Methane Gas Ebullition Dynamics From Different Subtropical Wetland Vegetation Communities in Big Cypress National Preserve, Florida Are Revealed Using a Multi‐Method, Multi‐Scale Approach

Methane (CH4) emissions from peat soils are highly variable in space and time and are influenced by changes in biogeochemical controls and other environmental factors. Areas or times with disproportionally high CH4 emissions in wetlands may develop where conditions are especially conducive for microbial processes like methanogenesis. Currently, eddy covariance methods are employed to quantify CH4 exchanges over several extensive subtropical forested wetland communities in the Big Cypress National Preserve, Florida. In this work, we investigate the importance of multi‐scale measurements to characterize CH4 ebullition dynamics from subtropical wetlands. Our approach uses a combination of gas traps, time‐lapse photography, and capacitance probes to characterize ebullition dynamics from two different wetland vegetation communities for comparison to eddy covariance CH4 measurements at the site. Ground‐penetrating radar surveys and soil sampling are used to assess differences in subsurface properties between sites that influence ebullition. Our results show that the mean measurement bias between fluxes measured in this study and the eddy covariance measurements over the same period was 10–14 times larger during the wet season when ebullition rates were greatest, than the dry season, when ebullition rates were smallest. This suggests that eddy covariance measurements may underestimate the CH4 contribution of ebullition across heterogeneous wetland vegetation communities and that the comparability of CH4 fluxes from methods varying in spatio‐temporal scale changes in response to subtropical Florida seasonality. Our work suggests that these methods can be used to complement eddy covariance measurements and improve the characterization of ebullition dynamics in subtropical wetlands.

There are three commonly employed methodologies for quantifying wetland CH 4 gas fluxes: (a) eddy covariance, (b) static chambers, and (c) gas traps.Each of these methods operate over different spatial and temporal scales and are thus sensitive to different CH 4 emissions pathways.Eddy covariance is a well-established method that uses rapid (10 Hz) wind velocity and gas concentration measurements within the atmosphere surface layer to measure the exchange of gases, like CH 4 between an ecosystem and the atmosphere (Campbell & Norman, 1998;Dyer, 1961;Tanner & Greene, 1989).The eddy covariance technique works by sampling turbulent air movements to determine average gas fluxes as the covariance between a chemical constituent mixing ratio and variations in vertical wind velocity (Baldocchi et al., 1988;Lai, 2009).Typically, eddy covariance measurements are collected on a continuous basis with measurements taken on the order of seconds to minutes.Thus, eddy covariance measurements have a high degree of temporal resolution.However, due to their large measurement footprints, which can often exceed hundreds of meters in diameter, the eddy covariance technique suffers from low spatial resolution.These factors make eddy covariance well suited to measure fluxes at the ecosystem scale via the plant-mediated transport and diffusion pathways, but smaller scale spatial variabilities and discrete ebullition events may be challenging to discern (Loescher et al., 2006;Tokida et al., 2007).
Similar to the eddy covariance technique, static chambers are typically used to investigate plant-mediated and diffusive CH 4 fluxes.However, in contrast to the eddy covariance technique, static chambers typically operate at much smaller spatial scales (<10 m 2 ), have lower temporal resolution, and are significantly cheaper to deploy.In the static chamber method, air samples from within the enclosed chamber are collected over time using a syringe and then subsequently analyzed via gas chromatography.Fluxes are then quantified as the rate of change in the concentration of a chemical constituent, like CH 4 , in the chamber over time (Lai, 2009).Static chambers have been widely applied to define CH 4 flux responses to varying environmental conditions and/or heterogeneities across plant communities (Griffis et al., 2000;Lai, 2009), but a high level of labor and time are needed by manual operators to accurately capture spatial and temporal CH 4 flux variability.Given the limitations of static chamber measurements in terms of space and time, rapid ebullitive fluxes are difficult to capture.While continuous automated chambers can sample at higher frequencies needed to capture ebullition events without manual labor, they are more expensive and require a high level of maintenance and site infrastructure, like a constant electrical source.Moreover, since automated chambers are continuously deployed, they may disturb the soil microenvironment by modifying the temperature, moisture, air flow, and diffusion gradients under the chamber (Davidson et al., 2002;Yu et al., 2013).
In contrast to the eddy covariance and static chamber methods, the gas trap method is specifically designed to capture ebullition events through direct measurements of gas bubble accumulations within a closed chamber.Gas traps typically consist of an inverted funnel connected to a transparent closed chamber with a sampling port to evacuate the chamber using a syringe.Samples taken from the chamber can then analyzed using gas chromatography.Similar to static chambers, individual gas traps typically cover a small spatial area (<1 m 2 ) and require manual sampling, which limits their temporal resolution.However, when multiple gas traps are deployed and paired with time-lapse cameras, significant gains in spatial and temporal resolution can be achieved such that individual ebullition events can be observed across heterogenous landscapes and related to varying environmental conditions.Due to their simplicity and low cost, gas traps have been widely deployed to measure ebullitive fluxes from lakes (e.g., Bastviken et al., 2004;Casper et al., 2000;Vachon et al., 2020), northern peatlands 10.1029/2023JG007795 3 of 18 (e.g., Stamp et al., 2013;Strack et al., 2005), and subtropical peatlands (e.g., Comas & Wright, 2012;Sirianni et al., 2023;W. Wright & Comas, 2016).Despite their wide usage and demonstrated success, the gas trap method suffers from several limitations that act as a source of error and uncertainty.For example, gas concentrations within the chamber can change between sampling campaigns due to factors like equilibration with underlying water, bacteria-mediated production or consumption of gases, and plant growth in the chamber that contributes additional non-ebullitive gas transport.
While numerous greenhouse gas studies in peat soils have been conducted over the past three decades, these studies predominantly employ single techniques which limit their analysis to a specific spatial and temporal scale.Thus, studies comparing CH 4 fluxes across varying spatial and temporal scales of measurement are highly valuable, yet are limited in tropical and subtropical wetland environments (Krauss et al., 2016).In this study, we hypothesize: (a) ebullitive gas dynamics may vary between two locations in BCNP due to differences in vegetation and soil types and (b) flux estimates may vary when different spatial and temporal measurement scales are considered.To estimate ebullitive gas fluxes from soils in cypress forest and cypress scrub vegetation communities, we utilized a combination of gas traps, time-lapse photography, and capacitance probes and tested their performance against ecosystem scale eddy covariance measurements.Differences in soil type between sites were investigated using field-and laboratory-based ground-penetrating radar (GPR) measurements and soil sampling to characterize variability in soil thickness, subsurface porosity, and organic matter content, which is hypothesized to vary between the two vegetation communities.Our results have implications for (a) characterizing variability in fluxes below the spatial resolution of eddy covariance methods, (b) refining CH 4 flux estimates in subtropical peat soils, and (c) improving eddy covariance data processing techniques to include ebullition.

