Urban watersheds have high potential for N retention (Bettez et al. 2015), but the processes driving N retention are not well constrained (Groffman et al. 2004; Hobbie et al. 2017). Permanently wet stormwater ponds (SWPs) are engineered ecosystems commonly constructed in urban watersheds (Collins et al. 2010; Sinclair et al. 2020) to provide various ecosystem services to society (Barot et al. 2017; Keeler et al. 2019), and often discharge to naturally occurring waterways. Although SWPs can provide a range of ecosystem services, they are primarily designed to mitigate downstream flooding by capturing and storing surface runoff generated in impervious watersheds and secondarily to protect downstream water quality by settling or removing pollutants (i.e., sediments, nutrients). In other words, SWPs are engineered to some of the ecological functioning naturally provided by riparian zones (Bettez and Groffman 2012), which act as a buffer through which runoff flows through before entering waterways. Despite their prevalence and importance to society, the internal biogeochemical processes occurring within SWPs are understudied. In particular, the processes controlling the fate of N within these ecosystems are not well understood (Gold et al. 2019, but see Gold et al. 2017a, b; Blaszczak et al. 2018; Hohman et al. 2021). Given their high spatial distribution and the assumption that SWPs retain and/or remove nutrients, it is important to characterize how SWPs contribute to N-cycling within urban watersheds.

Aspects of SWP morphology, such as a relatively small area (typically < 0.01 km2) and high sediment area to water volume ratio, suggest they are potential hotspots for N and phosphorus (P) cycling (Cheng and Basu 2017). Lentic ecosystems retain P via physical (sedimentation), chemical (precipitation, substrate adsorption), and biological (algal, microbial, and macrophytic assimilation) processes (Wetzel 2001). Similar patterns and processes are found in SWPs, leading to total P removal efficiencies of 60–90% in Florida SWPs dependent on water residence time (Harper and Baker 2007). The lack of dissimilatory transformations of P pathways and the importance of sorption and sedimentation for P dynamics may result in high storage P (i.e., sediment–water column fluxes). In contrast to P, the N cycle involves a multitude of biogeochemical processes, including assimilatory and dissimilatory transformations, complicating our ability to quantify the fate of N inputs in SWPs. Mass-balance studies report a wide range of N removal efficiencies using inflow vs. outflow responses. For example, SWPs in Florida exhibit low total N removal efficiencies of no more than 45% (Harper and Baker 2007), and ponds can increase the effluent concentration of nitrate (NO3), ammonium (NH4+), and/or organic N compared to inflows (Fairchild and Velinsky 2006; Reisinger et al. 2016a; Gold et al. 2017a). In addition to this variability in N removal, there is a lack of understanding of the biogeochemical mechanisms occurring within SWPs that alter the fate of N (Gold et al. 2019, but see Gold et al. 2017a, b; Blaszczak et al. 2018; Hohman et al. 2021).

Quantifying N cycling in SWPs helps inform considerations of their effectiveness when balancing intended services (e.g., floodwater prevention and water quality protection) vs. potential disservices (e.g., N production or increased N export). A mechanistic view of N cycling processes occurring within SWPs is particularly needed given the importance of SWPs as a tool for mitigating urban N pollution (Collins et al. 2010). An improved understanding of SWP functioning is particularly important in Florida (FL, USA), where there are an estimated 76,000 urban SWPs covering > 600 km2 of land area (Sinclair et al. 2020). Typically, SWPs receive a complex range of particulate and dissolved N from urban watersheds (Jani et al. 2020; Lusk et al. 2023). These N inputs are assumed to be removed from the water column via settling and subsequent storage in sediments, biotic assimilation, or denitrification. Denitrification represents a permanent removal of N from an ecosystem (N sink) driven by the dissimilatory reduction of NO3 to dinitrogen gas (N2; Reddy and DeLaune 2008). In contrast to denitrification, N-fixation, which is conducted by specialized bacteria (“diazotrophs”), represents an input of N into an ecosystem (N source) by fixing gaseous N2 into ammonia (NH3) that is incorporated into biomass or leaked into the environment (Howarth et al. 1988; Adam et al. 2016; Marcarelli et al. 2022).

SWPs often feature ideal conditions for denitrification (high N, low dissolved oxygen, a labile carbon source), leading to an assumption of high denitrification activity within SWPs. However, most studies focused on N-cycling in SWPs have used inflow-outflow mass-balance comparisons rather than quantifying mechanisms and rates of N transformations (Gold et al. 2019), especially at the ecosystem scale. Typically, denitrification is expected to occur within sediments but can also occur in the water column on anaerobic particle microsites (Ploug et al. 1997; Liu et al. 2013; Reisinger et al. 2016b). Under ambient conditions, however, SWP sediments have exhibited low denitrification rates, or even net N-fixation (Bezold 2021; Hohman et al. 2021; Gold et al. 2021). In contrast to ambient conditions, external NO3 additions (simulating stormwater runoff) have been shown to stimulate either denitrification (Hohman et al. 2021; Gold et al. 2021) or N-fixation (Gold et al. 2017b), although the mechanisms responsible for the net N transformation response is unclear. Despite the expectation of N-fixation typically occurring under N-limited conditions, SWP sediments have been previously shown to exhibit net N-fixation under ambient and elevated NO3 conditions (Gold et al. 2017b), and it has been shown that denitrification and N-fixation co-occur (Halm et al. 2009; Varga et al. 2023). The lack of dissimilatory transformations and permanent removal of P in SWPs may favor N-fixing conditions. For example, high sediment P retention leads to legacy P which can be released back to the water column under anoxic conditions, potentially decreasing N:P ratios regardless of N availability, as has been shown for agriculturally impacted P-rich lake sediments (Nifong et al. 2022), thus promoting N fixation.

SWPs are expected to provide similar functions as natural systems (Bettez and Groffman 2012), but comparisons with natural ponds are lacking. Differences in functionality between constructed and natural ecosystems are expected to relate to various physical, chemical, and biological characteristics such as the age of the system, watershed characteristics, hydrology, morphology, and food web dynamics. For example, a majority (85%, n = 248) of Midwestern lakes exhibited net denitrification, implying that lentic N cycling is reliant upon terrestrial N inputs (Loeks and Cotner 2020). The high degree of N removal in freshwater lakes via denitrification (Loeks and Cotner 2020; Zhang et al. 2022) may point to the high rates of respiration due to terrestrial C inputs, as denitrification is a dissimilatory catabolic process or that anthropogenic inputs of N saturate autotrophic N demand, leading to excess N available for denitrification. Furthermore, the seasonality and magnitude of terrestrial inputs into SWPs may differ from natural water bodies due to altered urban hydrology. Whether SWPs exhibit similar net denitrification at the whole-pond level in response to terrestrial inputs is unknown and is likely influenced by a range of factors. Further, urban ponds are expected to be more productive in their phytoplankton communities do to elevated nutrient inputs (Grogan et al. 2023), potentially altering where in the water column N transformations are occurring (e.g., water column versus sediments).

