Introduction

Methane (CH4) and carbon dioxide (CO2) fluxes are the two most important processes driving climate warming (Myhre & Shindell 2014), and lakes emerge as natural hot spots in the transmission of greenhouse gases (GHG; Rosentreter et al. 2021). Small lakes and ponds (< 5 ha) are globally abundant and have high fluxes relative to their areal cover, but these systems are poorly defined and often overlooked in investigations of GHG fluxes (Deemer et al. 2016; Richardson et al. 2022). Higher GHG fluxes in ponds, compared to larger lakes, may be due to the compounding effects of higher organic substrate concentrations and higher temperatures (Deemer & Holgerson 2021). Ponds have a higher shoreline-to-surface-area ratio and, consequently, higher inputs of terrestrial organic material relative to surface area and water volume, causing high rates of degradation and production of CO2 and CH4 (Kuhn et al. 2021; Sand-Jensen & Staehr 2009). Furthermore, they receive substantial inputs of dissolved CH4 and supersaturated CO2 through sub-surface flow from the surroundings (Abril & Borges 2019; Striegl & Michmerhuizen 1998). As a result of large lateral inputs and internal production of organic material, pond sediments produce CO2 and CH4 which fuels the emission of GHG (Jansen et al. 2020b).

Goeckner et al. (2022) found differences in GHG emissions and carbon storage between natural and artificial ponds, with artificial ponds being carbon sources when younger (< 20 years), but carbon sinks when they become older (> 20 years). Moreover, surrounding land use and soil types may influence GHG fluxes (Peacock et al. 2021). Thus, despite only comprising 8.6% of the global lake area, small ponds (< 0.01 km2) have been estimated to be responsible for approximately 40% of diffusive CH4 emissions (i.e., 13.2–21.2 Tg CH4 y−1), according to early global estimates (Holgerson & Raymond 2016; Rosentreter et al. 2021). However, numbers and sizes of smaller ponds are difficult to assess using satellite imagery and include in the global GHG budget, resulting in large uncertainties of their contribution (Grinham et al. 2018).

CH4 is formed by the degradation of organic substrates in the absence of oxygen and alternative electron acceptors (e.g., nitrate, ferric oxides and sulfate; Fenchel et al. 2012). The production is promoted by high rates of degradation at high temperatures and an ample supply of easily degradable organic substrates, when oxygen supply and alternative electron acceptors cannot metabolize the production of small organic acids (mainly acetate) and H2 to CO2 and H2O (Fenchel et al. 2012). This situation may account for the observed increase in CH4 fluxes and the CH4:CO2 flux ratio as lake size decreases, causing a higher warming potential per unit surface area (Holgerson & Raymond 2016; West et al. 2016). Hence, the higher input of organic substrates from external sources and in-lake production relative to surface area may increase CH4 formation in small, typically warmer lakes during summer (Beaulieu et al. 2019; DelSontro et al. 2016). Although CH4 and CO2 emissions may be high from lakes and ponds, substantial carbon sequestration in the sediments will reduce net emission from the sites and downstream in the hydrological network (Anderson et al. 2014; Céréghino et al. 2014; Gilbert et al. 2021; Goeckner et al. 2022; Taylor et al. 2019). Thus, the debate of pond construction as a climate mitigation strategy should be evaluated as part of carbon budgets for the entire hydrological network at relevant time scales. Nonetheless, the debate of pond construction as a climate mitigation strategy remains unsettled.

The rates of microbial CH4 production in sediments increase more with higher temperatures than CO2 production, thereby increasing the CH4:CO2 flux ratio (Yvon-Durocher et al. 2014). In eight Swedish lake sediments, CH4 production rates measured in incubations increased 10–100-fold between 4 and 20 °C and more strongly above 10 °C, though with considerable variation between lakes (Duc et al. 2010). The influence of temperature on CH4 fluxes from lake surfaces is even more complex because CH4 may be produced and oxidized by different aerobic and anaerobic processes in the sediment and the water column (Bogard et al. 2019; D'Ambrosio & Harrison 2021; Donis et al. 2017). Nonetheless, we expect substantially higher CH4 production and fluxes during summer than winter, and particularly higher CH4:CO2 flux ratios as higher summer photosynthesis may reduce CO2 emission (Jansen et al. 2020a; Wik et al. 2013, 2018, 2014).

A recent application of automated ventilated floating chambers deployed to 80 sites on a 10-ha lake revealed high spatial variability of CH4 and CO2 fluxes (Sø et al. 2023a, b). The chambers provided hourly diffusive and ebullitive fluxes throughout the diel cycle because the chamber headspace is automatically replenished with atmospheric air every hour allowing new measurements of gas fluxes between water and air. Sieczko et al. (2020) found that CH4 fluxes from Swedish lakes were 2.4 times higher in the daytime than in the nighttime. This may be due to higher daytime wind speeds causing higher water column turbulence and sediment shear stress promoting CH4 release from water and sediment surfaces (Joyce & Jewell 2003). When estimates of CH4 fluxes are based solely on daytime measurements, this may lead to an overestimation of diel values (Sieczko et al. 2020). Floating chambers with automated headspace venting are suitable to test whether diel variations in CH4 and CO2 fluxes also are pronounced in ponds where wind speeds are markedly reduced due to sheltering by riparian vegetation, and short wind fetches (Vachon & Prairie 2013). CO2 flux is likely lower during daytime due to photosynthesis and CO2 consumption and higher during nighttime due to respiratory CO2 release (Martinsen et al. 2022; Sand-Jensen et al. 2019). Measurements assessing variability of CO2 and CH4 fluxes within ponds are few.