Study Area
Located in southwestern Florida, BCNP is a 729,000-acre collection of subtropical wetlands that features extensive cypress-dominated and sawgrass communities that dry and flood annually in response to rainfall (Muss et al., 2003).Distinct wet and dry seasons are a consequence of the semipermanent subtropical anticyclone over the North Atlantic basin which drives subtropical seasonality in weather patterns (Parker et al., 1955).The typical wet season occurs from May to October when rainfall, temperature, humidity, and solar radiation are greatest.During October, the dry season generally begins which brings lower rainfall, temperature, humidity, and solar radiation conditions.This subtropical seasonality of weather conditions, along with their associated changes in plant phenology, influence carbon cycling in BCNP.
Two areas were selected for this study due to their pre-existing eddy covariance infrastructure; they are called Site 1 (25.8221,−81.1017;Ameriflux Site ID: US-TCS) and Site 2 (25.8136, −80.9078;Ameriflux Site ID: US-DCS) (Figure 1a).In this study, the source area for eddy covariance measurements was defined as the radial distance surrounding the tower that likely contributes greater than 90% of total flux measurement based on calculations derived from Schuepp et al. (1990) and conducted for the Site 1 and Site 2 by Shoemaker et al. (2011).The Site 1 source area extends 800 m radially from the tower base and is mainly comprised of dense cypress forest (Taxodium distichum) 18-25 m in height with a mixed hardwood subcanopy and various varieties of grasses, sedges, and forbs that give rise to organic peat soils found here (Figure 1b).Within the source area of Site 1, two representative locations of the dominate vegetation classification were chosen with Site 1a and Site 1b located proximally and distally from the tower, respectively (Figure 1b).The Site 2 source area extends 400 m radially

10.1029/2023JG007795
4 of 18 from the tower base and is dominated by cypress scrub which is characterized by scattered cypress (Taxodium distichum) 4-10 m in height and a sawgrass (Cladium jamiacense) understory less than 1 m in height, with distinct small and medium sized areas of tall dense cypress forest (Figure 1c).Site 2a is representative of cypress scrub with scattered and stunted cypress trees and a dense sawgrass understory.Extensive periphyton mats occur at Site 2a which give rise to the marl soils found in throughout cypress scrub areas.Site 2b is similar to Site 1a and 1b however the cypress trees are shorter, thinner in diameter, and the sawgrass understory is denser in comparison.In contrast to Site 2a, Site 2b does not have any visible periphyton mats and is characterized by the presence of organic peat soils.All study sites are underlain by topographically irregular limestone from the late Miocene-early Pliocene Tamiami Formation.

Geophysical Site Characterization
Ebullitive CH 4 gas dynamics can be influenced by factors such as soil thickness variability and subsurface porosity.To investigate this influence at Site 1 and Site 2, a geophysical site characterization was conducted which included field-and lab-based ground-penetrating radar (GPR) measurements and soil sampling.One field-based GPR survey was conducted at each study site using a 160 MHz HDR shielded antenna in the common offset mode.GPR is a minimally invasive geophysical method that uses a transmitting antenna that emits an electromagnetic wave and a receiving antenna that records the sequence of reflections generated from the contrasting dielectric permittivity of subsurface stratigraphy (Neal, 2004).Profiles collected in the common offset mode, where a constant distance between antennas is maintained, record the two-way travel time (in nanoseconds) of the transmitted electromagnetic wave to various stratigraphic reflectors.Electromagnetic wave velocity can be estimated from common offset data by considering the diffracted events recorded in the GPR profiles (Conyers, 2013).Based on the increasing path length as distance increases away from a point scatterer, a diffraction hyperbola fitting method was used to estimate velocity (v).With this (v) the bulk dielectric permittivity constant (ε b ) was modeled using Equation 1, where c is the speed of light in a vacuum, or 3.0 × 10 8 m s −1 .
By modeling v as ε b , we can use it as an input into the Complex Refractive Index Model (CRIM) (Equation 2), a petrophysical mixing model used here to model porosity by assuming fully saturated conditions.
where ε w , ε s , and ε a are the dielectric permittivity of gas, water and soil respectively,  is porosity, S w is the water saturation with values between 0 (unsaturated) and 1 (saturated), and α is a factor that accounts for the orientation of the electrical field and the geometrical arrangement of minerals, with values between −1 and 1; 0.5 is used in this research (Mount & Comas, 2014).Three peat soil cores were collected, using a Russian peat corer, and qualitatively analyzed in the field along each common offset GPR profile to constrain the interpretation of soil thickness from GPR images.Post processing of GPR data sets was done using ReflexW by Sandmeier Scientific and included (a) a "dewow" filter over a time window of 6.25 ns, (b) application of time-varying gain, (c) a bandpass frequency filter, (d) a background removal, (e) a time-zero static correction, and (f) an fk (or Stolt) migration.
To further constrain velocity estimates from the diffraction hyperbola fitting method, lab-based GPR measurements were carried out on one wetland soil monolith collected from two locations along each common offset GPR survey.In total, four soil monoliths (0.46 × 0.30 × 0.18 m) were collected following the extraction routine described in Comas and Slater (2007).First, an empty plastic container was placed on the soil surface and used as an outline for cutting the soil to match the dimensions of the plastic container (i.e., 0.46 × 0.30 × 0.18 m) using a machete.Next, and using the machete, the soil surrounding the outline was cut and pulled back in order to expose the base of the monolith.The exposed monolith base was then cut with the machete to free it completely from its surroundings and placed into the plastic container.Water from the field sites was added to the sample holders to mimic the saturated conditions and water chemistry in the field.The soil monolith was then transported to the laboratory within 5 hr of collection and stored in a temperature-controlled (25°C) environmental chamber.
While in the environmental chamber, laboratory-based GPR measurements were conducted using 1.2 GHz shielded antennas in the zero offset profile (ZOP) mode, where a transmitter and receiver antenna are placed on opposing sides of a sample at a constant separation and moved synchronously across the sample (Annan, 2009).See Sirianni and Comas (2020) for further details on a similar experimental setup for a different study.Through the known distance between the antennas and the recorded travel time of the electromagnetic wave, a velocity can be calculated.
On conclusion of the laboratory-based GPR data collection, four samples from each soil monolith were collected for porosity and organic matter content determination.Porosity was calculated by: (a) saturating a cylinder with a known volume, (b) weighting the sample at full saturation, and then (c) comparing the saturated weight to the dry weight after drying the sample in an oven at 105°C until no further changes in weight were observed (ASTM D4531-15, 2015).Organic matter content was then determined for each sample through loss on ignition tests.Samples were placed in a muffle furnace and heated at 550°C for 4 hr, cooled to room temperature in a desiccator, and weighed.Organic matter content was calculated as the difference between the oven-dry soil mass and the soil mass after combustion, divided by the oven-dry soil mass (Schulte and Hopkins, 1996).In all cases, weights were measured in quadruplicate and averaged.

