The 26 ponds in this study were all located within a 40-mile radius within the metropolitan area of St. Paul and Minneapolis, Minnesota, United States. The ponds varied in land use type, from stormwater ponds in urban environments to natural forested ponds in preserves and parks. Ponds were divided into two categories based on origin, “natural” ponds and “constructed” ponds. While the complete history of some ponds was unknown, ponds were classified as constructed if there were known human alterations to the pond. All constructed ponds in this study are functioning stormwater ponds, managed for water regulation or the reduction of nutrient and metal exports. Daily aerial imagery at 3 m resolution from Planet Labs (Planet Team, 2017) was used to examine pond inundation, surface area, canopy cover, and floating macrophyte coverage. All ponds remain inundated year-round, and pond surface area did not detectably change within the study period. Canopy cover within a 50 m buffer around each pond was estimated using the area covered by mature trees within the buffer, using imagery from the same day each pond was sampled. Pond surface area ranged from 0.04 to 1.5 ha and pond max depth ranged from 0.3 to 1.8 m deep (Table 1).

The ponds were all sampled once between mid-July and mid-August 2021. Ponds were sampled between 10:30 in morning and 14:00 in the afternoon, to remove as much hourly variation as possible. While diel cycles have been shown to influence both CO₂ and CH₄ emissions, both sample time and sample date did not show a significant trend with measured emissions (Supplementary Figure S1). Nonetheless, diel studies have shown that daytime fluxes of CO₂ underestimate total daily fluxes while overestimating total CH₄ fluxes (van Bergen et al., 2019; Sieczko et al., 2020). Percent cover of floating macrophytes on each pond was estimated using aerial imagery. Since surface area coverage of macrophytes can change day to day with wind speed and direction, the maximum percent cover within a week period around the sample date for each pond was used. Visual estimates of floating macrophyte coverage were also taken onsite during sampling to verify the areal estimates and assess types of floating macrophytes. Floating macrophytes were almost exclusively duckweed (lemna sp.) and watermeal (wolffia sp.), here-on referred to as “duckweed”.

Surface water samples were collected in the middle of each pond and used to measure total phosphorus (TP), dissolved organic carbon (DOC), total dissolved nitrogen (TDN), nitrate/nitrite (NO₃/NO₂), and ammonium (NH₄). TP was measured using the molybdenum blue reaction with acid-persulfate digestion (Murphy & Riley, 1962). For DOC, TDN, NO₃/NO₂ and NH₄ analysis water samples were passed through a muffled 0.45 μm filter. DOC and TDN was measured using a Shimadzu TOC-L model high temperature carbon-analyzer with a TNM-L module (Shimadzu Corp., Kyoto, Japan). NO₃/NO₂ was measured using colorimetric analysis by the cadmium reduction method, and NH₄ was measured using colorimetric analysis by the salicylate/nitroprusside method using a Lachat 8500 FIA (Lachat Instruments, Loveland Colorado, United States). Spectral scans of water samples and DOC concentrations were also used to calculate specific ultraviolet absorbance at 254 nm (SUVA₂₅₄) as a measure of DOC aromaticity (Weishaar et al., 2003).

Profiles were taken at the deepest point of each pond to measure water temperature, dissolved oxygen (DO), pH, chlorophyll a (chl a), and conductivity (Manta probe, Eureka Water Probes, Austin Texas, United States). Measurements were taken every 5 s, and profiles were taken slow enough to allow for measurements every 0.05–0.1 m. DO profiles were used to calculate the anoxic fraction of the pond, defined as the fraction of sediment exposed to anoxic conditions (anoxic sediment area/pond surface area) (Nürnberg, 1995). DO values under 2 mg L⁻¹ were considered anoxic (Nürnberg, 1995). The anoxic sediment area was determined using DO profiles and pond hypsographic curves determined from bathymetry. As an example, in a 1.5 m-deep pond in which the DO profile was less than 2 mg L⁻¹ within 0.5 m of the bottom of the pond, all sediment at a depth of 1 m or greater was considered the anoxic sediment area. Bathymetry of ponds was acquired from maps or previous measurements, however for six ponds bathymetry was not available. For these ponds bathymetry was estimated using functions in the LakeAnalyzer R package (Read et al., 2011).

Water temperature profiles were used to calculate relative thermal resistance to mixing (RTRM) as a measure of stratification strength. RTRM (unitless value) was calculated for the whole water column by taking the difference of the densities of the surface water layer (using surface water temperature) and bottom water layer (using temperature ∼5 cm above sediment) and dividing by the difference in densities of water at 4 and 5°C (Birge, 1916). An RTRM >1 implies that the water column is stably stratified (more stable at higher values), while an RTRM ≤1 implies the water column is unstable and will mix (i.e. the density difference between water layers is less than the density difference between water at 4 and at 5°C). A value of zero implies the water layers are the same density.