Our goal was to determine CO2 and CH4 fluxes at high spatial and hourly resolution in six temperate ponds during summer and winter using automated floating chambers. To cover the two main types of small, nutrient-rich ponds, we examined three ponds located in a deciduous forest and three ponds located in open landscape nearby. We expect that forest ponds receive a higher input of organic matter that supports CO2 and CH4 effluxes. Ponds situated in open country receive higher summertime solar irradiance, are warmer and support higher in-pond primary production that, in turn, lowers CO2 efflux. Overall, high input of organic matter and nutrient-rich conditions in both pond types may lead to profound CH4 flux by diffusion and ebullition during both summer and winter.

Study sites and methods

This study was conducted during the winter and summer of 2022 in six shallow, nutrient-rich ponds located soil consisting of meltwater gravel and freshwater peat in North Zealand, Denmark (Fig. 1, Table 1). In some cases, sensors or batteries failed after a short time due to rough weather conditions; thus, data from two winter periods are included in the analyses of some ponds (Number of fluxes for each station is indicated on Fig. S2-4). Summer measurements were completed in late May or June, while winter measurements were done from January to the beginning of March. Three forest ponds (F1–F3, Table 1) and three are in open land nearby (O1–O3, Table 1) were studied. The area surrounding the forested ponds was dominated by old deciduous forest with few to no buildings. The forest is used for recreation and some timber production. The open land ponds were located in open grasslands, without buildings but in urban transitions. The ponds vary in size (878–5897 m2) and all experience substantial variations in water level. The age of the ponds varied, with two of the ponds (one in a forest and one open) being natural ponds of unknown age, whereas the rest were human-made ponds created 16–26 years ago to restore the natural hydrology or increase the biodiversity (Table S1). Water depth was measured in summer at the sites of GHG measurements. Four sites were selected for each pond, two in shallower waters (30–50 cm) and two in deeper waters (50–100 cm), whenever possible. One very deep pond (O2; 274 cm) was equipped with two additional floating chambers to cover the both shallow, intermediate and deep areas (Table 1). Distance between chamber positions ranged from 10 to 48 m.

Fig. 1
figure 1

Arial imagery of the ponds and weather station (W). Blue areas indicate the ponds

Full size image
Table 1 Overview of location and physiochemical conditions in three small forest ponds (F) and three small open ponds (O). LOI is the sediment organic content (mean (± SD) as % of dry mass) at the 4–6 sites of gas measurements
Full size table

Environmental conditions

Wind speed, gust and direction were measured one meter above the water surface in the middle of one forest pond (F1; HOBO S-WSB-M003 and HOBO S-WDA-M003, Onset, USA) and at a nearby public weather station (4–6 km from the ponds; Davis Vantage Pro2, Lyngvig 2022). Wind speeds recorded at F1 were applied for the other two forest ponds, while public weather station data were used for all open ponds. Throughout the measuring period from 1 December 2021 to 30 July 2022, wind speeds in the forest remained lower than in open terrain. The barometric pressure was also acquired from the public weather station and used in Eq. 1 for all ponds. Each site was equipped with an oxygen and temperature sensor placed 50 cm above the sediment, when possible, and otherwise at the water surface. We measured temperature and oxygen in the pond water every 10 min (MiniDOT, PME, USA), while total phosphorus (TP), total nitrogen (TN) and dissolved organic carbon (DOC) were measured once during winter and summer following the methods of Kragh and Søndergaard (2004). TP and TN samples were stored at − 18 °C. Upon analyses, the samples were acid-boiled in an autoclave and measured on a SKALAR autosampler. The DOC samples were filtered through a GF/F filter (Whatman, UK) and conserved using 2 M HCl. The water sample was later analyzed on a Shimadzu TOC (Shimadzu, Japan). Chlorophyll a and light absorbance of colored dissolved organic material (CDOM; Jespersen & Christoffersen 1987; Kirk 1994) were measured only during summer. Chlorophyll a was measured by ethanol extraction and spectrophotometric measurements of absorbance at 665 and 750 nm, while CDOM was measured as absorbance of 440 nm. The organic content of the upper 10 cm of sediment was measured in four cores (21 cm2) for each pond in summer, with one core retrieved from each site within the 6 ponds and, thus, a total of 24 cores. The cores were split into two depth layers (0–5 cm and 5–10 cm) and the organic content was measured on duplicate samples as loss on ignition (LOI) at 550 °C. During winter, only one oxygen-temperature sensor remained in each pond. GHG measurements were not performed during the brief periods when the ponds were ice-covered which typically lasted less than one or two weeks. Submerged plant coverage in a plot of 2 × 2 m below the floating chambers was assessed visually during summer and categorized as: absent (0%), sparse (1–20%), high (20–90%), or complete coverage (90–100%). The most common submerged species present within the plot were Callitriche sp., Ceratophyllum demersum, Chara sp., Elodea canadensis, Myriophyllum spicatum, Potamogeton obtusifolius and Potamogeton pusillus.