CH 4 Gas Dynamics
In this study, a multi-method, multi-scale approach was used in order to better understand the influence varying spatial (scale of measurement) and temporal (sampling rate) scales have on CH 4 gas flux estimates.This approach consisted of an array of techniques including eddy covariance, gas traps coupled with time-lapse cameras, and capacitance probes.Each platform consisted of an array of two gas traps, one time-lapse camera, and one capacitance probe.As shown in Figure 2, each method is characterized by a unique measurement footprint and temporal resolution with very little overlap between them, thus showing the potential of each method for providing unique information in relation to CH 4 gas flux dynamics.
The eddy covariance method was used to compute CH 4 fluxes during this study and additional details for all the processing, filtering, and gap-filling procedures used to create this data set are fully described in Shoemaker et al. (2021).For brevity, we highlight several important considerations here.CH 4 fluxes were measured using an LI-7700 open-path CH 4 analyzer (LI-COR, Nebraska, USA) and were only conducted at the Site 2 due to (a) CH 4 analyzer cost, (b) safety issues associated with servicing instruments on tall towers, and (c) evidence that methane emissions from Site 1 were small (Shoemaker et al., 2021).Other key instrumentation installed at the Site 2 tower included sensors to measure soil temperature, air temperature, and water level.The Raw 10-Hz data were processed into mean half-hourly CH 4 fluxes following protocols which included uncertainty estimates, spiking filters, double coordinate rotations, blocked-average detrending, statistical filters, air density and oxygen corrections, and high-pass filtering (Baldocchi et al., 1988;Finkelstein & Sims, 2001;Tanner & Thurtell, 1969;Tanner et al., 1993;Webb et al., 1980).The mean half-hourly CH 4 fluxes were filtered to remove unrealistic fluxes which were defined as greater than 0.5 and less than −0.2 g C m −2 d −1 (Shoemaker et al., 2021).A local de-spike filter removed half-hourly CH 4 fluxes that were outside three standard deviations within a moving 7-day window, and a friction velocity (u*) threshold of 0.1 m s −1 was used for the Site 2 station (Shoemaker et al., 2015(Shoemaker et al., , 2021)).Roughly 17% of values were removed by the unrealistic value, local despike, and u* filters at Site 2. An artificial neural network (ANN) technique was used to gap-fill missing fluxes and estimate random error uncertainty (Anderson et al., 2016) and is fully described in Shoemaker et al. (2021).Data were processed using EddyPro software by LI-COR.
The gas trap methodology uses a water-filled graduated clear PVC pipe with an inverted funnel and cut-off sampling valve attached on opposite ends (W.Wright et al., 2018).The funnel is fixed approximately 20-30 cm above the soil surface where gas bubbles exiting the soil will enter and subsequently travel upward becoming entrapped in the clear PVC chamber (Comas & Wright, 2014;W. Wright & Comas, 2016).During a user defined sampling period, released gas entrapped in the clear PVC is removed from the gas trap using a graduated syringe and cut-off valve sampling port until the chamber was totally evacuated.The total gas volume is recorded and duplicate samples were processed in the field using a Landtec GEM 2000 portable gas chromatographer (Landtec, Colton, CA, USA) to determine CH 4 concentrations.By using a known sampling period duration, surface area of the gas trap funnel, total gas volume, CH 4 fraction of the extracted gas, and the ideal gas law, a flux measurement can be reported.
The time-lapse camera method works in combination with the gas trap method.The time-lapse camera photographs the progressive displacement of the water by gas bubbles in the graduated clear PVC gas trap chamber over time.In this study, time-lapse cameras were set to collect one photograph every 30 min during daylight hours from 0700 to 1900.Photographs from the cameras were downloaded during each gas trap sampling campaign and compiled using standard image processing techniques to record gas trap level, as indicated by the cm graduations on the clear PVC chamber.A custom calibration was then used to convert gas trap level to gas volume.By using a known sampling period duration, surface area of the gas trap funnel, total gas volume, CH 4 fraction of the extracted gas, and the ideal gas law, a flux measurement can then be reported.
Capacitance probes (i.e., Meter 5TE) measure the dielectric permittivity of bulk soil which can be used to infer the volumetric water content (VWC) after applying a custom calibration specific to soils at the site and temperature correction (Campbell et al., 2009;Cobos & Chambers, 2010).In this study, a custom calibration specific to soils in BCNP was applied (see Munzenrieder, 2016).When measurements are performed in saturated peat soils, any changes in VWC can be associated with changes in volumetric gas content (VGC).Similar techniques like time-domain reflectometry have been successfully employed to investigate the internal gas dynamics of peat soils at the laboratory (e.g., Baird et al., 2004;Kellner et al., 2006) and field (e.g., Kellner et al., 2005) scales however, to the authors' knowledge, this study presents a novel approach to estimating the internal gas content variability and gas bubble production rates in saturated peat soils using capacitance probes at the field scale.In our approach, capacitance probes were used to infer: (a) gas accumulation, shown in the probe record by progressive increases in VGC; and (b) gas releases or ebullition events, shown in the probe record as sudden decreases in VGC (i.e., gas replaced by water).However, it is important to note that capacitance probes cannot differentiate between gas accumulation and release to the atmosphere from gas migration into and out of its sampling volume.Inferred gas accumulation and release (i.e., changes in VGC) was reported as a flux using the ideal gas law and assuming an average CH 4 fraction of 0.19, 0.30, and 0.26 for Site 1a, Site 2a, and Site 2b, respectively (as measured in this study).Capacitance probes were programmed to collect data every minute and calculate an average data value for every 30-min period.