Greenhouse gas emissions of CO₂ and CH₄ were measured using a floating chamber technique (Erkkilä et al., 2018; Grinham et al., 2018; Gorsky et al., 2019). N₂O fluxes proved to be too low to get reliable emission rates using the floating chamber, and therefore N₂O emissions were estimated from surface water and atmospheric concentrations. The floating chamber was constructed from an inverted white 5-gallon bucket, with a headspace volume of 10.02 L within the chamber. The chamber was wrapped in Styrofoam to float on the surface of the water, with 3 cm extending below the water surface to reduce turbulence. The floating chamber was connected directly to a portable greenhouse gas analyzer (DX4040 FTIR Gas Analyzer, Gasmet Technologies Oy, Vantaa, Finland) via inlet and outlet tubing to create a closed loop system. To measure CH₄ and CO₂ emissions, the float was placed on the surface of the water and the pressure was allowed to equilibrate through the outlet port. The float was then connected to the gas analyzer, and the analyzer began pulling air through the loop with an approximate pumping rate of 1.5 L min⁻¹. No alterations were made to the water surface before placing the chamber, and the chamber was placed on top of any floating macrophytes (such as duckweed) or algal mats on the surface of the pond as well as open water. Gas measurements were taken every 5 s, and incubations lasted at least 5 min, and up to 10 min if rates were low. Chamber incubations were taken at three locations on each pond, one near the shoreline, one at the deep spot, and one in between those two points. Ponds were sampled with either a kayak or canoe, with the chamber floating freely connected by approximately 2 m of tubing. If the boat had to be anchored due to wind, the anchor was placed far away from the floating chamber to avoid ebullition events caused by disturbing the sediment.

Concentrations of CO₂, CH₄, and N₂O in both the surface water and bottom water were measured using the headspace technique (McAuliffe, 1971). Surface samples were collected 5–10 cm below the surface, and bottom water samples were collected approximately 10 cm above the sediment using a Van Dorn water sampler. Van Dorn sampling was done slowly and carefully to not produce bubbles or cause turbulence, and after retrieval one end was opened to directly collect the bottom water sample. For all samples, 125 ml of water was collected in a 140 ml plastic syringe. Any bubbles formed when drawing water collected in the top of the syringe, and then were removed by reducing the syringe volume to 105 ml. Atmospheric air (32.5 ml) was introduced into the syringe and then vigorously shaken for 2 minutes. Thereafter, 30 ml of the headspace was transferred into a separate syringe, and immediately injected onsite into the Gasmet portable gas analyzer, equipped with a closed loop injection system (Wilkinson et al., 2019).

The diffusive flux for all gases across the air-water interface can be expressed as: F = k × ( C w − C s a t ) (1) Where F is the gas flux, k is the gas transfer velocity, C _w is the concentration of gas in the surface water, C _sat is the concentration of gas in the surface water at equilibrium with the overlying atmosphere. For CO₂ and CH₄, gas flux (F) was calculated directly with the floating chamber measurements. For CO₂, concentrations always increased or decreased linearly over time, and a simple linear regression method was used to calculate the flux based on the slope of linear increase (Xiao et al., 2014). For CH₄, ebullition led to a non-linear increase in CH₄ concentration when bubbling occurred (example shown in Supplementary Figure S2). With the 5 s measurement resolution, emission from ebullition events could be separated from the diffusive flux due to the high sampling frequency, as shown by Xiao et al. (2014). First, the CH₄ diffusion rate was calculated by fitting a linear regression to a long straight segment of the sample curve (Supplementary Figure S2). Multiplying this diffusion rate by the total sample time gave the CH₄ concentration in the chamber due to diffusion. CH₄ concentration due to ebullition was the surplus CH₄ concentration, calculated by subtracting the diffusion concentration and original background concentration from the total concentration at the sampling endpoint (Supplementary Figure S2). The Ideal Gas Law was used to convert concentrations to mass, and all emission rates for CO₂ and CH₄ are expressed in mg of carbon (i.e. CO₂-C, CH₄-C).

As shown by Eq. 1, the rate of gas flux will change over time as the concentration changes inside the floating chamber (as C _sat increases or decrease). While linear models are often used calculating gas fluxes from floating chambers (e.g. Xiao et al., 2014; Attermeyer et al., 2016; Gorsky et al., 2019), it has been shown that exponential models may be more appropriate, and linear models may underestimate fluxes by 10—30% for short incubations (5—25 min), and by over 50% for hour long incubations (Xiao et al., 2016). To test the use of a linear or exponential model, emission rates were calculated using both linear and exponential models for every CO₂ incubation (n = 75) and every CH₄ incubation where ebullition did not occur (n = 26), and model goodness of fits were compared using coefficient of determination R ². Contrary to the results from Xiao et al. (2016), the linear models had a higher R ² value for most incubations, and initial slope values did not significantly differ between linear and exponential models (Supplementary Table S1). Therefore, linear models were chosen to calculate CO₂ and CH₄ emission rates from the float chambers.