Automated floating chambers

Floating chambers contained electronic equipment that continuously measured CH4 and CO2 as described in Bastviken et al. (2015), modified to perform hourly measurements enabled by automated venting of the headspace by using an air pump (surface area: 0.0615 m2, volume: 0.0135 m3; Martinsen et al. 2018; Sø et al. 2023a, b). All measurements were done in 2022, with winter measurements performed in January and February and summer measurements between late May and June. CH4, relative humidity, and temperature were measured every 2 s, while CO2 was measured every 20 s. CH4 and CO2 fluxes were calculated by the use of small sensors (CH4: NGM2611-E13, Figaro, USA, and CO2: K33 ELG, SenseAir, Sweden) mounted in the air headspace within floating chambers that were placed on the sites. The CH4 sensor was originally designed for high-concentration measurements, however, Bastviken et al. (2020) showed that using a new calibration and data processing technique the sensor can be used for CH4 fluxes in floating chambers. The CH4 sensor reacts to changes in absolute humidity, and thus measurements of relative humidity and temperature are needed to convert the voltage signal to CH4 concentration. Data logging and the air pump were controlled by an Arduino (Arduino, Italy) which is equipped with a 10-bit analog-to-digital converter, which limits the resolution. We used a 10-point moving average, applied five times to increase the resolution of data, as proposed by Kajiura and Tokida (2021). The sensor precision is lower than commercialized laser equipment, which might cause difficulties in measuring low rates of change; we quantified this uncertainty in our measurements to 2.1 ± 2.1 and 3.4 ± 3.5 µmol CH4 m−2 h−1 (see Sect. "Test of setup" below). Drift in the sensor has been shown to be within only 4–6 ppb per year for long-term measurements (Eugster et al. 2020), though this was not tested in our setup.

Each automated floating chamber measured CH4 and CO2 for a 40 min period, before a a small air pump was turned on for 20 min to ensure complete replenishment of headspace air with atmospheric air. Thus one measurement series was possible each hour (Sø et al. 2023a, b). The chamber had two hoses connected, one connected to the air pump, the other working as a pressure relief. The length and diameter (l: 1.5 m, Ø: 16 mm) of the hoses were tested to ensure no air would leave the chamber by diffusion. The air pump exchanged the air inside the chamber with atmospheric air, thus bringing the gas composition in the chambers close to that in the atmosphere and allowing continuous automated measurements of CH4 and CO2 fluxes. Chambers were tied to a sunk float that was held in place by an anchor, limiting the chambers to a one-meter float radius. All electronics were powered by an 18 Ah 4S LiFePO4 battery floating next to the chamber. All sensors were turned on for a minimum of one hour before deployment, enabling the CH4 sensor to warm up. Four chambers were distributed evenly between shallow and deep parts of each pond in areas without emergent vegetation.

Calibration of the gas sensors followed the procedure by Bastviken et al. (2020) and Sø et al. (2023a, b). All CH4 sensors were calibrated individually, as the sensors do not react the same to changing humidity and CH4 concentrations. During the calibration of the CH4 sensors, a Gasmet DX4030 FTIR Gas Analyzer (Version E1.22) and an Ultraportable Greenhouse Gas Analyzer (Los Gatos Research, USA) were used. The two CH4 analyzers were calibrated using standard CH4 gases (100 ppm).

CH4 and CO2 fluxes

Each flux measurement of CH4 was partitioned into ebullitive (FCH4-ebul) and diffusive fluxes (FCH4-diff), with their sum being the total flux (FCH4-total). Thus for each flux measurement, both FCH4-ebul and FCH4-diff could be calculated. Flux of CO2 (FCO2) only occurred by diffusion due to the high solubility of CO2 in water. All fluxes (µmol m−2 h−1) were calculated using Eq. (1).

$$F=\frac{dC}{dt}\frac{{P}_{amb}V}{{R}_{gas}TA}$$
(1)

The first term (dC dt−1) is the rate of change (µmol mol−1 h−1) in the headspace of the chamber, Pamb is the ambient pressure (Pa.), V is the chamber volume (m3), Rgas is the ideal gas constant (m3 Pa K−1 mol−1), T is the ambient temperature (K), and A is the surface area (m2) of the floating chamber. Positive fluxes indicate the gas leaving the water (source), whereas negative fluxes indicate the gas entering the water (sink).

Partitioning of FCH4-ebul and FCH4-diff was done according to Sø et al. (2023a, b). Specifically, ebullition events were characterized as a sudden increase in the headspace CH4 concentration (Fig. 2). Ebullitive events were identified by computing a 5-point running variance as sudden changes in concentration result in high variance, whereas FCH4-diff is characterized by low variance. We determined a cutoff value for the running variance by visually identifying ebullitive events and the running variance. All measurements which exceeded the running variance cutoff value were classified as ebullitive measurements. We determined the minimum and maximum concentration within this period and calculated FCH4-ebul as the difference between maximum and minimum concentration. All measurements one minute before and after the event were also included in the ebullitive event, to encompass the entire concentration change caused by an ebullitive event. Additionally, a diffusive threshold value was applied to exclude changes less than 1 ppm from being regarded as ebullitive events.

Fig. 2
figure 2

Raw data from one sensor, with black dots indicating the raw data and grey areas indicating areas of flushing the chamber air. Red lines indicate the linear model in which the slope was used to determine the diffusive exchange rates. Sudden changes in concentration were considered ebullitive events, which are shown by vertical blue lines

Full size image

FCH4-diff was calculated after the initial 200 measurements (6.5 min), i.e., the deadband, following the replacement of the chamber air, to avoid any initial instability, caused by large changes in humidity within the chamber which affects the CH4 sensor measurements. Ebullition was not determined either during this period. The next 400 measuring points (13 min) were used to calculate the slope (dC/dt in Eq. 1), using a linear regression model. If ebullitive events occurred during the 400 measuring points, only the points before the ebullitive event were used. Models with a well-defined linear fit (i.e., an R2 higher than 0.5 and more than 50 observations) were kept for calculation of FCH4-diff, while the others were discarded. All FCH4-diff were visually inspected to ensure that the headspace concentration did not approach saturation resulting in a lower slope. FCH4-ebul was calculated for the entire period after the initial 6.5 min.

FCO2 was calculated on all measuring points of each measurement period, resulting in approximately 120 measurement points over 40 min. The CO2 concentration in the chamber during the 40 min did not show any signs of saturation. A linear model was fitted for every 40 min period for each sensor. Models with an R2 higher than 0.5 and more than 50 observations were retained, and the rest were discarded.