Data Analysis
The CH 4 gas dynamics data for this study were collected from July 2016 to November 2016 (Sirianni et al., 2021).During this time, nine sampling periods occurred.Sampling periods were defined by the elapsed time between field campaigns for gas trap sampling and data downloading and are reported in Table 1.It is important to note that all comparisons between the methodologies used in this study occur over the sampling periods listed in Table 1.Due to the lack of CH 4 analyzer at the Site 1 eddy covariance tower, only Site 2 data were used for the comparative analysis between methods.For this analysis, data from each method were first averaged over each of the nine sampling periods.Next, fluxes observed from the gas traps, time-lapse cameras, and capacitance probes were scaled for comparison to eddy covariance measurements by using the spatial extent of each dominant vegetation community within the eddy covariance tower's radial source area reported in Shoemaker et al., 2011 as a spatial scalar (Equation 3).
where F T is the total scaled flux, s x and s y are spatial scalars of vegetation community x and y, and F x and F y are the gas trap, time-lapse camera, and capacitance probe flux from vegetation community x and y.The dominant vegetation types at Site 2 were found to be cypress scrub (83%) and cypress forest (15%) communities (Figure 1), thus the spatial scalars were 0.83 and 0.15, respectively.In order to compare estimates from the different methods at Site 2, scaled gas fluxes from the gas traps, time-lapse cameras, and capacitance probes were plotted against the eddy covariance flux for the same period.A segmented regression analysis (Oosterbaan, 2017) was performed to assess the data set for breakpoints.If no breakpoints were present, linear regression analyses (P < 0.05) were performed to determine relationships between CH 4 fluxes measured by gas traps, time-lapse cameras, and capacitance probes and CH 4 fluxes measured by eddy covariance during the same period.The Bland-Altman method (Altman & Bland, 1983) was used to assess differences in the mean biases between CH 4 flux measurement techniques during the wet and dry season and two-sample t-tests were carried out to assess statistical significance (P < 0.05).Additional linear regression analyses (P < 0.05) were performed to assess the relation between Site 2's scaled CH 4 fluxes measurements and the average water level, air temperature, and soil temperature during each sampling period, which are also summarized in Table 1.

Geophysical Site Characterization
Results from the common offset GPR profiles are shown in Figure 3 (top greyscale images) for both Site 1 and 2 and show the contrast in subsurface radar facies between the two study areas.The diffraction-based 2D velocity model (explained below) and inferred water saturated porosity for each site is also shown, followed by an inferred lithological model.Both profiles were collected during inundated field conditions using a small hand-towed boat to float the GPR unit on top of the water and water depth at collection is illustrated by the blue segment for each core shown in Figure 3.The GPR profile at Site 1 (Figure 3a) is characterized by two laterally continuous reflectors representing a fairly consistent thickness of approximately 0.25 m that are interpreted as the peat soil layer (brown color in the model in Figure 3a) as confirmed from coring results.The change in reflection signature below this layer is characterized by increased attenuation (or lack of reflections) and is interpreted as the underlying limestone (yellow in the model).The irregular appearance of these reflectors may also be due to the irregular nature of the peat-limestone interface.In contrast, the GPR profile at Site 2 (Figure 3b) shows a broadly undulating radar facies (composed also by two main individual reflections) that begins at a depth of 0.5 m at 0 m along the profile, reaches the surface between 80 and 100 m, and then dips down to approximately 0.5 m by the end of the profile.Following the results at Site 1, these reflectors are interpreted as the marl layer (gray color in the model) matching the velocity model and confirmed from coring.The underlying limestone (yellow) bedrock shows more topographic variability here (compared to Site 1), and includes peat filled solution pipes (cm to m scale) that were observed at the surface along the profile between 80 and 100 m (e.g., Lab Sample 4 location) underlying the cypress forest vegetation community.This sharp vegetation change corresponds with areas of the GPR profile with increased chaotic reflections near the surface which become deeper beneath the cypress scrub community.Surrounding cypress scrub communities generally correspond to areas with diminished reflection  strength (e.g., 0-40 m) in the GPR profile.Marl soils associated with cypress scrub communities may contribute to the energy loss in these areas due to their higher clay content relative to the more organic rich peat soils of cypress forest areas.
One unique aspect of the common offset GPR profiles collected at both sites was the consistent presence of diffractions, which allows for the characterization of subsurface velocities by fitting diffraction hyperbolae that can be converted into relative dielectric permittivity values to ultimately infer porosity using the CRIM model (Equation 3).These diffractions were used to generate 2D velocity/porosity models as shown in Figure 3 (middle colored images overlying the GPR common offset) and show the contrast between both sites.The 2-D model at Site 1 (Figure 3a) shows a relatively uniform surficial layer (i.e., warm colors) that follows the major reflectors shown in the common offset profile.Areas of diminished limestone porosity (i.e., cool colors) also occur heterogeneously throughout the profile (e.g., 160-180 m) but they are not associated with any major changes to the surficial vegetation.In contract, the 2-D velocity model at Site 2 (Figure 3b) shows areas of diminished limestone porosity (i.e., cool colors) underlaying the Cypress Scrub community while areas of enhanced limestone porosity (i.e., warm colors) are associated with the areas of exposed caprock and Cypress Forest communities.
An important distinction to make is that the velocities reported are non-Dix velocities and do not represent absolute velocities within the subsurface.Results from laboratory ZOP GPR measurements are also shown in Figure 3 (inset boxes).The velocities determined from collected samples at both sites match well with the inferred velocities from their respective 2-D velocity model, however no limestone samples were collected for ZOP GPR measurements.
Soil samples were collected from the surface at four locations within the eddy covariance radial source area at Site 1 and Site 2. The results from the porosity and organic matter tests show distinct differences in the soil composition and physical properties between sites (results shown in Figure 3).Site 1 samples had an average porosity and organic matter content of 80% and 61% respectively whereas, Site 2 samples had an average porosity and organic matter content of 70% and 24% respectively.Additionally, Site 2 samples were highly reactive to nitric acid suggesting they are CaCO 3 -rich likely due to the abundance of calcareous periphyton mats present at this site.No strong reaction to nitric acid was observed in Site 1 samples.