The objective of this study was to compare net N removal (denitrification) or production (N-fixation) dynamics between SWPs and natural ponds and lakes, and whether those dynamics differed among seasons. We predicted that (1) SWPs are more likely to be undersaturated with N2 causing influxes (net N-fixation) rather than effluxes (net denitrification) of N2-N gas due to high primary production leading to N limitation. Similarly, we predicted that (2) SWPs are more likely to exhibit net N-fixation compared to naturally occurring counterparts due to higher primary production and subsequently lower bioavailable N. Finally, we predicted that (3a) net denitrification would be more prevalent in the wet season due to more frequent inputs of C and N resources via stormwater runoff and (3b) N2 supersaturation would be more prevalent in hypolimnetic water column layers as denitrification often plays a larger role in sediments than surface water layers, although mixing dynamics may complicate this pattern. This study contributes to a growing understanding of the internal biogeochemical dynamics of SWPs. To our knowledge, this is the first study to measure whole-ecosystem N2 gas balances and fluxes in engineered aquatic ecosystems. This approach provides useful information on what N processes are occurring in SWPs, which can enhance our fundamental understanding of lentic N dynamics while also providing mechanisms to enhance the N removal benefits provided by SWPs.


Site description

We conducted this study in north-central and southwest Florida, a region with a subtropical climate that experiences distinct wet (June–October) and dry (November–May) periods with respect to rainfall. To compare N cycling between engineered and naturally occurring pond ecosystems, we sampled fifteen constructed SWPs in southwestern Florida and six naturally occurring ponds and small lakes in north-central Florida. Both regions have a P rich geology, with the southwestern FL sites located on the Peace River Group of the Hawthorn Formation (Scott 1988) and the P-rich Hawthorn formation also underlies much of north-central Florida, including the natural ponds (Scott 1983). We define ponds following Richardson et al. (2022) as those < 0.05 km2 in area (although some SWPs in this study are deeper than the cutoff for depth, < 5m), and small lakes as those above this cutoff (see Table 1). We sampled all sites at the end of Florida’s dry season (May 2021) and a subset of these sites (five SWPs and five natural) in the peak wet season (August 2021). All SWPs were located in a large, master-planned residential community (Lakewood Ranch, Bradenton, FL) that contains hundreds of SWPs. We selected easily accessible stormwater ponds with a range of littoral vegetation coverage (0 to ~ 35% areal cover) and no artificial aeration systems (i.e., fountains, bubblers). Natural sites were located on a 9500 + acre undeveloped sandhill landscape preserve at the UF/IFAS Ordway-Swisher Biological Station (OSBS; Melrose, FL). The high-relief region features a low gradient dark (or humic) water system of interconnected water bodies, and a clear water system that includes isolated and rainfed water bodies (OSBS 2023). Of the six natural sites, three were clear water and three were dark, or humic water sites (hereafter referred to as natural-clear and natural-humic). Natural-clear and natural-humic sites were both included to provide the two natural conditions of lotic ecosystems in this region for comparison to SWPs. Bradenton, FL and Melrose, FL receive an average of 1430 mm and 1390 mm of rainfall per year, respectively, based on 1991–2020 US Climate Normals (Palecki et al. 2021). We recognize that the ponds selected in this study are in different areas of Florida but due to substantial urbanization throughout Florida and particularly within the Tampa Bay area (which includes the SWPs sampled here; see Sinclair et al. 2020), agricultural activities in Southwest Florida, and karst geology limiting surface water accumulation in much of the state, OSBS provides a rare opportunity to find roughly similar sized and permanently wet water bodies within an undisturbed watershed.

Table 1 Site characteristics and water quality parameters measured from surface water samples during both wet and dry seasons

Site descriptions including coordinates, physical, chemical, and morphometric characteristics can be found in Table 1. Dry season sampling occurred over eight days between 17 and 27 May 2021. During that time Bradenton (SWP sites) had not received rainfall since mid-April, and Melrose (natural sites) received 8.4 mm from two events during the first two weeks of May. The average atmospheric temperatures for Bradenton and Melrose during the month of May were 25.8 and 23.6 °C, respectively. Wet season sampling occurred over five days between 9 and 16 August 2021. In the month prior to sampling, Bradenton received 232 mm rainfall and Melrose received 178 mm. Average atmospheric temperatures in the wet season for Bradenton and Melrose during the month of August were 27.9 and 27.0 °C, respectively. Natural-humic sites supported moderate to heavy areal coverage of American white waterlily (Nymphea odorata), a floating attached macrophyte. Natural clear sites supported tall emergent grass in shallow littoral areas. Morphology and vegetation within and around SWPs was highly variable, ranging from no littoral shelves, littoral shelves with bare sand and no vegetation, littoral shelves and sporadic emergent vegetation and little floating attached vegetation. Submerged macrophytes were not assessed.

Sample collection

At a central point of each pond or lake, we characterized thermal profiles using 0.2m increment temperature measurements taken with a Pro-DSS multiparameter water quality meter (YSI, Yellow Springs, OH). We collected triplicate water samples from each of three depths in the water column (surface, middle, bottom), leading to nine total water samples from each pond or lake. Sample replicates from a single layer are averaged to produce one observation per layer per site (three total observations). Sampling depths were selected based on thermal profiles of the water column. If the water column appeared stratified based on preliminary profiles produced in the field, we collected samples from the epi-, meta-, and hypolimnion, taking epilimnetic and hypolimnetic samples from central depths for each layer, respectively. If the water column was fully mixed (similar temperature throughout), we collected samples from the near-surface (within 0.4m of the surface) as well as middle and bottom layer samples that were evenly distributed along the depth profile. Water samples were collected at depth using a 2.2 L Van Dorn sampler while shallow surface samples were collected directly from the water within 0.4 m of the surface. Within each season, each site was sampled once between the hours of 10:00 and 15:00. Using the Pro-DSS, we measured temperature, dissolved oxygen (DO) concentration and saturation, specific conductivity, pH, and turbidity at the same depth and time that we collected water samples. At each depth, water samples were collected to be analyzed for N2 gas, inorganic nutrient ions (NOx, NH4+, PO43-), dissolved organic carbon (DOC), and total dissolved nitrogen (TDN). We estimated phytoplankton biomass as chlorophyll-a (chl-a), with samples collected in triplicate from the surface water of each site only, and we measured Secchi depth to characterize water clarity. We also measured atmospheric temperature, barometric pressure, and relative humidity with an SD200 datalogger during each sampling event (Extech, Nashua, NH, USA).