In all, 1390 measurements of FCH4-diff and 2701 FCH4-ebul were collected during summer and 1569 FCH4-diff and 4588 FCH4-ebul measurements in winter. There were 1942 hourly CO2 fluxes in summer and 1009 in winter. The number of days with flux measurements averaged 21, 51, and 21 days per pond of continuous hourly measurements of FCH4-diff, FCH4-ebul, and FCO2, respectively. The large difference in available flux measurements is due to diffusive fluxes of CH4 and CO2 having R2 values lower than 0.5 that were discarded.

The total GHG emission from the lake was calculated in terms of CO2-equivalents (CO2e) using the conversion of 27.2 as the global warming potential for non-fossil CH4 for a time period of 100 years (IPCC 2022). The annual emissions of CO2 and CH4 were estimated as the mean of summer and winter fluxes.

Test of setup

The uncertainty of our setup was tested on two floating chambers, which were deployed in an open container filled with tap water at a depth of 20 cm. The container was initially equipped with two air pumps to ensure equilibrium in partial pressure of CH4 between the atmosphere and the water. The container was open and exposed to natural weather conditions. Furthermore, to mimic higher piston velocity, a water circulation pump was turned on after 24 h. As anticipated, increasing piston velocity did not affect fluxes in this experimental situation with equilibrium in partial pressure between the water and the gas space. A total of 83 fluxes were measured with the two floating chamber sensors. Only 5% of the fluxes had an R2 above 0.5, and 39% of the calculated fluxes were negative and 61% were positive. We assumed that no CH4 formation would take place in the container, and expected no CH4 flux between the water and the headspace in the chambers. The mean fluxes (± 95% confidence intervals) for the two sensors were low, close to and not significantly different from zero (i.e., 1.9 ± 2.1 and 4.8 ± 3.5 µmol m−2 h−1). Notably, under conditions with very high temperatures inside the chamber (> 35 °C) the CH4 sensor may need longer time (> 15 min) to stabilize, due to rapid changes in absolute humidity.

Calculations of fluxes are difficult, even with high-precision equipment. This is further complicated by the low-resolution analog-to-digital converters used here. The signal-to-noise ratio is not proportional: the noise has a larger impact in situations with low fluxes. Thus, at low fluxes, linear estimation of the concentration change over time is difficult and may results in poor model estimates, causing a bias towards higher fluxes with accepted R2 > 0.5 (Lammirato et al. 2017). This phenomenon was also noticeable in our dataset, which showed a similar inverse relationship between the diffusive fluxes of CH4 and R2 (Fig. S1).

Statistical analysis

Linear mixed models (LME) were used to determine the variables influencing FCH4-ebul, FCH4-diff, and FCO2. Prior to any statistical modeling, all variables were examined for intercorrelation using Spearman rank correlation. An arbitrary Spearman ρ range between − 0.39 and 0.39 was treated as uncorrelated and kept for analysis. Only one variable was kept if the ρ value was not within this range. Due to the large number of observations, the p-value was a poor measure of significance and was thus disregarded. To ease the interpretation of the model parameter estimates, all predictor variables were centered and scaled by subtracting the mean and dividing by the standard deviation, resulting in the same unit for all predictor variables. Sediment organic content (LOI) variables at 0–5 cm and 5–10 cm, DOC, and CDOM were found to be intercorrelated; LOI at 0–5 cm was retained for modeling.

Models were fitted using either hourly or average daily values of FCH4-ebul, FCH4-diff, and FCO2 as the response variable. As predictor variables, all models consisted of average water temperature, LOI, water depth, and plant coverage. Additionally, FCH4-ebul was used as a predictor variable in both models of FCH4-diff. For the modeling using daily values, all predictor variables were also averaged. Model selection was performed using a stepwise backward elimination approach, removing predictor variables that were not significant until all remaining variables were significant (using likelihood ratio tests, p < 0.05). LME were used to account for the hierarchical nesting of the data, using sites nested within a pond as the random effects. An additional LME was created to determine if meteorological variables affected the ratio of ebullitive to total fluxes. This was done using the data from the public weather station. The model consisted of daily mean pressure, air temperature, and wind speeds. However, as seasonal differences in temperature and wind speeds can be large, we also created two models examining the meteorological variables' effect for each season. As previous results have shown differences in pond GHG fluxes based on them being artificial or natural, we created three LME with FCH4-diff, FCH4-ebu, FCO2 as the dependent variables, respectively, and type (natural vs artificial) as the independent variable. To account for differences between ponds, we used pond as a random effect. For all models, the goodness of fit was assessed using the marginal (R2marg, fixed effects only) and conditional (R2cond, fixed and random effects) R2.

We further used LMEs to determine differences in fluxes between forest and open ponds. FCH4-ebul, FCH4-diff, and FCO2 were used as dependent variables, and type (forest vs. open) and season were used as independent variables; season was included to take into account that some of the ponds had a higher number of fluxes available.

Diel variability of GHG-flux was assessed by generalized additive mixed modeling (GAMM) for each type of flux. The relationship between fluxes (response variables) and hour of the day (predictor) were modeled using cyclic cubic spline smoothing, which was evaluated for forest and open ponds during summer and winter. Moreover, ponds and sites within ponds were incorporated to account for nested data, similar to the linear mixed models. Significant diel correlation found in GAMM models was hypothesized to be due to variable wind speeds. Thus, ponds with diel variation in fluxes were modeled with wind speeds using LME. In the models, each flux was used as a response variable and wind speed from that type of pond was used as a predictor variable; we accounted for the nestedness of data like in the previous models.