To calculate the concentration of CO₂, CH₄, and N₂O in the surface and bottom water, gas concentrations in the sample headspace were first calculated using the following equation (Wilkinson et al., 2019): X h e a d s p a c e = ( V l − V s V s ) ∆ X + X 0 (2) Where X _headspace is the concentration of the sample headspace, V _l is the volume of the closed loop system, V _s is the volume of the injected sample, ∆X is the change in gas concertation after sample injection, and X ₀ is the initial gas concentration. Inputs from the added atmospheric air was subtracted out from the headspace concentration using onsite measurements of atmospheric gas concentrations (usually around 400 ppm CO₂, 2 ppm CH₄ and 0.33 ppm N₂O). The partial pressure of gas in the headspace was used to calculate the moles of dissolved gas in the water (mol _aq ) according to Henry’s law, and this was added to the moles of gas in the headspace (mol _headspace ) and divided by the original sample volume to find the concentration of gas in the original sample (C _w ) (Johnson et al., 1990): m o l a q = P h e a d s p a c e × K H (3) m o l h e a d s p a c e = P h e a d s p a c e R T × V h e a d s p a c e (4) C w = m o l a q + m o l h e a d s p a c e V w (5) Where P _headspace is the partial pressure of gas in the headspace (atm), K _H is Henry’s constant (mol l⁻¹ atm⁻¹), R is the universal gas constant (0.082, L∙atm mol⁻¹ K⁻¹), T is the temperature (K), V _headspace is the volume of the headspace (L), and V _w is the volume of the water sample (L). Henry’s constants were calculated for each gas: CO₂ (Weiss, 1974), CH₄ (Wiesenburg & Guinasso, 1979), and N₂O (Weiss & Price, 1980) and corrected for water temperature and pressure. Constants were also corrected for salinity, as particularly the bottom waters of urban ponds can have high salinity due to road salt inputs. The highest recorded conductivity in the study ponds was 10,000 μS cm⁻¹. Conductivity for each sample was converted to salinity using equations from Fofonoff and Millard Jr (1983), and Hill et al. (1986). Molar concentrations for all gases were then expressed in mg of carbon or nitrogen (i.e. CO₂-C, CH₄-C, N₂O-N).

Using Eq. 1, the estimate C _w , and calculating C _sat according to Henry’s law, the gas transfer velocity (k) was estimated for each pond. Since CO₂ chamber fluxes were often influenced by floating macrophyte photosynthesis, CH₄ diffusion fluxes and surface concentrations were used with Eq. 1 to calculate k values specific to CH₄ and the sample temperature. k coefficients were then normalized to k ₆₀₀ values for comparison among gases and at different water temperatures, with the following equation (Jähne et al., 1987): k 600 = k g , T ( 600 S c g , T ) − n (6) Where k _g,T and Sc _g,T are the gas transfer velocity and Schmidt number of a given gas (in this case CH₄) and temperature (Wanninkhof, 1992). Temperature specific Sc values were calculated using equations in Wanninkhof (1992), and an n of 2/3 was used for all calculations, as this factor is appropriate for a smooth liquid surface (Deacon, 1981). These site-specific k ₆₀₀ values were then used to estimate emissions for N₂O, based on N₂O surface concentrations and using Eq. 1.

For each site, equilibrium concentrations of each gas were calculated with Eq. 3, using the gas and temperature specific K _H (mol l⁻¹ atm⁻¹) and the partial pressure of the gas in the atmosphere measured at each site.

To evaluate the total greenhouse gas effect of emitted gases, gas emissions (floating chamber method for CO₂, CH₄, concentration method of N₂O) were then converted to CO₂eq (mg C m⁻² d⁻¹) on a mass basis by multiplying the CH₄ emissions by 37 (Derwent, 2020) and N₂O emissions by 265 (Myhre et al., 2013).

To test for differences in the chamber emission estimates among the three sample sites (near-shore, intermediate point, and the deep point) on each pond, linear mixed effect models were used with ponds included as a random effect (Zuur et al., 2009). For the rest of analyses, chamber emission estimates from the three sample sites were averaged, to provide a single mean CO₂ and CH₄ emission rate for each pond. To evaluate individual relationships among all greenhouse gas metrics and environmental variables, a Pearson correlation matrix was used. Relationships among all variables were first visually inspected with scatter plots to check for a linear relationship, and then correlations were calculated with the raw values. Environmental variables included surface area, max depth, TP, DOC, SUVA₂₅₄, TDN, NH₄, chl. a, anoxic fraction, RTRM, duckweed coverage, and canopy coverage.