CH 4 Gas Dynamics
Gas fluxes inferred from eddy covariance, gas traps, time-lapse cameras and capacitance probes during our study period are shown in Figure 4. Due to the lack of a CH 4 analyzer installed at Site 1 (Figure 4a), the eddy covariance results are limited to Site 2 (Figure 4b).The gas trap (gray boxes) results from Site 1a, Site 2a, and Site 2b are shown in Figures 4c and 4e, respectively.The reported gas trap fluxes represent an average of the two gas traps located at each platform location in Figure 1.Over the course of the study, average gas trap fluxes from the Site 1a platform (Figure 4c) show CH 4 emissions that range from 2.6 to 15.9 mg CH 4 m −2 d −1 .In comparison, average gas trap fluxes from the Site 2a platform (Figure 4d) show much larger CH 4 emissions that range from 18.8 to 214.0 mg CH 4 m −2 d −1 exceeding one order of magnitude when compared to the Site 1a platform.Average gas trap fluxes from the Site 2b platform (Figure 4e) range from 2.3 to 65.3 mg CH 4 m −2 d −1 .In both cases at Site 2, the gas trap fluxes follow a similar trend to the eddy covariance data where measurements from 7/7 to 10/3 (highlighted by blue on the x-axis) show larger CH 4 fluxes than the measurement periods from 10/3 to 11/20 (highlighted by yellow on the x-axis).Acquiring consistent data sets at Site 1b was problematic due to technical failures resulting in scattered data with very little overlap between methods and for that reason, and for the sake of brevity, are not presented here.
One of the unique aspects of the high temporal resolution of time-lapse cameras is its ability to capture fast ebullitive releases shown as individual peaks for Site 1a, Site 2a, and Site 2b in Figures 4f and 4h, respectively.The time-lapse camera results from Site 1a platform (blue bars) (Figure 4f) show that ebullition events tended to be infrequent (35 ebullition events observed) during the study period with an average and maximum flux of The results from the capacitance probes at Site 1a, Site 2a, and Site 2b are shown (dark gray bars) in Figures 4i  and 4k, respectively.Changes in gas content within the sampling volume of the probe are interpreted as gas release (i.e., ebullitive fluxes) given the shallow nature of the probes; however, it is important to consider they could represent vertical migration without release to the atmosphere.Inferred gas bubble releases at Site 1a (Figure 4i) are characterized by the smallest magnitude fluxes (3,656 inferred ebullition events observed) of all sites with an Note that dates listed on x-axis align with sampling periods listed in Table 1.
average and maximum flux of 7.8 and 289.9 mg CH 4 m −2 d −1 , respectively.In comparison, Site 2a (Figure 4j) shows the largest magnitude fluxes (3,487 inferred ebullition events observed) with an average and maximum flux of 92.8 and 4,691.8mg CH 4 m −2 d −1 , respectively.Lastly, Site 2b (Figure 4k) shows smaller fluxes compared to Site 2a (2,967 inferred ebullition events observed) with an average and maximum flux of 59.9 and 6,377.3mg CH 4 m −2 d −1 , respectively.
Capacitance probes also provide additional information on volumetric gas content and gas bubble accumulation over time that can be associated to either gas bubble production or upward gas bubble migration from deeper layers.A comparison of results from each site is shown in Figure 5.At Site 1a (Figure 5a) the volumetric gas content (gray line) relative difference ranged over 3.8% and the inferred bubble production rates (highlighted in red Figure 5) ranged from 7 to 39 mg CH 4 m −2 d −1 (average = 20 mg CH 4 m −2 d −1 ).In comparison, the volumetric gas content relative difference at Site 2a had the largest range (i.e., 7.3%) and largest inferred bubble production rates which ranged from 8 to 223 mg CH 4 mg m −2 d −1 (average = 69 mg CH 4 m −2 d −1 ).Lastly, average volumetric gas content relative difference at Site 2b ranged over 5.3% and inferred bubble production rates ranged from 10 to 56 mg CH 4 m −2 d −1 (average = 29 mg CH 4 m −2 d −1 ).

Data Analysis
In order to compare estimates from the different methods at Site 2, the scaled gas flux from the gas traps, time-lapse cameras, and capacitance probes were plotted against the eddy covariance flux for each sampling period as shown in Figure 6.A 1:1 relationship is shown as representative of a perfect match between scaled flux estimates, as measured by gas traps, time-lapse cameras, and capacitance probes, and the eddy covariance fluxes.Visually, two general trends can be inferred from the data set: (a) a subset that closely follows a 1:1 relationship between scaled gas flux and corresponding eddy covariance flux during the dry season (October to April) which are highlighted by the yellow oval in Figure 6; and (b) a subset with steeper slope when compared to the 1:1 relationship during the wet season (May to October) which are highlighted by the blue oval in Figure 6.Further investigation of these groupings using a segmented regression analysis did not statistically support a breakpoint in the data suggesting that the data are best represented by a single regression line.The results of these linear regressions are summarized in Figure 6.The gas trap linear regression produced significant (P < 0.05) and well correlated (R 2 = 0.74) results with a slope of 3.1.The time-lapse camera linear regression produced significant (P < 0.05) and well correlated (R 2 = 0.92) results with a slope of 2.6.Lastly, the capacitance probe linear regression produced significant (P < 0.05) and well correlated (R 2 = 0.76) results with a slope of 1.5.
To evaluate the influence of wet/dry seasonality on the mean measurement bias the Bland-Altman method was used and the results are summarized in Table 2.When comparing the eddy covariance fluxes with the scaled gas trap fluxes, the mean bias was 14 times larger during the wet season than the dry season and significantly different (P < 0.05).Similarly, when comparing the eddy covariance fluxes with the scaled time-lapse camera fluxes, the mean bias was 10 times larger during the wet season than the dry season and significantly different (P < 0.05).In contrast to the scaled gas trap and time-lapse camera fluxes, the scaled capacitance probe flux mean bias was not significantly different (P > 0.05) between the wet and dry season and remained elevated throughout the study.In order to investigate the relation of scaled gas fluxes from eddy covariance, gas traps, time-lapse cameras, and capacitance probes on three environmental parameters measured at Site 2, linear regression analyses were performed.The results are summarized in Table 3. Water level was inversely related with scaled CH 4 flux from each method; however, the relations were poorly correlated and not significant.In contrast, air and soil temperature were both directly and significantly related with scaled CH 4 flux from each method with moderate to high correlation.