Spatiotemporal variability was examined between sites within ponds and between types of ponds (forest and open) by calculating the coefficient of variation of the daily average fluxes, described as the standard deviation divided by the mean.

All data analysis and statistics were made using R (R Core Team 2018). For the linear mixed effects model, the lme4 package was used (Bates et al. 2015), while the mgcv package was used for GAMM models (Wood 2011). R2marg and R2cond were calculated using the MuMIn package (Bartoń, 2018).

Results

Environmental conditions

All six ponds were alkaline, eutrophic, high in DOC, variable pH and with no systematic differences in water chemistry between forest and open ponds (Table 1). On average, the sediment organic contents at the 4–6 measuring sites in each pond were significantly higher in forest ponds (48.9–66.3%) than in open ponds (10.5–22.7%; t-test: df = 4, t = 6, p = 0.04). During gas flux measurements, mean water temperatures were slightly higher in open ponds than in forest ponds during both summer (23.8 and 21.1 °C, respectively) and winter (9.9 and 9.4 °C, Table 2). Two forest ponds had winter periods of hypoxia (< 10% oxygen saturation), while the other ponds were close to air saturation (Table S1). During summer, most pond sites had oxygen concentrations above air saturation during daytime and below during nighttime, but several high-plant-coverage sites in open ponds were permanently supersaturated (Table S1). One open pond (O3) had no submerged plants, while all forest ponds and one open pond (O1) had variable coverage between sites (ranging from no to high coverage) and another (O2) had full coverage (Table S1). Atmospheric pressure measurements were collected from the local weather station, showing on average a 1% difference in daily pressure from maximum to minimum, with higher average values during winter (1.3%, range: 0.2–3.1%) compared to summer (0.5%, range: 0.1–1%). The LME showed no significant relationship between FCH4-diff, FCH4-ebu, FCO2 and type (natural or artificial) of pond, indicating no differences in fluxes between natural and artificial ponds.

Table 2 Mean daily emissions of CH4 (ebullitive and diffusive) and CO2 of three forest and three open ponds in summer and winter
Full size table

CH4 flux

CH4 flux took place by diffusion (FCH4-diff) and ebullition (FCH4-ebul). FCH4-diff was usually lower than FCH4-ebul and occasionally negative, indicating diffusive uptake from the atmosphere (Fig. 3, Table 2 and Fig. S5). 13% of the diffusive fluxes showed negative values; however, the median of these values was low at -25 µmol CH4 m−2 h−1. FCH4-diff was higher in forest ponds than open ponds in both summer (range: 58–156 versus 31–53 µmol m−2 h−1, respectively) and winter (47–57 versus 23–29 µmol m−2 h−1). FCH4-diff was significantly higher in forest ponds compared to open ponds (estimate: 47.7 µmol m−2 h−1, df = 4, t value = 4.9, p = 0.007).

Fig. 3
figure 3

Average daily fluxes of ebullitive and diffusive CH4 and CO2 of each pond during summer and winter. The solid horizontal line indicates the median, upper and lower hinges indicate the 25th and 75th quartile

Full size image

FCH4-ebul varied extensively, both between sites within ponds and between ponds, and the variability was much higher than the variability of FCH4-diff and FCO2 (Fig. 3 and Table S2). Mean summer FCH4-ebul greatly exceeded winter FCH4-ebul, however, this was not significant (range: 104–1162 versus 32–118 µmol m−2 h−1, paired t-test: t = 2.1, df = 5, p-value = 0.09), and forest ponds had higher mean FCH4-ebul than open ponds (range: 62–441 versus 71–126 µmol m−2 h−1), but this difference was not significant (df = 4, t value = 1.4, p = 0.25). Higher rates of FCH4-ebul are reflected by ebullition events in 69% of the measurements in forest ponds in summer and 21% in winter, compared with 54% and 17% in open ponds during the two seasons (Fig. 4).

Fig. 4
figure 4

Average FCH4-total plotted against average FCH4-ebul: FCH4-total for each individual day for all ponds during summer and winter. Black line is fitted with a local polynomial regression model

Full size image

In summer, FCH4-diff was significantly higher around noon than during nighttime in forest ponds (GAMM, edf = 1.6, F = 4.5, p = 0.004) and showed a tendency towards higher fluxes around 6 in the morning in open ponds (GAMM, edf = 1.2, F = 1.5, p = 0.08; Fig. 5). Only forest ponds showed a significant diel pattern during winter, with highest fluxes in the evening around 18 (GAMM, edf = 2, F = 72, p < 0.001). Diel variation in wind speeds were measured in open and forest ponds during summer, which were significantly related to FCH4-diff (LME, estimate 15.4 µmol CH4 m−2 h−1 (m s−1)−1, df = 4272, p < 0.001), showing higher FCH4-diff with increasing wind speeds (Fig. 5 and S2), however, during winter, wind speed correlated negatively to FCH4-diff (LME, estimate − 3.2, df = 6189, p = 0.01). Forest ponds showed significant diel changes of FCH4-ebul, during winter (GAMM, edf = 1.7, F = 13, p < 0.001), peaking at around noon. Similarly to FCH4-diff, one LME showed a tendency in wind speeds towards FCH4-ebul (LME, estimate − 7.1 µmol CH4 m−2 h−1 (m s−1)−1, df = 6642, p = 0.06); however, the estimate was negative, indicating higher ebullitive fluxes during periods of low wind speed.