Water chemistry and algal biomass

Water chemistry samples for NOx, NH4+, PO43−, DOC, TDN were filtered through a 0.22 μm polyethylene sulfonate (PES) membrane filter (MilliporeSigma, Burlington, MA, USA) in the field into acid-washed sample containers (Reed et al. 2023). Inorganic nutrient samples were filtered into 20 mL scintillation vials, whereas DOC and TDN samples were filtered into 60 mL amber HDPE Nalgene bottles. We also collected 1L of water from each layer to be analyzed for chl-a. After returning to the lab, up to 1L of site water was filtered for chl-a onto a 0.7 μm Whatman GF/F filters (Cytiva Life Sciences, Marlborough, MA, USA), which were subsequently wrapped in foil. Water chemistry and chl-a samples were stored frozen at – 20 °C until analysis.

Nutrient samples were thawed to room temperature and analyzed colorimetrically with a Seal AQ400 discrete nutrient analyzer (Seal Analytical Inc., Mequon, WI, USA) for NH4+ using the phenol-hypochlorite method (Solórzano 1969), for NO2 + NO3 (hereafter referred to as NOx) using the cadmium reduction method (APHA 1995), and for PO43− using the ascorbic acid method (Murphy and Riley 1962). DOC and TDN samples were analyzed on a Shimadzu TOC-L total organic carbon analyzer coupled with a TNM-L total nitrogen unit (Shimadzu Corp., Kyoto, Japan). DOC was analyzed as non-purgeable organic carbon with samples acidified with 2% HCl to purge carbonates. Detection limits were 0.006 mg/L for NOx, NH4+, and PO43-, 0.005 mg L−1 for TDN, and 0.004 mg L−1 DOC.

We quantified chlorophyll- a phytopigment as a surrogate for phytoplankton biomass. Chlorophyll a was extracted from filters using a cold-methanol extraction and acidification approach (Wetzel and Likens 1991) and analyzed for absorbance at 665 and 750nm on a Horiba Aqualog (Horiba Ltd., Kyoto, Japan). Chl-a concentrations were calculated using standard calculations (Steinman et al. 2017) modified for methanol extraction.

Dissolved gases

We collected water samples for N2, O2, and argon (Ar) gases to characterize ecosystem-scale N2 dynamics (using N2:Ar ratios) and to understand groundwater inputs of gases to ponds (using O2:Ar ratios), which have the potential to influence how we interpret N2:Ar ratios. Water samples from each water column layer were transferred from the Van Dorn bottle via 50mL syringes into 12 mL exetainer vials (Labco Limited, Lampeter, United Kingdom). To prevent atmospheric contamination, Tygon tubing attached to the 50 mL syringe was inserted to the bottom of the Exetainer vial and sample water was pushed into the exetainer, allowing the exetainer to overflow 3 ×, leaving a positive meniscus. We added 100 μL 7 M ZnCl2 to exetainers to preserve samples to cease biological activity, preventing any further changes to N2 and O2 concentrations. After sealing vials we verified that no air bubbles were present in exetainers. We stored samples submerged upside down in water at 4 °C until analysis, within 6 months of collection.

We used a quadrupole membrane inlet mass spectrometer (MIMS; Bay Instruments, Maryland, USA) to quantify dissolved N2, O2, and Ar concentrations from water samples and used the N2:Ar and O2:Ar ratio approach to estimate field concentrations of dissolved N2 and O2 (Kana et al. 1994). The ratio approach uses MIMS to quantify N2 (or O2) concentrations with high precision by using the ratio of N2 (or O2) measured by the mass spectrometer relative to a conservative and biologically inert gas, Ar (Kana et al. 1994). The observed field N2 and O2 concentrations are estimated by multiplying the MIMS detection signal ratio of N2:Ar (or O2:Ar) by the expected solubility of Ar given the relevant pressure, temperature, and salinity conditions when samples were collected (Kana et al. 1994). Ar solubilities were calculated using Henry’s law and Ar-specific solubility coefficients (Hamme and Emerson 2004).

After calculating N2 concentrations, we calculated the N2 saturation ratio, which is the concentration of N2 gas measured in field samples divided by the equilibrium N2 gas concentration given the pressure, temperature, and salinity conditions when samples were collected. The equilibrium concentration of N2 was calculated using Henry’s law and N2-specific solubility coefficients (Hamme and Emerson 2004). The N2 saturation ratio represents the degree to which N2 gas dissolved in water deviates from equilibrium solubility, which is expected to be a result of biological activities that produce or consume N2 gas (Loeks and Cotner 2020; Taylor et al. 2023). N2 saturation ratios greater than 1 indicate that N2 gas has been produced within the ecosystem, suggesting net denitrification (and/or anammox). In contrast, N2 saturation ratios less than 1 indicate that N2 gas is consumed within the ecosystem, suggesting net N-fixation. Net denitrification (N2 saturation > 1) and net N-fixation (N2 saturation < 1) represent outputs and inputs of N at the ecosystem scale, respectively. We calculated MIMS N2:Ar and N2 saturation ratios following the approach presented by Loeks and Cotner (2020) and Taylor et al. (2023). For gas samples collected below the water column’s mixing depth, we included a pressure correction based on sample depth in the solubility equations of N2, Ar, and O2 (Colt 2012) that is dependent on the depth of the sample below the water’s surface. We calculated the depth of the mixing layer, or the depth of the epilimnion, from thermal profiles measured during sample collection using the thermo.depth() function of the RlakeAnalyzer package (Winslow et al. 2019). We identified stratification within a pond based on a water density gradient change of > 0.1 kg m−3 m−1 using the thermo.depth() function in RlakeAnalyzer (Winslow et al. 2019).