To further investigate the effect of duckweed on pond greenhouse gas emissions, ponds were classified as “duckweed ponds” and “non-duckweed ponds” using a k-means cluster analysis of pond duckweed coverage, resulting in 10 duckweed ponds and 16 non-duckweed ponds. Unequal variance t-tests were used for all comparisons between natural/constructed categories and duckweed/non-duckweed categories, as variances were not homogeneous between categories for many variables. To test for interactions among the categorical variables (natural/constructed and duckweed/non-duckweed), two-way anova tests were used. p-values < 0.05 were considered significant for all analyses. All statistics were performed using R statistical software (R 4.4.1, R Core Team 2021). The lme4 package was used for mixed effect models along with the lmerTest package to test for significant differences between sampling sites, and the stats package was used for the unequal variance t-tests, k-means cluster analysis, and two-way anova tests.

The 26 ponds varied widely in water chemistry and physical characteristics (Table 1). The ponds were generally small and shallow with a mean (range) surface area of 0.51 (0.04–1.5) ha and a mean max depth of 1.2 m (0.3–1.8). Most ponds were high in total phosphorus (TP) (mean 150 [44—470] μg L⁻¹), dissolved organic carbon (DOC) (mean 12 [7.1–20] mg L⁻¹)), and total dissolved nitrogen TDN) (mean 0.79 [0.26–1.7] mg L⁻¹). All but one pond (New Seminary) had undetectable levels of NO₂/NO₃ (below 10 μg L⁻¹). Most ponds showed evidence of stratification, with relative thermal resistance to mixing (RTRM) greater than one in all ponds (mean RTRM of 77, range of 6.8–180).

Natural ponds and constructed ponds did not significantly differ in TP levels (p = 0.66), but natural ponds did have significantly higher levels of DOC and TDN (p = 0.045, 0.016). On average, natural ponds had a larger surface area than constructed ponds (0.92 vs. 0.32 ha, p = 0.017), but constructed ponds had a deeper max depth (1.3 vs. 0.97 m, p = 0.14), though not significantly.

Duckweed coverage on the ponds ranged from 0 to 100%, though most ponds had either close to no duckweed present or almost full duckweed coverage (Table 1). Ponds were split into two duckweed categories using k-means cluster analysis, and duckweed ponds had an average duckweed coverage of 95.5% (n = 10), while non-duckweed ponds had an average duckweed coverage of 2.84% (n = 16). Compared to non-duckweed ponds, duckweed ponds had significantly higher TP concentrations (p = 0.023), significantly lower surface oxygen concentrations (p = 0.037), and a significantly larger anoxic fraction (p < 0.001; Table 2). There were no significant interactions for environmental variables or greenhouse gas metrics between the natural/constructed and duckweed/non-duckweed categories.

Both concentrations in the surface and bottom water and emissions varied by several orders of magnitude across the ponds for each greenhouse gas (Table 3). In the surface water, mean concentrations of gases across all ponds were 2.03 mg CO₂-C L⁻¹, 0.250 mg CH₄-C L⁻¹, and 0.501 µg N₂O-N L⁻¹. Mean concentrations in the bottom water were 10.3 mg CO₂-C L⁻¹, 1.38 mg CH₄-C L⁻¹, and 0.313 µg N₂O-N L⁻¹ (Table 3). The difference between surface and bottom concentrations was significant for each gas, with CO₂ concentrations significantly higher in the bottom water (mean difference 8.24 mg CO₂-C L⁻¹, p = 0.002), CH₄ concentrations significantly higher in the bottom water (mean difference 1.13 mg CH₄-C L⁻¹, p = 0.002), and N₂O concentrations significantly lower in the bottom water (mean difference -0.188 µg N₂O-N L⁻¹, p = 0.017). Equilibrium concentrations were on average 162 μg L⁻¹ for CO₂-C, 0.034 μg L⁻¹ for CH₄-C, and 0.24 μg L⁻¹ for N₂O-N, meaning that the ponds were generally super saturated with all three greenhouse gases. In the surface water, CO₂ concentrations were on average 11-fold supersaturated, CH₄ concentrations were on average 7,000-fold supersaturated, and N₂O concentrations were on average 2-fold supersaturated.