Fig. 5
figure 5

Hourly standardized fluxes of forest and open ponds during summer of FCH4-ebul, FCH4-diff and FCO2. Black lines indicate the overall mean. Grey area indicates night hours

Full size image

High FCH4-total was primarily due to ebullition, as seen in comparisons between days, sites, ponds, and seasons (Fig. 4). No ebullition took place when FCH4-total was at the lowest level (15.5 µmol m−2 h−1), and the percentage of ebullition in FCH4-total increased with higher FCH4-total, attaining a maximum of 94%, and an asymptote of 90% at approximately 400 µmol m−2 h−1. The LME containing air temperature, atmospheric pressure, and wind speeds showed a significant relationship with the daily average proportion of ebullition, with an R2marg of 0.23 and an R2cond of 0.6 when including all the data. When examining the relationship for each season, no meteorological variables had significant influence.

CO2 flux

FCO2 showed substantial variation between ponds (Figs. 3 and S5), however, no significant differences were found between forest and open ponds (df = 4, t value = 0.63, p = 0.57). Summer mean values of forest ponds ranged from 1786 to 6508 µmol CO2 m−2 h−1, while lower fluxes in open ponds ranged from − 3440 to 4198 µmol CO2 m−2 h−1. One open pond (O2) with dense aquatic vegetation took up CO2 in summer, while winter measurements revealed consistent CO2 release from all forest ponds (1414–2471 µmol CO2 m−2 h−1) and open ponds (330–3236 µmol CO2 m−2 h−1).

Maximum FCO2 in forest and open ponds in summer occurred at about 9:00, and the minimum at about 17:00, with both forest and open ponds showing a significant decline over the days (GAMM; forest: edf = 2, F = 45, p < 0.001, open: edf = 1.9, F = 26, p < 0.001, Fig. 5). A significant diel variation was also found during winter, in which higher fluxes were found in the evening or during the night, and lower fluxes during the day or early morning (GAMM; forest: edf = 1.9, F = 45, p < 0.001, open: edf = 1.9, F = 24, p < 0.001).

Drivers of CO2 and CH4 fluxes

Two models were fitted to each flux type, with the dependent variable being either hourly or daily fluxes (Table 3). Initially, all models included water temperature, oxygen concentration, sediment organic content (LOI), water depth, and plant coverage. R2marg was low for all hourly models (< 0.2), while R2cond was slightly higher (0.16–0.71), indicating that random effects accounted for some of the variation. The results from the ebullition model were particularly striking when daily rather than hourly values were used, increasing R2 values of 0.11 and 0.34–0.17 and 0.68 for R2marg and R2cond, respectively. Both R2marg and R2cond were higher in the CH4 models that applied average daily values, but this was not the case for R2marg in the CO2 model. Oxygen was positively related to fluxes in the hourly CH4 models, but not in the CO2 model. When water depth was introduced in the final model, it had a positive influence on the diffusive flux, while sediment organic content showed a negative correlation with daily FCH4-ebul and a positive correlation with FCH4-diff. There were no obvious differences in fluxes with age or natural vs. human-made ponds, however, we only had a few ponds with little differences in age, thus more research is needed to evaluate the relationship.

Table 3 Linear mixed effect models for CH4 and CO2 fluxes at hourly and daily scales at 4–6 sites and 6 ponds over several days in summer and winter. Estimates are scaled and centered, and can therefore be compared within each model
Full size table

Total greenhouse gas fluxes

On average, daily summer values of FCH4-ebul were 5.1-fold higher than FCH4-diff in forest ponds and 4.5-fold higher in open ponds (Table 2). During winter, the FCH4-ebul:FCH4-diff ratios were around 1.0 in the forest (1) and open ponds (1.2).

The mean FCO2:FCH4-total ratios (expressed in terms of moles) in forest and open ponds were lower in summer (5.5) than in winter (19.8 and 13.9, Table 2). The mean FCO2:FCH4-total ratios in forest and open ponds expressed in terms of atmospheric warming potential were below 1.0 in summer (0.55 and 0.56) and above 1 during winter (2.0 and 1.4, Table 2). Thus, the warming potential of GHG fluxes was dominated by CH4 during summer and by CO2 during winter.

Discussion

Seasonal variations in CH4 fluxes

Daily FCH4-total was sixfold higher during summer than winter in the small forest ponds and 3.9-fold higher in open ponds. However, winter fluxes were not negligible, while the increase of FCH4-diff from winter to summer was modest (1.5–twofold). Our findings suggest a marked increase in CH4 formation at higher temperatures in the sediments, directly supporting ebullition, and only a relatively lower increase of diffusive CH4 release of dissolved CH4 in the pond water. The continued CH4 release in winter agrees with data from more than 900 subarctic, boreal and temperate lakes and ponds showing CH4 accumulation under winter ice despite low temperatures (< 4 °C; Denfeld et al. 2018a, b). However, despite low winter fluxes, Ollivier et al. (2019) showed that FCH4-diff in agricultural farm ponds increased 25-fold from winter to summer. An increase from winter to summer was observed in our data, but it was far less, most likely due to higher lability and amounts of the organic material entering the ponds with higher inflow in autumn and winter.

While FCH4-ebul at our sites declined substantially during winter, it remained higher than FCH4-diff, likely because winter temperatures were not particularly low (mean: 9.4–9.9 °C). CH4 bubbles are produced even at very low temperatures under winter ice in lakes (Walter Anthony et al. 2010; Wik et al. 2011). Spatial variability of CH4 ebullition is high and localized bubbles are often present in high densities trapped in the ice (Walter Anthony et al. 2010; Wik et al. 2011). High coefficients of variation generally indicated high spatiotemporal variability of CH4 ebullition, both in this study and in an eutrophic lake examined previously (Sø et al. 2023a, b). Ebullitive release results from anaerobic degradation in the sediment and is not affected by dissolved CH4 in the water column, but rather by sediment dissolved CH4 concentration, the ambient pressure and the micro-environment (organic concent, sediment texture, production depth, etc.) where the bubbles are formed (Boudreau 2012). Previous studies have shown ebullitive events being primed by sudden drops in atmospheric pressure (Mattson & Likens 1990; Wik et al. 2013). In our study, however, we only observed modest daily changes in atmospheric pressure (0.1–3 Pa.), and only slightly higher during winter, which is why we did not test for an effect of pressure changes on ebullitive fluxes. The higher spatial variability of ebullition than diffusion of CH4 is a general pattern also observed in other studies (Bastviken et al. 2004; Sø et al. 2023a, b; Yang et al. 2020).