Prior N2 dissolved gas studies have assumed dissolved gas dynamics are driven by internal production and air–water gas exchange. However, groundwater inputs (and gases dissolved in the groundwater) have the potential to complicate these patterns, particularly in hypolimnetic waters. To understand the potential impact that groundwater inputs of gases has on our MIMS-estimated dissolved gas concentrations measured at depth, we compared the O2 concentrations that were produced using O2:Ar ratios from the MIMS to in situ O2 measured with the YSI optical DO sensor. YSI O2 measurements were recorded at the same depth and time that we took water samples. We divided MIMS DO by YSI DO to assess the magnitude of their difference. Any deviation from 1 would suggest that O2 estimated via MIMS relative to Ar saturation leads to under estimation (if groundwater is Ar-enriched) or over estimation (if groundwater is Ar-depleted). Comparing O2 estimated using MIMS and other in-situ methods has previously been used to assess the effectiveness of using Ar as a conservative tracer to compare MIMS and sensor-based measurements (Reisinger et al. 2016b).

Air–water N2 gas flux

We calculated the flux of N2 between surface water and the atmosphere using surface water samples and Eq. 1 (Wanninkhof 1992):

$$F = k*\left( {C_{ measured} - C_{ expected} } \right)*0.1$$

where F is the flux of N2 (µg N2-N m−2 h−1), k is the gas transfer velocity (cm h−1; Cole and Caraco 1998), Cmeasured is the measured concentration of N2 dissolved in the surface water, Cexpected is the equilibrium solubility of N2 (µg L−1; Hamme and Emerson 2004), and 0.1 is a unit conversion factor. We used the gas transfer velocity equation by Cole and Caraco (1998) over other options based on a recent comparison of different approaches for calculating gas transfer velocity to estimate N flux in ponds (Zhang et al. 2022). To calculate k, we coupled our wind speed measurements from the field with the normalized wind speed function, wind.scale.base(), from the LakeMetabolizer package (Winslow et al. 2016), and the temperature-adjusted Schmidt number for N2 (Wanninkhof 1992).


We conducted all statistical analyses in the statistical software R (version 4.2.3, R Core Team 2023). Using the suite of water quality variables measured in the field, we characterized the water quality fingerprint of each site using a principal component analysis (PCA) conducted with the prcomp() function of the base R stats package. We visualized the contribution of individual water quality variables to each PCA axis using the fviz_contrib() function of the factoextra package (Kassambara and Mundt 2020). To test for differences in environmental characteristics (morphology, surface layer water chemistry and algal biomass) between the dry and wet season, we use analysis of variance (ANOVA; one for each parameter included in Table 1 for each site type) with season as a factor and the characteristic variable as the response.

We developed two generalized linear mixed-effects models (GLMM) to understand factors affecting N2 saturation ratios. First, we tested how season (wet, dry), site type (urban, natural-humic, natural-clear), and water column layer (surface, middle, bottom) influenced N2 saturation ratios (GLMM-A). Second, we tested how environmental variables influenced N2 saturation ratios (GLMM-B). For GLMM-B, we used the water quality PCA to identify environmental predictor variables relevant to denitrification and N-fixation. For both models, pond site was a random effect, and a gamma distribution was used with a log link function. We included season as an interaction for each explanatory variable in GLMM-B. The GLMMs were produced using the glmmTMB() function of the glmmTMB package (Brooks et al. 2022). We diagnosed model fit using qq-plots and simulated versus observed residual plots from the DHARMa package to assess deviations of the response variable to the selected distribution and to infer if a model is correctly specified (Hartig 2022). Marginal (variance explained by fixed effects alone) and conditional (variance explained by the entire model including fixed and random effects) R2 values for GLMM’s were calculated using the r.squared GLMM() function of the MuMIn package (Barton 2022). Estimated means and various Tukey pairwise comparisons were conducted using the emmeans function and package (Lenth 2022). With N2 flux data we do not test for effects of experimental design or environmental variables as these values are derived from surface layer N2 saturation, and thus we assume relationships remain between fluxes and explanatory variables in GLMM-A and GLMM-B. For all statistical analyses, we interpret p-values < 0.05 as significant effects and we report all p-values < 0.10 to include additional detail on marginal results.


Environmental characteristics

Study sites varied in physical, chemical, and biological parameters (Table 1). SWP area varied between 0.003 to 0.015 km2, natural-humic sites from 0.008 to 0.012 km2, and natural-clear sites from 0.037 to 0.084 km2. Water column depths at the center sampling location were variable among SWPs, ranging from 0.8 to 6.8 m deep, whereas natural-humic and natural-clear ponds ranged from 1.3 to 2.6 m and 3.3 to 7.3 m, respectively (Table 1). Water column depths were not significantly different between seasons (ANOVA; df = 1.18, p = 0.73).

All three pond types exhibited distinct surface water chemistry, as seen in the PCA, with some overlap in characteristics between SWPs and natural-clear ponds (Fig. 1). The first principal component axis explained 30.6% of the variation in water chemistry data and was primarily represented by N availability with the top four variables (10 to 23% contributions) including DIN, NH4+, DOC:TDN, and NOx (Fig. S1). The second principal component axis explained 26.7% of the variation in water chemistry data with the top four variables (14 to 23% contributions) including %DO, DOC, TDN, and pH (Fig. S1). The only seasonal difference in surface water quality was found for pH (ANOVA; df = 1.18, p < 0.01), which was higher in SWPs in the dry season (7.58 ± 0.12) than the wet season (6.77 ± 0.19; Tukey’s HSD post-hoc comparison; p < 0.05). Water quality parameters of SWPs tended to be intermediate between natural-clear and natural-humic ponds (Fig. 1). However, SWPs had the highest means and most variable values for conductivity (683 ± 48 µS cm−1), chl-a (122 ± 20.3 µg L−1), and pH.

Fig. 1
figure 1

Principal component analysis (PCA) revealed differences in surface water chemsitry collected among the three pond types. Pond types are identified by different colors and shapes: natural-clear (green circles and ellipsoid), natural-humic (yellow triangles and ellipsoid), and urban (grey squares and ellipsoid). Water chemistry parameters are identified with red text, and the direction and length of arrows reveal how water quality parameters influenced the ordination of sites along the two primary axes

Thermal profiles of pond water columns (0.2 m increments; Figs. S2, S3) indicated that ten sites were stratified (water density gradient change of > 0.1 kg m−3 m−1 using the thermo.depth() function; Winslow et al. 2019) at some point above the bottom layer sampling depth during dry season sampling (five urban, five natural). Five sites (all natural ponds) were stratified during wet season sampling. DO profiles (0.2 m increments; Figs. S2, S3) revealed that seven of fifteen urban ponds and all natural-humic ponds fell below 5% DO at the bottom of the water column during dry season sampling. In the wet season, the bottom of the water column in four of five urban ponds, one natural-clear pond, and two natural-humic ponds fell below 5% DO. While the predicted thermocline for a given site did not always agree with the plotted thermocline (depth of rapid temperature change; Figs. S2, S3), shifting thermocline depths (observed vs. modeled) did not affect how we calculate gas saturation at depth (i.e., whether it depth was incorporated into the saturation equation).