Emission rates using the floating chamber method did not significantly vary by location within each pond (Supplementary Figure S3), with no significant differences between the shallow, middle and deep locations (p = 0.245, 0.782, and 0.315 for CO₂ emission, CH₄ diffusion, and CH₄ ebullition, respectively). Though not significant, CH₄ ebullition emissions were highest at the near-shore shallow location, while CH₄ diffusion emissions were highest at the deep point location (Supplementary Figure S3). When all three floating chamber incubations were averaged to get a mean rate for each pond the mean emission rates for CO₂ and CH₄ were −25.9 mg CO₂-C m⁻² d⁻¹ and 705 mg CH₄-C m⁻² d⁻¹, with mean rates of 219 mg CH₄-C m⁻² d⁻¹ for diffusion and 484 mg CH₄-C m⁻² d⁻¹ for ebullition. Using surface concentrations and calculated k ₆₀₀ values, the average emission rate for N₂O was 0.398 mg N₂O-N m⁻² d⁻¹. The mean k ₆₀₀ across all ponds was 1.12 m d⁻¹ but ranged from 0.240—2.82 m d⁻¹ (Table 3). k ₆₀₀ did not significantly correlate with any environment variables but was lower in duckweed ponds compared to non-duckweed ponds (0.924 vs 1.24 m d⁻¹), though not significantly (p = 0.251).

The mean total daytime emission rate for all three greenhouse gases in CO₂ equivalents was 26,100 mg C m⁻² d⁻¹, with CH₄ accounting for over 94% of the total emissions in CO₂ equivalents on average. CO₂ accounted for about 4% of the total emissions in CO₂ equivalents, but ranged from a sink of −12%–57%. N₂O accounted for 2.4% of the total emissions in CO₂ equivalents, ranging from a sink of −2.2%–11% (Table 3).

CO₂ emission rates from the floating chamber method was significantly negatively correlated with TP, chl. a, anoxic fraction, RTRM, and duckweed coverage, with the strongest correlations being duckweed, and chl. a (Table 4; Figure 1). However, surface concentrations of CO₂ were significantly positively correlated with TP and anoxic fraction, as well as ammonium concentration (Table 4; Figure 1).

CH₄ emission rates and surface concentrations correlated with similar variables, with surface CH₄ concentrations significantly positively correlated with surface area, TP, RTRM, anoxic fraction, and duckweed coverage. For emission rates of CH₄, total emission, ebullition, and diffusion were all significantly positively correlated with TP, anoxic fraction, and duckweed coverage, with ebullition also negatively correlated with maximum depth, and diffusion also positively correlated with ammonium and chl-a (Table 4; Figure 2).

N₂O surface concentrations were not significantly correlated with any environmental variables. All three greenhouse gas bottom concentrations were correlated with maximum depth (positively for CO₂ and CH₄, negatively for N₂O). CO₂ bottom concentration was also positively correlated with SUVA values, and N₂O bottom concentration was negatively correlated with chl-a, anoxic fraction, and RTRM (Table 4).

Duckweed coverage was significantly correlated with many of the greenhouse gas metrics, as well as significantly positively correlated with TP and anoxic fraction (Table 4; Figure 2). Duckweed ponds had significantly higher CH₄ emissions for both total emission (p = 0.04) and diffusion (p = 0.0085), but not ebullition (p = 0.095). Duckweed ponds also had significantly lower CO₂ emissions (p = 0.0055; Figure 3). For greenhouse gas concentrations, duckweed ponds had significantly higher CH₄ surface concentrations (p = 0.004), and significantly lower N₂O bottom concentrations (p = 0.043; Figure 4).

Most of the 26 ponds in this study were super-saturated with all three greenhouse gases during the study period. Concentrations of greenhouse gases in the surface water were comparable to other recent studies of pond greenhouse gases, including ponds in Connecticut (mean of 4.32 mg CO₂-C L⁻¹ and 0.396 mg CH₄-C L⁻¹ (Holgerson, 2015)), and ponds in Denmark (mean of 1.94 mg CO₂-C L⁻¹, 0.044 mg CH₄-C L⁻¹, and 0.8 µg N₂O-N L⁻¹ (Audet et al., 2020)). However, surface concentrations of CO₂ and CH₄ were higher than values typically reported for lakes. In a survey of 1,835 lakes, the mean CO₂ concentration of the upper 10% of the samples were 16-fold above atmospheric equilibrium, while the average of all lakes was about 2-fold above equilibrium (Cole et al., 1994). For the ponds in this study, the highest surface CO₂ concentration was 40-fold above atmospheric equilibrium, while the average of all 26 ponds was 11-fold above equilibrium. For CH₄, a survey of surface methane concentrations from 48 lakes had a mean concentration of 8.3 µg CH₄-C L⁻¹ with 27.9 µg CH₄-C L⁻¹ as the highest concentration (Bastviken et al., 2004). The ponds in this study had an average surface concentration of 250 µg CH₄-C L⁻¹, showing ponds can be extreme hotspots of CH₄ production. These high CH₄ concentrations may also signal the lack of CH₄ oxidation in ponds, potentially driven by anoxia and/or a lack of other oxidants such as nitrate, or shallow depth that allows for rapid transport of CH₄ from the sediment to the surface (Bastviken et al., 2004). Surface N₂O concentrations were not much higher than atmospheric equilibrium, and four of the 26 ponds were undersaturated in surface N₂O. These ponds overall had lower surface N₂O concentrations compared to many lakes. A study of 15 Swiss lakes found an average of 434% saturation for surface N₂O (Mengis et al., 1997), while the ponds in this study had an average saturation of 209%.