Comparability of Methods Across Spatial and Temporal Scales of Measurement
The significant change in mean measurement bias between the scaled gas trap and time-lapse camera fluxes and the eddy covariance flux that coincides with the end of the typical wet season and start of the typical dry season in south Florida is an interesting result of this study.Similar to previous studies in other wetlands (e.g., Goodrich et al., 2011;Männistö et al., 2019), this suggests that ebullition patterns in BCNP cypress forest and cypress scrub vegetation communities are responsive to subtropical seasonality.We attribute this change in mean measurement bias pattern to the limitations of the eddy covariance method for capturing ebullition.The datapoints that plot near the 1:1 relationship (yellow oval; Figure 6) correspond to smaller less frequent ebullition events recorded by the gas traps, time-lapse cameras, and capacitance probes (i.e., 10/3 to 11/20 in Figure 4) during the dry season when methanogens are becoming less active and thus producing less CH 4 that can be subsequently release in ebullition events.On the other hand, the datapoints that deviate from the 1:1 relationship (blue oval; Figure 6) correspond to more frequent ebullition events recorded by the gas traps, time-lapse cameras, and capacitance probes (i.e., 7/7 to 10/3 in Figure 4) during the wet season when methanogens are most active and ebullition rates become significantly greater and a more important flux pathway.It is clear that having scaled flux measurements comparable to the eddy covariance in inherently dependent upon the relative sensitivity to different CH 4 emission pathways of each method.For example, while gas trap and time-lapse camera measurements are designed to target the ebullition pathway, eddy covariance measurements are sensitive to all CH 4 emissions pathways occurring within their measurement area.Short-lived ebullition events are difficult to discern using eddy covariance methods since ebullition events may not persist in air concentrations long enough over the typical 30-min measurement averaging time (Foken & Wichura, 1996;Loescher et al., 2006;Tokida et al., 2007).Moreover, the standard filtering employed in many eddy covariance processing workflows to remove high values that represent "unrealistic" fluxes may itself also contribute to the masking of ebullition events and thus measurement mismatch between methods.In many cases, static chambers may also miss rapid ebullition events since: (a) nonlinear changes in CH 4 concentrations within the chamber headspace air, which can arise due to episodic ebullition, are often excluded from analysis (Korrensalo et al., 2018), and (b) the discontinuous nature of static chamber measurements makes quantification of sporadic rapid ebullition events difficult (Sun et al., 2013).
Another important consideration is the effect of differences in spatial resolution and the representative percentages of important ecosystem components within the measurement areas of the methods being compared.These effects are exemplified well in our data set.For instance, despite operating over the same 30-min measurement interval, individual ebullition measurements from time-lapse cameras reached up to two orders of magnitude greater than eddy covariance measurements.Mismatch between time-lapse camera and eddy  covariance data may be due, in part, to the disparity in spatial resolution of each method which is 0.027 versus 502,485 m 2 , respectively.This result is consistent with previous studies on gas dynamics in subtropical peat soils that show gas content and emissions rates can vary over multiple orders of magnitude when different spatial and temporal measurement scales are considered (Comas & Wright, 2012, 2014;Mustasaar & Comas, 2017).It is also similar to the relation between scale of measurement and variability of physical properties that has been explored in many hydrological studies (e.g., Bakalowicz, 2005;Ford & Williams, 1989;Mount & Comas, 2014;Robinson et al., 2008;Worthington, 1999).Measurement mismatch may also be due to the relative amount of ecosystem complexity represented within each measurement footprint.For example, systems with higher degrees of ecosystem complexity, like the sharp differences in vegetation and soil types at Site 2, may cause fluxes measured by gas traps to be biased due to location within the eddy covariance measurement footprint and up-scaling across a representative area.It is important to note that the limited number of gas traps employed in this study may not adequately characterize flux across the eddy covariance footprint.Additionally, vegetation maps used in this study (i.e., Ruiz et al., 2019) are based on a 50 m × 50 m grid that may limit the accuracy of the up-scaling process.
To further explore the comparability between different spatio-temporal scales of CH 4 flux measurement, a compilation of results from previous wetland studies that compare gas trap and/or static chamber measurements with eddy covariance measurements are summarized in Figure 6 (i.e., open circles with letter reference; see figure caption for citations).Similar to our results, these previous studies plot in three main groupings, (a) near 1:1 relationship with eddy covariance (e.g., Hartley et al., 2015;Hendriks et al., 2010;Korrensalo et al., 2018;O'Shea et al., 2014;Schrier-Uijl et al., 2010;Yu et al., 2013), (b) >1:1 relationship with eddy covariance (e.g., Chaichana et al., 2019;Krauss et al., 2016;Meijide et al., 2011;Sun et al., 2013), and (c) <1:1 relationship (e.g., Acosta et al., 2019;Männistö et al., 2019).While some studies shown in Figure 6 (i.e., Chaichana et al., 2019;Meijide et al., 2011;Sun et al., 2013) discuss the potential influence of ebullition on their measurements, others (i.e., Hartley et al., 2015;Hendriks et al., 2010;O'Shea et al., 2014;Schrier-Uijl et al., 2010;Yu et al., 2013) do not mention or explicitly exclude (i.e., Korrensalo et al., 2018) ebullition effects in their analysis.The only study in Figure 6 to quantify ebullition (i.e., Männistö et al., 2019) used floating gas traps but found that ebullitive fluxes did not significantly contribute to the total ecosystem CH 4 emissions and were thus orders of magnitude lower than static chamber and eddy covariance measurements.In most cases, studies that do not consider rapid ebullition show a near 1:1 relationship between static chamber and eddy covariance measurements (i.e., Hartley et al., 2015;Hendriks et al., 2010;Korrensalo et al., 2018;O'Shea et al., 2014;Schrier-Uijl et al., 2010;Yu et al., 2013) suggesting that the good agreement between methods could be due, in part, to ebullition not being a distinctive component of flux, or both methods overlooking rapid ebullition.In comparison, studies that discuss the effect of ebullition (i.e., Chaichana et al., 2019;Meijide et al., 2011;Sun et al., 2013) tended to show a >1:1 relationship between static chamber and eddy covariance measurements suggesting that static chamber measurements in these studies may be capturing ebullition to some degree and thus accounting for, in part, the flux mismatch.This interpretation is further supported by visual evidence of ebullition during measurements by Sun et al. (2013), and the common and significant occurrence of ebullition in rice fields by Meijide et al. (2011) and Chaichana et al. (2019).However, studies like Krauss et al. (2016) (>1:1 relationship) and Acosta et al. (2019) (<1:1 relationship) do not fit this trend and suggest that the vast differences in the measurement footprint size, footprint composition (i.e., representative percentages of open water, bare soil, and vegetation), and sampling density between the static chamber and eddy covariance methods could account for the differences in fluxes.