Comparing O2 saturation estimated via MIMS (using O2:Ar) to that of in situ YSI measurements (using an optical DO sensor) indicate site and depth specific differences. When O2 concentrations measured using both methods are plotted against each other (Fig. S4a), data fall closely to the 1:1 line for natural-clear sites (YSI and MIMS O2 estimates are the same). However, these data shift slightly above the line (MIMS O2 values are higher than the same YSI O2 estimate) for natural-humic sites and some SWPs (Fig. S4a). Box plots of the ratio of MIMS O2 to YSI O2 show that ratios increased with depth for natural-humic sites and in some SWPs (Fig. S4b). The overall mean ratio of MIMS O2 to YSI O2 in natural-clear ponds was 1.14. Surface layer MIMS:YSI O2 ratios were 1.47 and 1.06 in natural-humic and SWPs, respectively, while bottom layer means were 4.43 and 1.81, respectively (Fig. S4b). An elevation of MIMS O2 relative to YSI O2 would suggest that the MIMS approach is overestimating O2, potentially due to groundwater Ar differing from expected equilibrium Ar due to different water sources.

N2 saturation ratios

Of 93 observations collected from 21 sites spanning two seasons (all sites sampled during the dry season, a subset of ten sites sampled in the wet season) and three depths within each site, N2 saturation ratios ranged from 0.89 to 1.11 (Fig. 2a). Across all sites and seasons, 61% of observations exhibited supersaturated conditions (N2 ratios > 1; net denitrifying) and 39% exhibited undersaturated conditions (N2 ratios < 1; net N2-fixing; Fig. 2a). In natural ponds, 82% of the 33 observations were supersaturated with N2 (Fig. 2b). In contrast, 50% of the 60 observations from SWPs were N2 supersaturated and 50% under saturated (Fig. 2b). All samples exhibiting undersaturated N2 ratios in natural ponds (6 of 33 total samples) occurred in the wet season and were equally distributed between clear and humic sites (three from clear and three from humic waters; Fig. 3b). Focusing on surface layer samples only, SWPs still exhibited 50% super-saturated and 50% under saturated N2 ratios (n = 20; Fig. 3), whereas 91% of surface samples from natural sites were supersaturated and 9% were under saturated (n = 11; Fig. 3).

Fig. 2
figure 2

Observations of N2 saturation ratios reveal that both net N-fixation (N2 saturation ratio < 1) and net denitrification (N2 saturation ratio > 1) were common across all samples (a). When separating naturally occurring ponds (green) from SWPs (gray), natural ponds were more likely to exhibit N2 super-saturation than SWPs (b). The red line indicates the equilibrium N2 saturation based on Henry’s law

Fig. 3
figure 3

Saturation ratios of N2 exhibited differing patterns in layers of each site type and season. Individual points represent N2 saturation ratios (y-axis) measured at different sites (x-axis) during a dry (a) and wet (b) season. Larger points circled in black indicate the mean of points and black standard error bars are shown. The red lines indicate the equilibrium N2 saturation based on Henry’s law. Pond sample sizes (n) are included at the bottom of the plots

The fixed effects of site type, water column layer, and season and their interactions (GLMM-A) explained 52% of the variation (marginal R2) in N2 saturation ratios, and the whole model (including site as a random effect) accounted for 64% of the variation (conditional R2). Site type (GLMM-A; df = 2.73, p < 0.01) and season (GLMM-A; df = 1.73, p = 0.03), but not water column layer (GLMM-A; df = 2.73, p > 0.10) affected N2 saturation ratios. These individual fixed effects were complicated by interactions among factors, however, as there were significant two-way season × site type (GLMM-A; df = 2.73, p < 0.01) and three-way water column layer × site type × season (GLMM-A; df = 4.73, p < 0.01) interactions.

The differences among site types and seasons were driven by the dry season where natural-humic ponds had higher N2 saturation ratios than SWPs (emmeans pairwise comparison; p < 0.01) and natural-clear ponds (emmeans pairwise comparison; p = 0.02; Fig. 3a). The season × site type interaction was driven by natural humic sites having higher N2 saturation ratios during the dry season (emmeans pairwise comparison; p < 0.01), whereas N2 saturation ratios in natural-clear and urban sites did not differ seasonally (emmeans pairwise comparisons; p > 0.1 for both comparisons). The three-way interaction between water column layer, site type, and season was driven by seasonal shifts in the N2 saturation of middle (natural-humic sites) and bottom (natural-humic and clear sites) depths (emmeans pairwise comparisons; Fig. 3). While there was no significant difference between layers in SWPs, N2 saturation ratios were lower in the bottom layer than in surface layers for 8 of 15 SWPs (Fig. S5).

Based on the PCA (Fig. 1) we selected DO (indicator of redox conditions; mg L−1), TDN (indicator of N availability; mg L−1), and DOC (indicator of C availability; mg L−1) to be included as environmental fixed effects in GLMM-B for their association with potential drivers of denitrification and N-fixation. Environmental variables also influenced N2 saturation (GLMM-B), with fixed effects in the GLMM-B model explaining 41% of the variation in N2 saturation ratios (marginal R2), and the full model (including site as a random effect) explaining 50% of the total variation in N2 saturation ratios (conditional R2). Dissolved oxygen content (df = 1.83, p < 0.01) and TDN (df = 1.83, p = 0.05) were negatively related to N2 saturation ratios, while there was a marginal but non-significant effect of DOC (df = 1.83, p = 0.08). The effect of DO on N2 saturation ratios differed between seasons (DO × season interaction; df = 1.83, p < 0.01; Fig. S6). In the dry season, DO was significantly and negatively correlated to N2 saturation ratios (emtrends p < 0.01, df = 83) and had a steeper slope than that of the wet season (emtrends p = 0.051, df = 83; Figure S6), indicating a stronger effect of DO in the dry season. The explanatory power of GLMM-B was similar to but lower than GLMM-A.

