (Figs. 3f & 4d).
Reflecting the annual growth cycle in the wetland, increased photosynthesis and CO 2 uptake during daytime may explain the higher pH and oxygen concentrations measured in spring (Fig. 2), while plant decomposition and ammonification may explain the lower pH and net TAN production measured in late summer, when temperatures were high and oxygen concentrations low (Reddy & Patrick 1984). Processes other than plant uptake and nitrification also remove TAN. As discussed in Section 4.2, anammox removes TAN directly (Kuenen 2008, Rambags et al. 2019) and minor removal of TAN by anammox possibly occurred downstream in the wetland where denitrification presumably was carbon limited. In addition, un-ionized ammonia is easily lost through volatilization (Reddy & Patrick 1984), but such losses were probably insignificant here given that pH was generally below 7.5.

OM DISS removal
As for OM PART , OM DISS removal was largest in the first part of the wet land while it diminished in an exponential manner along the wetland. Unlike particu-   Table 2 for abbreviations late matter, OM DISS k 1,A values fluctuated in a seasonal manner, and using Eq. (1) it can be estimated that approximately 5 times more BOD 5-DISS was removed during winter when k 1,A values were high than in summer/autumn when they were low (applying similar inlet concentration). Other temperate FWS wetland studies have found similar seasonal patterns in OM DISS removal, coupling it to oxygen solubility at different temperatures (lower oxygen solubility at higher temperatures). In addition, leaching of less degradable OM from plant material may contribute to reducing net removal (Pinney et al. 2000, Kadlec & Reddy 2001. This probably also applied to the present study, where BOD 5-DISS k 1,A values were slightly negatively and positively correlated, respectively, with wetland water temperatures (r = −0.634) and oxygen concentrations (r = 0.433). Furthermore, longitudinal concentration profiles indicated that there was a net production of less degradable COD DISS at times (data not shown) presumably deriving from the decomposition of dead plant biomass and a release of less degradable humic substances (Kadlec & Wallace 2009). Humic substances are a natural source of acidity, and a production of humic substances would explain the reduction in pH observed in summer and autumn (Fig. 2d) when temperatures were high.

Nutrient mass removal by infiltration and evapotranspiration
It was an unusually hot summer in 2018, including the most extended drought ever registered in Denmark (Cappelen 2019), and up to 50% of wetland water was lost in April−July (Fig. 2c). Wetlands treating MTF effluents are non-lined, non-planted, streamlike systems and measuring water balances for 2 full years in 8 type III MTF wetlands, Svendsen et al. (2008a) deduced that influxes and downward infiltration dominated the balances, whereas impacts of evapotranspiration and rainfall were insignificant, contributing at maximum ± 0.3 l s −1 on a yearly basis. Consistent with this, we assume that water losses in the present study were primarily due to downward infiltration along with transpiration from macrophytes growing along the banks, as there was principally no emergent aquatic vegetation within the wetland channel (Fig. 1). Downward infiltration and transpiration do not affect dissolved nutrient concentrations in the water column per se, as water lost by infiltration and transpiration has the same nutrient concentration as the water column. In addition, dissolved nutrient turnover in an FWS wetland is pre-dominantly associated with processes occurring in plants, biofilms, and at sediment interfaces rather than processes occurring in the water column. A change in water volume (i.e. water column depth) will therefore not proportionally affect the turnover of dissolved nutrients (Kadlec & Wallace 2009). Similarly, downward infiltration supposedly had a minimal effect on residual particulate nutrient concentrations given that most particulate nutrients, as discussed in Section 4.1, settled in the first part of the wetland with concentrations rapidly approaching a constant background level. We therefore assume that downward infiltration and transpiration had minimal ef fects on the particulate and dissolved k 1,A values derived by fitting Eq. (1) to the measured nutrient concentrations. Mass removals calculated using these k 1,A values will, however, be underestimated (i.e. a conservative measure), as dissolved nutrients re moved by downward infiltration and transpiration are not accounted for. Combined downward infiltration and transpiration losses were therefore estimated separately using Eq. (3), and as seen in Table 3, there was high consistency between the estimated and measured total removal. This shows that nutrients removed by sedimentation and internal turnover processes were adequately accounted for by the simplistic first-order plug flow model (within ob served concentration ranges), while predicted overall mass removal will be underestimated in the case of large downward infiltration and transpiration losses. Table 2 summarizes the nutrient loads (g m −2 d −1 ) and constant removal rates (g m −2 d −1 ) obtained in different MTF wetland studies. Loading rates in the present study were largely comparable to those in a faster-flowing type I MTF wetland , where lower inlet concentrations typically accompany a higher inflow. The faster-flowing wetland was, however, evidently less efficient, especially in removing nitrate. As in the present study, nitrate entered the faster-flowing wetland in seemingly nonrate limiting concentrations (> 3 mg NO 3 -N l −1 ) and carbon loading was similar, if not higher, than in the present study, suggesting favourable conditions for denitrification. Hydraulic conditions in the fasterflowing wetland, however, resulted in higher oxygen levels (inlet oxygen concentrations > 8 mg l −1 throughout the year), presumably promoting nitrification rather than denitrification and leading to a net pro-duction of nitrate at times. Nutrient loadings and removal rates were also largely comparable to those obtained by Svendsen et al. (2008a) in 8 similar slowflowing MTF wetlands. Therefore, we assume that the k 1,A values obtained in the present study generally apply to other type III MTF wetlands, except for nitrate, where k 1,A values depend on site-specific carbon availability and oxygen conditions.