Bottom water concentrations of CO₂ and CH₄ were on average much higher than surface concentrations, while N₂O concentrations were often lower in the bottom waters. While all ponds were less than 1.8 m deep, all ponds still showed evidence of stratification, with relative thermal resistance to mixing (RTRM) greater than one in all ponds during the day. Thus, many ponds had substantial storage of CO₂ and CH₄ in the bottom waters, as well as undersaturation of N₂O. Bottom water concentrations of greenhouse gases are less often measured or reported for lakes and ponds, but the ponds in this study had high concentrations of CO₂ and CH₄ that compare to or exceeded anoxic hypolimnion concentrations of stratified lakes. Hypolimnion concentrations of CO₂ from northern lakes can reach 3.6 to 5.4 mg CO₂-C L⁻¹ (Ducharme-Riel et al., 2015), while these study ponds had a maximum bottom CO₂ concentration of 49 mg CO₂-C L⁻¹ (mean 10.3 mg CO₂-C L⁻¹). The highest hypolimnion CH₄ concentrations in a study of three Wisconsin lakes in late summer ranged from 3.6—8.4 mg CH₄-C L⁻¹ (Bastviken et al., 2008), while the highest bottom concentrations in these study ponds was 6 mg CH₄-C L⁻¹ (mean 1.38 mg CH₄-C L⁻¹). This shows that some small ponds can still build up large concentrations of CO₂ and CH₄ in the bottom waters, which likely are emitted during fall turnover or other mixing events. Intermittently mixed ponds may mix during the night or early morning when the surface water cools (Holgerson et al., 2022), and nighttime emissions could be higher due to mixing events in some of these ponds. N₂O concentrations were on average lower in the bottom waters compared to the surface waters, and 14 of the 26 ponds were undersaturated in the bottom waters, compared to only four ponds undersaturated in the surface waters. This pattern agrees with many findings for lakes and reservoirs, where N₂O concentrations peak in the surface water or oxic-anoxic interface, and are undersaturated in the permanently anoxic part of the hypolimnion (Mengis et al., 1997; Beaulieu et al., 2015).

CO₂ using floating chamber estimates suggested that the ponds are a net sink (Table 3), though these ponds were sampled during the day and in the peak growing season and this likely does not reflect overall daily or annual CO₂ fluxes from these ponds. Fluxes of CO₂ would likely be the lowest during day when photosynthesis is high and would peak during the night when respiration predominates. Therefore, the daily CO₂ fluxes measured in these ponds are likely underestimated due to the lack of nighttime measurements. The negative fluxes of CO₂ contrasted with the CO₂ supersaturation in the surface waters of most ponds, due to duckweed creating a CO₂ sink above the surface of the duckweed covered ponds. Despite middle floating chamber incubations and surface concentrations being measured in the same spots, using surface concentrations to estimate CO₂ fluxes would have greatly overestimated daytime CO₂ emissions for the duckweed covered ponds.

Fluxes of CH₄ were always positive, and the CH₄ emission rates from these ponds were extremely high for freshwater ponds and lakes. CH₄ emissions have been shown to peak during the day and regress at night (Sieczko et al., 2020), and daytime sampling of these ponds may overestimate daily emission rates. Nonetheless, ponds may mix at night due to convective cooling unlike larger lakes (Andersen et al., 2017), which may lead to higher emissions during nighttime as gases escape from the bottom waters. CH₄ emissions also peak during the growing season (van Bergen et al., 2019), and therefore the CH₄ emission rates found in these study ponds may represent the peak emission rates that these ponds experience. However, mixing during the shoulder seasons could also lead to the release of CH₄ from the bottom waters, and fall and spring CH₄ emission rates can be higher than other seasons (Riera et al., 1999; Jansen et al., 2019). Mean total CH₄ emissions from these ponds were among the highest reported for studies involving multiple ponds (Table 5). Diffusive CH₄ emission were high compared to a global assessment of lakes and ponds, where waterbodies less than 1 ha in size had a mean diffusive emission rate of 21.8 mg CH₄-C m⁻² d⁻¹ (Holgerson & Raymond, 2016). However, emissions were similar to diffusive CH₄ emissions found in Virginia stormwater ponds (mean 271.8 mg CH₄-C m⁻² d⁻¹) (Gorsky et al., 2019). Ebullition CH₄ emissions are less often measured in ponds, and ebullition rates are highly variable both spatially and temporally. Ebullition CH₄ emissions from these study ponds were higher than mean ebullition fluxes in Canadian ponds (mean 55.2, max 204 mg CH₄-C m⁻² d⁻¹) (DelSontro et al., 2016), but within a similar range of ponds across Canada and Missouri (highest median pond ebullition rate of 485 mg CH₄-C m⁻² d⁻¹) (Baron et al., 2022). Ebullition rates can be strongly correlated with temperature (DelSontro et al., 2016), and sampling in mid-summer during the day likely facilitated the peak ebullition rates observed from these ponds.