Measurements of FCH4-diff were occasionally slightly negative, indicating an uptake of CH4 from the floating chambers’ headspace to the water. Similar results have been reported in areas with high plant cover or high oxygen concentrations (Miller et al. 2019; Sø et al. 2023a, b). While most of our negative fluxes occurred during winter in pond O3, which had a substantial winter plant cover, most of these negative values were close to the detection limit of the floating chambers. Also, when fluxes are close to zero, the error of the regression slope of CH4 concentration versus time increases and negative values may occur by chance. Moreover, CH4 uptake occurs if an inward gradient from air to pond surface is formed due to elevated atmospheric CH4 concentration just above the ponds’ surfaces and, thus, an elevated initial CH4 concentration in the chamber’s headspace after flushing. It is known that a dense, very active microbial neuston community is formed in a narrow 1 − 100 µm layer immediately below pond surfaces, where CH4 oxidation may take place (Conrad 1996).

Relationship between CO2 and CH4 fluxes

FCH4-total decreased more than FCO2 during winter, leading to a higher CO2:CH4 molar ratio during winter (open ponds = 13.9 and forest ponds = 19.8) than in summer (open = 5.5 and forest = 5.5). This may result from the lower temperature dependence (i.e., lower Q10 values and lower activation energy) of CO2 formation than CH4 formation (DelSontro et al. 2016; Jansen et al. 2020b; Yvon-Durocher et al. 2014). Importantly, CH4 formation continued at low winter temperatures and should be addressed in the annual balances (Denfeld et al. 2018a; Karlsson et al. 2013). Quantification of winter FCO2 is even more important, as summer fluxes were only marginally higher than winter fluxes in small open ponds (ratio = 1.7) and appreciably lower in forest ponds (ratio = 1.5), where CO2 was consumed by photosynthesis during summer, but less so during winter with lower solar radiation and temperatures. In our study, daily air temperature indicated that during 2022, similar temperature conditions like the ones we measured during winter (1–7 °C) occur almost one-third of the year (summer 32%), whereas mean solar radiation similar to our winter measurements (0–80 W m−2) occurred almost 50% of the year (summer 48%).

Pathways of CH4 fluxes

In summer and winter comparisons among all sites in the six ponds higher FCH4-diff was accompanied by even higher FCH4-ebul. Thus, the proportion of FCH4-ebul increased from zero at low total CH4 release to more than 90% of FCH4-total, in accordance with our hypothesis. A likely explanation is that the diffusive CH4 release across the sediment surface only keeps pace with a modest CH4 production rate. However, when CH4 production increases at high temperatures, CH4 concentrations build up and bubbles are formed and released, bypassing possible oxidation in an oxic sediment surface and water column (Langenegger et al. 2019). This pattern was noticeable in our dataset as the summer FCH4-total were generally higher than winter total CH4 emission, most likely a result of the increased CH4 production at higher temperature. While bubble release is limited mainly by the rate of CH4 production and physical restraints before being released from the sediment (Wik et al. 2013, 2018), diffusive flux is constrained by several biological and physical processes, including oxidation in surface sediments and water column, and low gas transfer velocities at the sediment–water and air–water interface (Bastviken 2009). Our initial models showed a significant relationship with meteorological variables; however, this relationship vanished when differentiating between seasons, which suggests that it was caused by coupled differences in CH4 and meteorological variables between summer and winter.

Diel variations in CO2 and CH4 fluxes

The automated floating chambers measure GHG fluxes every hour and are particularly suited to detect possible diel patterns. In summer, FCH4-diff was significantly higher in the early morning than during the evening in forest ponds possibly related to convective mixing, as the ponds do not stratify and the air might become colder than the water during the night (Podgrajsek et al. 2014). In open ponds, a maximum was seen around noon, which correlated well with increasing wind speeds. Maximum FCO2 in forest ponds in summer occurred late in the morning following nighttime respiration, and the minimum was observed late in the afternoon after extended daytime photosynthesis. Our diel patterns show similar results to those of Miller et al. (2019) who found that diel fluxes of CH4 and CO2 from drawdown ponds can vary widely throughout the day while being dependent on water temperature, hour of the day, pH and oxygen concentration.

Overall, the diel variability of FCH4-diff was not nearly as pronounced as that observed in larger lakes, where wind exposure and gas transfer velocities are higher (Sieczko et al. 2020). If diel patterns observed in previous literature (Rudberg et al. 2021; Sieczko et al. 2020) are driven by wind shear and variable gas transfer velocities, open ponds may exert a more pronounced diel pattern as they are more susceptible to wind due to the lack of terrestrial sheltering. However, our results indicate lower CH4 fluxes in open ponds, which could restrict the detection of diel patterns because variability was low. By contrast, FCH4-ebul is a highly stochastic process independent of wind day and night and, as anticipated, it did not vary systematically during the diel cycle.