CONCLUSIONS
Danish MTF effluent regulation is based on wetland discharges of TN, TP, and total BOD 5 , and it is assumed that nutrients are removed at constant rates independent of wetland size and season (Danish Ministry of the Environment 2016). As shown in the present study, nutrient removal is, however, a combination of 2 processes happening at different rates. Particulate nutrient removal largely takes place by sedimentation and typically happens at faster rates than that of dissolved nutrients. Dissolved nutrient removal largely relies on biologically mediated turnover, and k 1,A values tend to fluctuate in a seasonal manner, coupling to biotic and abiotic processes. Improving wetland treatment performance must therefore take account of these differences.
Approximately half of the P and OM entering a type III MTF wetland is typically in particulate form, and as concentrations rapidly ap proach a residual background level, enlarging a wetland beyond this point will not improve particulate nutrient removal any further. Rather, a larger wetland/longer HRT may augment the growth of phytoplankton and elevate discharge concentrations of OM PART and P PART . Improving wetland treatment performance therefore principally comes down to enhancing dissolved nutrient removal. Nitrate is typically the 'first limiting nut rient' in Danish MTF wetlands with respect to production expansions. Denitrification in MTF wetlands is presumably restrained by readily available carbon, and while a C:N ratio of ~2.5 may enable complete denitrification using methanol as carbon source (Halling-Sørensen & Jørgensen 1993), a C:N ratio of at least 5 is recommended under field conditions to account for carbon lost in other processes (Baker 1998, Hang et al. 2016. In comparison, the COD DISS : NO 3 -N ratio in the present study (measured in the water column) averaged 1.4 ± 0.4 across all sampling stations and sampling days, indicating that denitrification was limited by readily available carbon. Previous studies have demonstrated that, with little effort, fish sludge can be used as an endogenous carbon source to improve denitrification in MTFs (Suhr et al. 2013, Letelier-Gordo et al. 2015, von Ahnen et al. 2020. To optimize the utilization of sludge for denitrification in wetlands, freshly ob tained sludge ideally hydrolysed for about 3 d to maximize the formation of volatile fatty acids is preferable compared to non-optimized sludge basin overflows. Installation of woodchip bioreactors in an MTF wetland is an additional way to supplement carbon and achieve year-round denitrification (e.g. Christianson & Schipper 2016, Hang et al. 2016, Lepine et al. 2018. Woodchip bioreactors should preferably receive effluents with a low C:N ratio to reduce risks of clogging, making woodchip bioreactors especially suited in situations where there is a shortage of hydrolysed sludge.
The present study revealed that there was hardly any removal of P DISS or TAN in the wetland averaged over a year. In such a case, and if an MTF cannot meet discharge limits, the amounts of P DISS and TAN entering the wetland should be reduced. In some cases, that might be achieved relatively simply by emptying the overflowing sludge basin more frequently. With respect to phosphorus specifically, in flow concentrations might be reduced by adding some kind of chemical precipitant or binding/adhesion agent, or establishing a biological phosphate removal unit where applicable, taking system-specific flows and concentrations into account. Furthermore, phosphorus entering an MTF wetland originates from fish feed in the first place, and previous studies have shown that dietary phosphorus levels can be optimized (re duced) to minimize the excretion of phosphorus without harming the fish (Dalsgaard et al. 2009, Dalsgaard & Pedersen 2011, essentially solving subsequent removal issues. Finally, as all type III MTFs operate with nitrifying biofilters, TAN discharges can be reduced by optimizing biofilter size and performance.