The mean gas transfer velocity, k ₆₀₀ , of the study ponds was high but many of the ponds fell within the range of other measurements of k ₆₀₀ in small, shallow waterbodies (Xiao et al., 2014; Holgerson et al., 2017). Many studies that use a literature value of k ₆₀₀ to calculate greenhouse gas fluxes based on surface concentrations use a k ₆₀₀ value of less than 1 m d⁻¹ for small waterbodies, such as the 0.36 m d⁻¹ used by Holgerson and Raymond (2016). The high average k ₆₀₀ in these ponds of 1.12 m d⁻¹ was driven by a few ponds with exceptional large k ₆₀₀ values (up to 2.82 m d⁻¹), and the median value of 0.89 m d⁻¹ is closer to what studies assuming a constant k ₆₀₀ may use for small waterbodies. Gas transfer velocities can be extremely variable, and the high k ₆₀₀ values seen in these ponds may mean that assuming a low constant k ₆₀₀ could underestimate greenhouse gas emissions based on surface water concentrations.

CO₂ concentrations and emissions correlated with similar variables, but in opposite directions. While surface CO₂ concentrations were significantly positively correlated with TP, CO₂ emissions were significantly negatively correlated with TP. This is likely due to a relationship between duckweed and TP, with phosphorus facilitating duckweed growth (Lasfar et al., 2007), that then reduces CO₂ emissions. Neither CO₂ concentrations nor emissions were correlated with DOC concentration, a relationship that has been found in many lakes (Raymond et al., 2013). Surface CO₂ concentrations did correlate negatively with surface oxygen concentrations (r ² = 0.46, p = 0.00014), a relationship that has been shown to best predict CO₂ concentrations in ponds (Kankaala et al., 2013; Holgerson, 2015). Low oxygen is likely consequence of the combination of high rates of aerobic respiration in ponds, where sediment respiration can affect the whole water column (Holgerson, 2015), and low ventilation rates, especially in the systems that were protected from wind and with high abundance of duckweed.

Both CH₄ concentrations and emissions were predicted by similar variables. CH₄ diffusion and TP concentration was the strongest correlation among any greenhouse gas metric and any environmental variable, and TP also correlated with ebullition rates and surface CH₄ concentrations. This trend agrees with previous research that showed higher CH₄ emissions in more eutrophic lakes and ponds (DelSontro et al., 2016, 2018; Beaulieu et al., 2019). While it is unknown if high phosphorus levels stimulate CH₄ production directly, excess phosphorus likely helps create conditions conducive to CH₄ production, including stimulating algal growth, which when degraded can lead to more labile organic matter and increased anoxia (Davidson et al., 2018). The pond anoxic fraction was also significantly positively correlated with all CH₄ gas metrics. As methanogenesis is primarily an anoxic process, more anoxic water can lead to increased CH₄ formation in the water column and the sediments, as well as decreased methane oxidation (Bastviken et al., 2004).

While no environmental variables correlated with surface N₂O, N₂O concentrations from the ponds in this study compare well to other studies that have not found significant N₂O emissions from ponds (Singh et al., 2005; Audet et al., 2020). Despite high nutrients in these ponds, many which are hyper-eutrophic based on phosphorus concentrations (Carlson, 1977), there was almost no detectable levels of NO₃/NO₂ in any of the ponds, which can be a strong driver of N₂O emissions (Mccrackin & Elser, 2010; Beaulieu et al., 2015; Wang et al., 2021). Furthermore, almost all of the ponds were stratified to some degree during the day, similar to conditions in Canadian farm ponds that became N₂O sinks in late summer (Webb et al., 2019).

There was no significant difference in greenhouse gas concentrations or emissions between constructed ponds and natural ponds, which was likely due to very similar water chemistry on average between natural and constructed ponds. There were no significant differences in TP among constructed and natural ponds, highlighting that most ponds are often highly eutrophic, even among different land use types. While there can be many reasons why constructed ponds may function differently compared to natural ponds (Clifford & Heffernan, 2018), constructed ponds can vary greatly in physical and chemical characteristics just like natural ponds, potentially mitigating differences between the two at larger scales.