Differences in Ebullition Dynamics in Relation to Vegetation Community and Soil Type
Results from this study (Figure 4) also show marked differences in ebullition gas dynamics at each study site that seem to be associated with vegetation community.When comparing cypress scrub versus cypress forest, the first showed regularly spread ebullition events (i.e., approximately every 1-2 days) that were larger in magnitude.In comparison, the cypress forest vegetation community exhibited scattered ebullition events that were  lower in magnitude.In this study, CH 4 bubble production rates were inferred in the field using a novel approach with capacitance probes.Overall, inferred bubble production rates observed in this study (shown in Figure 5) ranged from 7 to 223 mg CH 4 m −2 d −1 and are consistent with previous studies in the Everglades like, Comas and Wright (2014) and W. Wright and Comas (2016) that found CH 4 production ranged from 8 to 642 mg CH 4 m −2 d −1 and 20-470 mg CH 4 m −2 d −1 , respectively using field-scale time-lapse GPR measurements in different Everglades peat soils.Our results show clear differences in CH 4 bubble production rates between sites, with Site 2a having roughly 3 times higher average bubble production rate than Site 1a and Site 2b (i.e., 69, 20, and 29 mg CH 4 m −2 d −1 , respectively).This is also consistent with previous work in the Everglades that found marl soils may have up to a 6 times greater potential rate of methanogenesis than peat soils (Bachoon & Jones, 1992) and CH 4 production rates in periphyton can be up to 4 times higher than in detritus (A.L. Wright & Reedy, 2008).
Clear differences in gas dynamics between the CaCO 3 -rich marl soil at Site 2a and the more organic-rich peat soil at Site 1a and Site 2b suggests that differences in peat type could account for some of the observed differences in gas dynamics between study sites.This seems reasonable considering previous work in boreal systems (e.g., Baird et al., 2004;Green & Baird, 2013;Waddington et al., 1996) has ascribed hot spots for gas production, accumulation, and release to differences in peat type.Hot spots may also arise in relation to peat thickness.For instance, at Site 1 the peat-limestone interface was interpreted as laterally continuous with fairly consistent soil depths (about 0.25 m depth) along the 200m GPR transect (Figure 3) whereas the GPR data at Site 2 suggests that the marl-limestone interface is more undulating in nature (about 0.5 m depth) with areas of surficial bedrock exposure (Figure 3).This interpretation is consistent with previous work in the Everglades that found differences in peat thickness between sites helped to account for differences in observed fluxes between sites and between gas traps (Comas & Wright, 2014).
Porosity has been shown to play an important role in controlling gas dynamics in peat soils with small porosity differences producing large differences in gas storage and release (Ramirez et al., 2015(Ramirez et al., , 2016(Ramirez et al., , 2017;;W. Wright et al., 2018).In this study, Site 1a and Site 2b were found to have higher soil porosity than Site 2a which could allow for more gas to be held within the soil matrix leading to sporadic gas release.Conversely, the lower soil porosities observed at Site 2a may prevent gas from being stored as readily leading to a more continuous gas release.This suggests that some peat soils, despite having a lower total porosity, may have a more "open" matrix that allows gas to move more freely upward though the peat column while others may have a more "blocked" matrix system where gas bubbles become entrapped and build up until mobilized to the surface (W.Wright & Comas, 2016).The complex interwoven mass of widespread shallow horizontal and vertical roots associated with the cypress trees in combination with the root systems of other emergent plants at Site 1a and Site 2b may add to the tortuosity of the bubble pathways and decrease gas mobility thus acting to trap bubbles, increasing the amount of gas stored within the peat matrix.This is commonly observed in northern peatland systems where layers of highly decomposed peat or woody materials act as gas traps due to their enhancement of the peat matrix's shear strength (Comas et al., 2014;Glaser et al., 2004;Rosenberry et al., 2003).However, other studies have found the presence of roots to be linked with higher CH 4 emissions (Noyce & Megonigal, 2021) and the presence of vegetation to be linked with low porewater CH 4 concentration due to the rapid loss of CH 4 through stems (Bansal et al., 2020), highlighting context dependency and the need for more research.
Labile carbon may also help to explain the ebullition regime differentiation between the cypress scrub and cypress forest vegetation communities.For example, despite Site 2a having the lowest organic matter content (i.e., 15%) it had higher average emissions than Site 1a or Site 2b which had higher organic matter content (i.e., 61% and 33%, respectively).One explanation for this difference is that the extensive periphyton mats present at Site 2a (and not present at the other sites) may provide organic compounds that are more readily fermented for methanogenesis than others (e.g., lignin) which may be found in higher concentrations in peat derived from macrophytes (e.g., sawgrass) and woody plants (e.g., cypress trees) similar to what is present at Site 1 and Site 2b (Bachoon & Jones, 1992;Burke et al., 1988).This explanation is consistent with previous work in the Everglades (i.e., A. L. Wright & Reedy, 2008) that found CH 4 production rates in periphyton were roughly four times higher than in detritus despite periphyton having 11% less organic matter content.Thus, it is conceivable that having less concentrated but higher quality organic matter at Site 2a could account for the differences in gas production and gas release observed at each site, however a more detailed analysis of organic matter quality between sites is needed.
The association of ebullitive CH 4 fluxes to different vegetation communities and soil types offers a simple approximation for estimating total ebullitive CH 4 flux at the regional scale.To exemplify this, average fluxes from the gas trap, time-lapse camera, and capacitance probe method were upscaled using BCNP vegetation classifications from Ruiz et al. (2019) to estimate the yearly ebullitive CH 4 contribution of cypress scrub and cypress forest vegetation communities across BCNP.An average between ebullitive CH 4 fluxes measured at Site 1a and Site 2b was used in the upscaling process due to their similar vegetation and soil composition.Together, the cypress scrub and cypress forest vegetation communities were estimated to release 0.023, 0.026, and 0.030 Tg CH 4 yr −1 using the gas traps, time-lapse cameras, and capacitance probes respectively.Previous work by Bartlett et al. (1989) applied a similar approach using chamber-based CH 4 flux measurements from different vegetation communities across the Shark River Slough area of Everglades National Park and estimated a total contribution of about 0.022 Tg CH 4 yr −1 from this area.Gas traps, which operate on a similar spatial scale to chambers, are comparable in magnitude, demonstrating the potential of employing a combination of gas traps, time-lapse cameras, and capacitance probes to estimate regional CH 4 ebullition fluxes from subtropical peat soils.Assuming the greater Everglades emits 0.5 Tg CH 4 yr −1 (Burke et al., 1988), ebullition from peat and marl soils in cypress forest and cypress scrub vegetation communities of BCNP could represent between 4 and 6% of yearly CH 4 emissions using the gas traps, time-lapse cameras, and capacitance probe estimates from this study.However, this comparison approach is limited due to factors like differences in vegetation communities present and methodologies used between the studies and lacks full characterization of seasonal and spatio-temporal variability.Given these considerations, CH 4 flux measurements from the methods used in this study should therefore be viewed as complementary techniques to the eddy covariance method, rather than fully comparable, that can provide more detailed information about rapid ebullitive fluxes and the internal gas dynamics of subtropical wetland soils.