Air–water N2 gas fluxes

The flux of N2 gas between surface waters and the atmosphere ranged from -500 to 433 μg N m−2 h−1 (Fig. 4) with both the minimum and maximum fluxes occurring in SWPs. The range in fluxes were lower in natural-humic (41 to 407 μg N2-N m−2 h−1) and natural-clear (− 68 to 74 μg N2-N m−2 h−1) sites. Similar to N2 saturation ratio patterns, the average natural-humic flux (202 ± 64.9 μg N2-N m2 h−1) was higher than natural-clear (39.7 ± 22.7 μg N2-N m2 h−1) and SWPs (− 1.67 ± 50.3 μg N2-N m2 h−1). Further, mean N2 flux was higher in the dry season than the wet season across all site types. Dry season mean ± SE fluxes were 52.7 ± 13.1, 293 ± 60.3, and 8.16 ± 65.6 μg N m−2 h−1 in natural-clear, natural-humic, and urban sites, respectively. Natural-clear and natural-humic pond dry season fluxes were 6 and 36 times higher than in SWPs respectively. Wet season means were in the same order as dry season, but lower for all site types, with 26.7 ± 47.3, 66.8 ± 26.1, and -31.2 ± 50.9 μg N2 m−2 h−1 in natural-clear, natural-humic, and SWPs, respectively (Fig. 4).

Fig. 4
figure 4

Air–water N2 fluxes were more variable in urban ponds than natural clear or humic ponds. Points represent air–water N2-N flux (µg N2-N m−2 h−1; y-axis) estimates from individual ponds for each site type (x-axis) during a dry (orange points) and wet (blue points) season. The red line indicates equilibrium fluxes and horizontal orange or blue bars represent means. Fluxes were calculating using surface layer samples only. Samples sizes are the same as shown in Fig. 3


N2 saturation ratios and fluxes

Stormwater ponds are overlooked nodes in urban drainage networks that connect developed watersheds to downstream ecosystems. These engineered ecosystems are often assumed to be sinks for N via denitrification (Gold et al. 2019) despite limited empirical support. In this study, we found that half of the samples from SWPs were supersaturated with N2, indicating that SWPs can be N sinks via net denitrifying activity (removal of bioreactive N from the ecosystem). However, the remaining 50% indicated net N-fixing activity, a biological introduction of N into the ecosystem (Fig. 2b). This result was consistent whether including all sampled depths or when only focusing on surface layer N2 saturation, which is less influenced by groundwater inputs (Fig. S4). Although this study focused solely on N2 saturation ratios, this MIMS-based ratio approach has been shown to successfully infer net denitrification or net N-fixation in pond mesocosms when compared to established approaches to measuring these N cycling processes (Taylor et al. 2023). The consistency of the N2 saturation approach with traditional N-cycling process rate measurement methods supports our conclusions that N2 undersaturation and supersaturation indicate ecosystem-scale net N-fixation and net denitrification, respectfully. However, there is a potential for diel variation in N2 dynamics (Levine and Lewis 1984; Reisinger et al. 2016b; Knee et al. 2018; Nifong et al. 2020), and diel sampling would provide valuable insight into the balance of denitrification and N-fixation in SWPs and other ecosystems.

Unlike SWPs, a majority of samples from undisturbed and naturally occurring ponds and small lakes were net denitrifying (82%), which is comparable to the 86% of samples collected at multiple depths from temperate midwestern lakes found to be net denitrifying (Loeks and Cotner 2020). Our results indicate that SWPs are at times more likely to support N-addition via N-fixation in excess of N-removal via denitrification, an unintended disservice compared to the functions SWPs are expected to provide to society. This result of net N-fixation within SWPs is also supported by previous research showing net N-fixation in SWP sediments using core incubation approaches (Gold et al. 2017a, b; Hohman et al. 2021).

With this study, we compared net N cycling in SWPs to natural ponds that were either humic or clear water. All three site types exhibited differences in water chemistry (Fig. 1), hydrology, and watershed characteristics, which likely played a large role in observed differences in N2 patterns. Differences in water chemistry were largely driven by aspects of N, C, and DO availability, and pH (Figs. 1, S1). Natural-clear lakes represent rain-fed ecosystems that do not receive substantial sediment inputs due to their isolated position. Natural-humic lakes represent shallow groundwater and wetland-fed systems that support floating attached vegetation (Nymphaea odorata in ponds of this study) with DOC-rich and dark colored water. SWPs are primarily fed by water generated in developed landscapes that are built to funnel runoff and associated dissolved and suspended materials directly to SWPs. This engineered hydrology allows SWPs to rapidly accumulate sediment and receive particulate and dissolved materials characteristic of urban vegetation, urban karst (i.e., weathered asphalt, concrete, and other human built infrastructure; Kaushal et al. 2017), and other activities that contribute chemicals to the landscape (e.g., fertilizers, detergents, pesticides), or directly into ponds (e.g., algaecides, herbicides, dyes). While natural sites were acidic, urban sites were neutral to slightly alkaline (Table 1), which could be a result of carbonates and basic ion contributions from weathering of urban karst (Kaushal et al. 2017), or higher rates of photosynthesis as suggested by higher chl-a concentrations in SWPs.

Fluxes of N2 between surface waters and the atmosphere in SWPs were nearly zero on average (− 1.67 ± 50.3 μg N2-N m2 h−1) but highly variable when compared to natural ponds (202 ± 64.9 and 39.7 ± 22.7 μg N2-N m−2 h−1 in humic and clear, respectively). The range of fluxes in SWPs indicated that they could emit and remove up to 433 μg N2-N m2 h−1 from a pond, or fix and add (indicated by a negative flux) up to 500 μg N2-N m2 h−1 into a pond. The N2 fluxes from SWPs found in this study were lower than previously found in residential ponds using the same approach (3280 μg N2-N m−2 h−1; Zhang et al. 2022). This prior study also found more consistent evidence of net denitrification (Zhang et al. 2022). In an experimental study testing the effect of N:P ratios on N cycling in limnocorral mesocosms within constructed ponds, the sediment–water column flux of N2 was − 550 μg N2-N m2 h−1 (N uptake by sediments indicating N fixation) in N-limited treatments (low N:P), similar to the minimum recorded air–water flux in SWPs of this study (Taylor et al. 2023). In the same limnocorral study, N fluxes increased with increasing N:P up to 5900 μg N2-N m2 h−1 in N-enriched conditions (high N:P; Taylor et al. 2023). Thus, it is likely that N-fluxes (and N2 saturation ratios) in SWPs are associated with N:P and N availability. Lower positive N fluxes in our SWPs compared to other pond studies may indicate that they either don’t receive enough N to stimulate high denitrification (indicated by low inorganic N concentrations), or that the form of N available is not in a form ideal for denitrification (i.e., organic N), as has been shown previously (Jani et al. 2020; Lusk et al. 2023).