CO2 is consumed in the water by daytime photosynthesis and commonly changes CO2 uptake from the atmosphere to nocturnal release (Martinsen et al. 2022). Thus, when both CO2 and CH4 are measured in freshwater habitats, it is essential to include measurements during both day and night as gas flux often exhibits profound diel variability related to the influence of solar radiation on photosynthesis, temperature, and ventilated gas flow of CO2 and CH4 through floating-leaved plants (Brix et al. 1992; Dacey & Klug 1979; Martinsen et al. 2022).

Pond greenhouse gas fluxes

Combined, FCO2 and FCH4-total showed a significant carbon loss from ponds and a high warming potential. Assuming that the mean of summer and winter fluxes is a fair measure of annual fluxes, they attained 2453 g CO2e m−2 y−1 from the forest ponds and less, 911 g CO2e m−2 y−1, from the open ponds (Table 2). A higher mean carbon flux (i.e., 6600 g CO2e m−2) was calculated from several small ponds (< 1000 m2); however, these values are tied to large uncertainties in CH4 fluxes (Holgerson & Raymond 2016; Rosentreter et al. 2021). These ponds with very large carbon flux resemble phytoplankton primary productivity in nutrient-rich lakes (Kalff & Knoechel 1978), reflecting the high input of terrestrial and in-pond production of organic material (Mulholland & Elwood 1982).

The annual global warming potential (GWP) of released CH4 was comparable to that of CO2, representing 60% of the combined GWP of both gases in the studied forest ponds, 44% in the open ponds, and, in comparison, 37% as the average of < 1000 m2 ponds compiled by Holgerson and Raymond (2016). Previous research have shown a change in GHG emissions from ponds of different age and type (natural vs. artificial) (Abril et al. 2005; Goeckner et al. 2022; Phyoe & Wang 2019), but this was not observed for our ponds. Our ponds were mostly created to restore the natural hydrology rather than digging holes, as the artificial ponds examined by Goeckner et al. (2022) were. An initial outburst of GHG emission could be expected when sites are inundated, but our ponds are many years old and if they had experienced an initial outburst, we are likely far beyond the relevant time to measure this. On the other hand, our forest ponds might not have decreased in emissions over time due to the substantial terrestrial input they receive.

In discussions of GHG, it has been pointed out, that the re-establishment of small lakes and ponds leads to a substantial release of CO2 and CH4 to the atmosphere (Goeckner et al. 2022; Pi et al. 2022). CO2 and CH4 release is a natural consequence of the high terrestrial input of organic carbon and supersaturated CO2 relative to pond surfaces. In carbon budget evaluations, it is necessary to account for the permanent burial of organic carbon in pond sediments (Sand-Jensen & Staehr 2012). Furthermore, the origin of the carbon is important, as autochthonous carbon does not remove external carbon, which would be the case when the inputs are allochthonous (Prairie et al. 2018). However, due to the surroundings, we expect the carbon input to be primarily allochthonous in the measured ponds, although this was not tested. Annual sedimentation rates for nine small eutrophic Danish lakes as a function of maximum water depth (Anderson et al. 2014) suggest a sedimentation rate of about 550 g CO2e m−2 y−1 by extrapolation to a water depth of 1 m, typical of the ponds studied here. Direct measurements performed by digging out the organic material accumulated over a few years after the establishment of ponds in the English countryside yielded annual rates of 290–906 g CO2e m−2 y−1 (Taylor et al. 2019). Other studies have found rates varying from 29 to 1638 g CO2e m−2 y−1 (Goeckner et al. 2022; Ljung & Lin 2023). Though we assume that annual accumulation in small ponds is in the upper part of this range due to the high terrestrial input and the gradual expansion of littoral plants across the water surface, ultimately turning ponds into moors and bogs (Wetzel 1983), we apply a conservative mean annual sediment accumulation of 550 g CO2e m−2. We compared the resulting warming potential of ponds with that of drained cropland, drained unutilized land, and rewetted soils (Tiemeyer et al. 2020) that are commonly converted to ponds to increase local biodiversity and reduce downstream eutrophication and flooding risk (Schmadel et al. 2019). If estimated organic matter deposition in our ponds and measured GHG fluxes are accounted for, the annual GWP of open ponds is low (361 g CO2e m−2 y−1; Table 4), while that of forest ponds is higher (1903 g CO2e m−2 y−1). The annual GWP of rewetted organic soils (686 g CO2e m−2) resembles that of open ponds, while drained cropland and uncultivated wetland on organic soils have much higher warming potentials (3718 and 2799 g CO2e m−2) than both our pond types. The deep sediment cores (5–10 cm, data not shown) were expected to portray the surrounding soils, which in all forest ponds showed organic contents higher than 20%, while in open ponds organic content was from 10.1, 5.3 and 11.4%, the lowest content was in the deepest of ponds, which was artificially dug. We therefore believe that the comparison with Tiemeyer et al. (2020) is appropriate, as their estimation is based on organic-rich soils (12–18% C; IPCC 2006).

Table 4 Annual warming potential of CO2, CH4, and N2O exchange from three types of wetlands in Germany (Tiemeyer et al. 2020, their Table 2) compared to that of small Danish ponds in this study
Full size table

Conversion of drained cultivated and uncultivated organic soils to wetlands and ponds shifts organic carbon degradation into carbon retention and, though more CH4 is released, the carbon, nitrogen, and phosphorus retention in pond sediment and the reduction of N2O fluxes when the addition of nitrogen fertilizers ceases will markedly reduce the warming potential. In conclusion, to establish a fair evaluation of conservation and management initiatives to form new ponds to replace the many lost it is essential to account for overall GHG budgets, biodiversity and downstream eutrophication and flooding risk. We propose that everyone will benefit from the conversion of drained cultivated organic soils into natural wetlands or ponds.