Duckweed coverage had a discernable effect on the physical and chemical states among ponds, as well as a significant effect on all three greenhouse gases. Among the 26 ponds, conditions either supported the complete coverage of duckweed over the pond surface, or ponds were not suitable for duckweed to have significant growth. Conditions for duckweed growth may depend on terrestrial canopy cover to block wind, as well as available nutrients (Hillman, 1961) (Table 4). Very few ponds had only partial duckweed coverage, though larger ponds with duckweed often had some open water if there was any wind. Duckweed coverage had clear effects on oxygen levels, with duckweed ponds having significantly less surface oxygen and a larger anoxic fraction (Tables 3, 4). Duckweed mats are known to reduce oxygen levels in small waterbodies (Pokorny & Rejm, 1983; Ceschin et al., 2019), likely due to duckweed shading out photosynthetic organisms in the water column and leading to almost no oxygen production within the pond. A blanket of duckweed coverage could also reduce atmospheric exchange of oxygen, and duckweed ponds did have slightly lower mean gas transfer velocity, though not significantly. Stratification was also stronger in duckweed ponds, though not significantly (Table 4), and this could also play a role in developing a more anoxic hypolimnion. Duckweed ponds were also significantly higher in TP (Tables 3, 4), likely due to increased anoxia which led to a positive feedback loop by stimulating further duckweed growth (Lasfar et al., 2007). Greater anoxia can stimulate higher rates of internal loading of phosphorus from the sediments (Taguchi et al., 2020) (Figure 5).

Through these physical and chemical effects on the water column, duckweed coverage created two distinct states among the 26 ponds in this study. The increased anoxia and high phosphorus concentrations promoted by duckweed coverage also coincided with increased CH₄ emissions from duckweed ponds (Figure 2), and duckweed ponds had mean CH₄ emissions over twice that of non-duckweed ponds for each of total, ebullitive, and diffusive emissions (Figure 3). Degrading duckweed may also create a large pool of labile organic matter in duckweed ponds similar to submersed macrophytes (Hilt et al., 2022), which could further promote methanogenesis and ebullition (West et al., 2012; Zhou et al., 2019). Duckweed ponds were on average CO₂ sinks, despite having higher CO₂ concentrations in the surface waters compared to non-duckweed ponds (Figures 3, 4). Though duckweed ponds were a daytime CO₂ sink, higher CO₂ concentrations may mean that these ponds release more CO₂ during the night, potentially offsetting the CO₂ captured by the duckweed. Further, CH₄ accounted for over 95% of the total emission rate in CO₂eq on average for these ponds, showing that CH₄ dominates the emissions in small ponds. Overall duckweed ponds had about three times the emission rate in CO₂ equivalents compared to non-duckweed ponds (mean 31 vs. 11 g C m⁻² d⁻¹).

Duckweed ponds also had higher concentrations of CH₄ and CO₂ in the bottom waters, though not significantly. Duckweed ponds did have significantly lower concentrations of N₂O in the bottom waters, with most bottom waters of duckweed ponds undersaturated. This could again be due to anoxic conditions in the hypolimnion of duckweed ponds, which has been linked to undersaturated N₂O (Beaulieu et al., 2015).

Floating macrophyte coverage, including duckweed, is increasing worldwide in recent decades, correlated with expanding urbanization and increased eutrophication of aquatic systems (Kleinschroth et al., 2021). While not all floating macrophytes may create similar dynamics, other floating macrophytes that shade out water column oxygen production could lead higher anoxia, higher phosphorus concentrations, and higher CH₄ emissions as seen in the ponds in this study. For urban stormwater ponds, floating macrophytes may undermine the goal of capturing and preventing nutrients from flowing into downstream waterways by increasing internal loading of from phosphorus-rich sediments (Taguchi et al., 2020). Further, duckweed and other floating macrophytes may be causing an unforeseen increase in greenhouse gas emissions from both constructed and natural ponds.

In constructed ponds, management of duckweed and other floating macrophytes can be difficult, as weed control can be costly and labor intensive (Kleinschroth et al., 2021). Duckweed may be utilized as an obvious “indicator” of degraded conditions in constructed ponds, likely signaling high nutrients, poor water quality, and high greenhouse gas emissions, and duckweed has already been used as an indicator of pollution and heavy metal contamination (Garg & Chandra, 1994). While canopy cover blocking wind may be an important influencer of duckweed growth, data for this study suggest phosphorus levels may be one of the most important drivers of duckweed coverage. While management of stormwater ponds already often focuses on reducing nutrient levels, the duckweed state in ponds may act similarly to stable states seen in shallow lakes (Scheffer et al., 1993), where duckweed takes over as the dominant primary producer, and the ensuing degraded water quality state is more resistant to reduced nutrient levels. Oxygen depletion may be the largest factor in increasing CH₄ emissions caused by duckweed, and increasing oxygenation in ponds may limit phosphorus loading as well as reduce greenhouse gas emissions.