Assessment of groundwater influence on bottom layer gases

Groundwater inputs can contribute water that contains N2, O2, and Ar at different concentrations than equilibrium conditions, depending on the age of the groundwater and the conditions experienced throughout groundwater flow paths (Gardner et al. 2016; Knee et al. 2018). N2:Ar and O2:Ar ratios are based on the assumption that Ar is in equilibrium with the atmosphere, following Henry’s law (Kana et al. 1994). However, groundwater inputs may alter the ratio of N2 and O2 relative to Ar, causing N2:Ar and/or O2:Ar ratios to appear more or less saturated, which can be incorrectly interpreted as being driven by biogeochemical processes (Gardner et al. 2016; Knee et al. 2018). If groundwater inputs contain more enriched Ar than equilibrium, N2 and O2 become underestimated, and Ar-depleted water would overestimate N2 and O2. Additional approaches to account for groundwater contributions, such as using stable isotopes of water (18O, 2H) and radon (222Rn), would provide additional support for the relative importance of groundwater contributions in these ecosystems (Petermann et al. 2018; Knee et al. 2018).

By comparing O2 saturation estimated using MIMS (which uses Ar as a baseline following the same approach as for N2) to in-situ optical DO sensor measurements, we demonstrate that the potential influence of groundwater is site specific and most important for bottom layer samples (Fig. S4). In a majority of samples that deviated from 1, O2 was overestimated, indicating the potential groundwater influence of Ar-depleted water. For N2, this would indicate that N2 could also be overestimated (and potentially result in interpreting a site as net denitrification when it is net N-fixation). The natural-clear sites in this study appeared to have no influence of groundwater inputs, which agrees with the fact that these sites are at a slightly higher elevation and have been described as rainwater fed lakes (OSBS 2023). In contrast, natural-humic sites are low-elevation lakes with a high degree of hydrological connection to the landscape and appeared to have the highest potential groundwater influence based on the difference between MIMS O2 and in-situ O2 estimates (Fig. S4). Natural-humic ponds in the wet season exhibited net N-fixing N2 ratios (N2 undersaturation). While these N2 ratios in the natural-humic site during the wet season may have been overestimated based on the Ar-depleted groundwater results, hypolimnetic waters, in particular, may be even more under-saturated than measured here. Further, we suspect that systems with low groundwater connectivity such as the natural-clear sites in this study allow for reliable hypolimnetic N2:Ar and O2:Ar interpretation. These results suggest that N2:Ar approach used here should be interpreted cautiously in highly groundwater influenced systems.

The occurrence and frequency of groundwater connectivity in SWPs is unclear and likely depends on engineering design. For example, some SWPs are built with plastic liners isolating the pond from groundwater, while others are intentionally designed to interact directly with the water table. We do not have direct information about the lining (or lack thereof) of SWPs in this study but would predict that plastic liners are uncommon in this region based on prior experience (e.g., Goeckner et al. 2022). Sixteen of twenty SWP samples appeared to have groundwater influence in bottom layer samples based on MIMS:YSI DO ratios higher than 5% difference between MIMS and YSI readings (1.05 ratio), although most of these (12) were within 20% of each other (1.06–1.20 MIMS:YSI ratio), and may not result in meaningful differences between MIMS and YSI values. While the O2 comparison allows for an assessment of the likelihood of groundwater inputs and the interpretability of direct N2:Ar data, future research could incorporate additional inert gases to constrain the relative importance of groundwater inputs on N2:Ar ratios (Knee et al. 2018).

Considering drivers and spatial patterns of denitrification and N-fixation

Denitrification is primarily thought to take place in sediments but can occur on suspended particles in the water column of rivers (Liu et al. 2013; Reisinger et al. 2016a, b), and denitrifying bacteria have been consistently found in reservoir water columns (León‐Palmero et al. 2023). Regardless of whether a site was net denitrifying or net N-fixing, N2 saturation increased with depth in eleven of fifteen SWP sites and all six natural sites during the dry season (although sampling depth was not significant in GLMM-A). This pattern suggests that sediment denitrification can increase N2 saturation at the pond scale Fig. S5a). However, this pattern changed in the wet season when all natural sites (n = 5) and some SWPs (3 of 5 sites) had higher N2 saturation at the surface than the bottom of the water column (Fig. S5b). These results suggest surface layer denitrification activity may increase (i.e., by the increased input of particles or sediments that support denitrification microsites), that hypolimnetic/sediment N-fixation was occurring faster than equilibrium could be achieved, or that changes to groundwater gas content and connectivity to ponds led to a decrease in N2:Ar saturation ratios.

Stormwater ponds are typically assumed to be hotspots for denitrification due to carbon rich and anoxic sediments (Gold et al. 2019), and there is an expectation of ample N in the form of NO3 due to stormwater runoff. Denitrification rates are strongly correlated to nitrate and oxygen availability (Cavari and Phelps 1977; Nakajima et al. 1984). Despite an expectation of ample NO3 associated with urban runoff, a majority of the N in stormwater runoff in the study region is in dissolved and particulate organic N form, while a smaller fraction is present as NO3 and NH4+ (Jani et al. 2020; Lusk et al. 2023). In this study, NOx concentrations were often below detection or negligible (< 0.01 mg L−1). Thus, the low effluxes of N2-N and occurrence of net denitrification in only half of SWPs may be a result of insufficient nitrate concentrations, oxygen inhibiting denitrification, or N-fixation consuming N2 produced via denitrification, as has been previously found in sediments of SWPs and eutrophic lakes (Varga et al. 2023; McCarthy et al. 2016). SWPs supported higher phytoplankton biomass in surface waters than natural ponds (chl-a, Table 1). Therefore, it is likely that new inputs of inorganic N are rapidly assimilated into biomass, contributing to denitrification being limited by NO3-N, as has been shown previously in SWPs (Hohman et al. 2021). For this assimilated N to become available for denitrification (or any dissimilatory transformation), it would need to undergo organic matter mineralization followed by nitrification to supply NO3. We found that TDN concentrations were negatively correlated to N2 saturation, contrary to our expectations, as higher concentrations of N typically favor denitrification (Bezold 2021; Hohman et al. 2021). This contrary result may be due to the N existing in an unfavorable form (organic) or some unidentified factor driving both TDN and denitrification.

N-fixation has the potential to reduce N removal efficiency by consuming the N2 that is produced via denitrification or shifting a system to being a net source of N. Prior research has suggested that a portion of N2 produced by denitrification in SWP sediments can be recycled into the system by acting as a source of N2 for N-fixation, as observed by the associated expression of nifH (Varga et al. 2023). For example, N-fixation was found in eutrophic lake sediments, but rates were not sufficient to supersede N2 producing (N sink) processes despite N-fixation consuming 25 to 30% of the N2 produced by denitrification (McCarthy et al. 2016). Thus, even in cases of net denitrification, N-fixation is a biological process that reduces the N-removal efficiency of engineered ecosystems designed to retain nutrients from the runoff.

In surface waters, N-fixation is performed by free-floating diazotrophs, mostly cyanobacteria, or by heterotrophs attached to suspended particle microsites (Ploug et al. 1997), whereas in sediments it is mostly performed by heterotrophic bacteria and/or cyanobacteria if light is available. In SWPs, we found ten cases of net N-fixation in surface waters (seven dry season and three wet season), and ten cases of net denitrification (eight dry season and two wet season). Although SWPs receive ample N from the watershed, N-limitation—which stimulates the need for N-fixation—can persist with high productivity if new N inputs are rapidly assimilated into biomass.

This region of Florida is characterized as having a P-rich geology, and a previous study in a low-density residential catchment (also in southwest FL) found that urban runoff had ample concentrations of P and low TDN:TDP ratios (ranging from 0.3 to 29) that indicated N-limitation and NP co-limited conditions (Yang and Toor 2018). SWPs sufficiently remove P via sedimentation (Harper and Baker 2007) and when sediments undergo anoxic conditions, PO43− can be released from mineral associations into the overlying water (Reddy and DeLaune 2008). In this study PO43− was below detection in SWPs, although that could be due to a variety of reasons including rapid biotic assimilation, abiotic sorption to sediments, or lower than expected inputs from the watershed. Low N:P ratios in SWPs can thus either be a result of low N inputs and/or high sediment P release (contributing to benthic N-fixation). This coupled with negligible NOx and high chl-a (Table 1) supports the idea that SWPs are highly productive ecosystems with high assimilatory demand for N and P.

The potential for N-fixation to support highly productive ecosystems may be particularly important when diazotrophic cyanobacteria are present. Compared to naturally occurring ponds and lakes, SWPs in this study had higher algal biomass, indicated by chl-a concentrations, and had low to non-detectable concentrations of NH4+ and NOx (Table 1). Large, productive lakes have also reported high pelagic N-fixation rates correlated with algal biomass and light availability (Mugidde et al. 2003). In SWPs, it is possible that N inputs from stormwater runoff support an algal dominated ecosystem, which then enhances the need for N-fixation during periods of limited N inputs to sustain algal biomass. This explanation for N-fixation may be supported by a recent SWP survey that found a widespread occurrence of diazotroph-dominated blooms, the most common cyanobacterial bloom found across 39 SWPs in southeastern USA (NC, USA; Grogan et al. 2023). However, chl-a was not related to N2 saturation in the current study, implying that these patterns may be complicated by other factors in urban SWPs.

Natural-humic sites in this study exhibited a distinct change between seasons, moving from increasingly net denitrifying with depth in the dry season to increasingly net N-fixing with depth in the wet season. Enhanced wet season N-fixation is contrary to what we predicted, as increased precipitation during the wet season should deliver more N and C from the watershed, pushing the system towards net denitrification. Previous studies have reported higher quantities of nitrogenase (N-fixing enzyme) encoding genes in a humic lake versus eutrophic lake (Linz et al. 2018), which also had higher nitrogenase abundances in the anoxic hypo- versus epilimnion (Fernandez et al. 2020). Similar to our study, these hypolimnetic nitrogenase abundances were highest in late summer (July, August) and related to N:P ratios (Fernandez et al. 2020). In this study, wet season natural-humic sites also experienced anoxic hypolimnions, and despite having the highest TDN concentrations, DIN was negligible. Previous lake sediments have reported summertime N-fixation associated with sediment P-release (Nifong et al. 2022). While we do not have seasonally distinct environmental data to explain why natural-humic sites showed different patterns between seasons, we hypothesize that variation in N2 saturation may have been a result of sediment-hypolimnetic redox conditions influencing P availability, as forms of N did not vary seasonally.

Limitations and opportunities for future research

Although this study highlights patterns of N-cycling across ponds of engineered or natural origin, future studies could build on these findings by modifying temporal, analytical, or other experimental approaches. In this study we collected daytime synoptic samples, yet denitrification and N-fixation rates have reported varying diel patterns in freshwater ecosystems (Levine and Lewis 1984; Oremland 1990; Reisinger et al. 2016b; Taylor et al. 2023). Higher frequency sampling over shorter periods of time may improve our understanding of daily net N-cycling processes. While we saw no significant seasonal differences in N2 saturation ratios or fluxes, a higher frequency of synoptic campaign (e.g., monthly) may enhance our understanding of seasonal patterns. In order to better understand biological changes to hypolimnetic N2:Ar and account for the potential effect of groundwater gas saturation, others should consider adding groundwater sampling of gases to their study if possible. Finally, our results suggest that both surface water and sediment N-fixation may contribute N to SWPs. Directly measuring N-fixation and denitrification rates from SWP surface waters and sediments can provide insight into the magnitude of these rates (as opposed to proportional ratios) as well as the importance of each component to N production and removal in these ecosystems.


A better understanding of N-cycling processes within SWPs is critical to predicting water quality outcomes in engineered landscapes. In this study, SWPs were more likely to be net N-fixing compared to natural ecosystems and had more variable and extreme influxes and effluxes of N2-N gas. This study adds to the growing body of literature highlighting that SWPs may not consistently provide the N removal service that is typically assumed. This result, combined with potential N production via N-fixation and the considerable number of SWPs that exist in urban landscapes, prompts further investigation into the impact that SWPs and other engineered ecosystems or stormwater control measures have on downstream waterbodies. Overall, these results highlight that knowledge of N-cycling processes and the associated drivers in SWPs should be considered when weighing actual benefits to water quality improvement for downstream ecosystems. An improved understanding of the biogeochemical processes driving N cycling is needed to inform SWP design and management in urban landscapes to maximize intended benefits.