Conclusions
In this study, we used an array of methods that vary in spatio-temporal resolution to characterize spatial heterogeneities and temporal variations in the frequency and magnitude of ebullition events that are an important component of the total ecosystem CH 4 flux in subtropical wetlands.Eddy covariance measurements may not properly represent ecosystem variability within their measurement footprint when rapid ebullition events are more common.Gas traps, time-lapse cameras, and capacitance probes can reveal CH 4 flux heterogeneities between vegetation communities that are otherwise difficult to discern in the eddy covariance data record.Our work suggests that these methods can be used to complement eddy covariance to improve the characterization of CH 4 fluxes across vegetation communities in subtropical wetlands.This work has been partially supported by National Park Service U.S. DOI17-440, NOAA (GC11-337), DOE (TES 10959421), USGS (Cooperative Agreement: Carbon Dynamics of the Greater Everglades), the FAU Center for Environmental Studies Walter and Lalita Janke Innovations in Sustainability Science Research Fund and the FAU GRIP Grant.The authors would like to thank the Editor, Associate Editor, Dr. Sheel Bansal, and one anonymous reviewer for the constructive reviews towards the improvement of this manuscript.We also thank Dr. Scott Prinos for the comments provided during the review of this article.Lastly, we thank W. Barclay Shoemaker.His support and guidance throughout this project was instrumental to its success.

Figure 1 .
Figure 1.(a) Map of South Florida showing boundaries for Big Cypress National Preserve, Everglades National Park, the Greater Everglades, and study locations (white triangles) along Loop Road (black inset box with aerial imagery).(b) Vegetation map of Site 1 eddy covariance (EC) tower radial source area showing tower location (white triangle) and Site 1a and 1b platform locations (black circles).(c) Vegetation map of Site 2 EC tower measurement radial source area showing tower location (white triangle) and Site 2a and 2b platform locations (black circles).For each site, relative percent coverage of major vegetation communities are shown in table to the right of the map with total radial source area.Vegetation classifications for each site were taken from Ruiz et al. (2019).Note difference in radial scale between sites.

Figure 2 .
Figure 2. Schematic representation of typical spatial and temporal resolutions of each gas flux measurement methodology used in this study.

Figure 3 .
Figure 3. 160 MHz GPR common offset profiles (top image), 2-D velocity models (middle image and rainbow color scale), and interpreted cross sections (bottom image) along Site 1 (a) and Site 2 (b).Schematic representations of vegetation communities are shown above the profiles.Soil sampling locations (indicated by open circles) and results (reported in inset boxes) are also shown.Coring results show differences in soil type and soil-limestone interface (as indicated by colored boxes in the legend) between sites.
Half hourly CH 4 fluxes (black line) are plotted.Over the course of the study, the half-hourly CH 4 fluxes ranged from 7.1 to 82.8 mg CH 4 m −2 d −1 with an average CH 4 flux of 45.3 mg CH 4 m −2 d −1 .From 7/7 to 10/3 (highlighted by blue on the x-axis) the average CH 4 flux ranged from 51.6 to 60.4 mg CH 4 m −2 d −1 .From 10/3 to 10/15 (highlighted by yellow on the x-axis) there is a slight decrease in the average CH 4 flux to 40.5 mg CH 4 m −2 d −1 that becomes even smaller beyond 10/15 that persists through the last three measurement periods.During this time, average CH 4 fluxes are 25.7, 21.8, and 15.5 mg CH 4 m −2 d −1 , respectively.
6.1 mg CH 4 m −2 d −1 and 1,305.5 mg CH 4 m −2 d −1 respectively.In comparison, the time-lapse cameras from Site 2a platform (red bars) (Figure 4g) reveal ebullition events occurred almost 3 times more frequently (91 ebullition events observed) and with larger average and maximum fluxes of 108.3 mg CH 4 m −2 d −1 and 7,000.3mg CH 4 m −2 d −1 respectively.Lastly, time-lapse cameras from Site 2b (red bars) (Figure 4h) show less frequent ebullition events (42 observed) than the cypress scrub platform with the smallest average and maximum fluxes of 41.5 mg CH 4 m −2 d −1 and 4,499.7 mg CH 4 m −2 d −1 , respectively.

Figure 4 .
Figure 4. CH 4 gas release results from Site 1 and Site 2. (a) No eddy covariance CH 4 results at Site 1.(b) Eddy covariance 30-min mean CH 4 flux at Site 2. (c) Gas trap CH 4 results from Site 1a.(d) Gas trap CH 4 results from Site 2a.(e) Gas trap results from Site 2b.(f) Time-lapse camera results from Site 1a.(g) Time-lapse camera results from Site 2a.(h) Time-lapse camera results from Site 2b.(i) Capacitance probe results from Site 1a.(j) Capacitance probe results from Site 2a.(k) Capacitance probe results from Site 2b.Note order of magnitude differences between y axis.Colored brackets with asterisk indicate times with missing time-lapse camera data.Blue and yellow coloring along x-axis shows wet and dry season, respectively.Note that dates listed on x-axis align with sampling periods listed in Table1.

Figure 5 .
Figure 5. Volumetric gas content percent (gray line) expressed as relative difference from Day 0 and inferred CH 4 bubble production rates (highlighted in red with inset text) from capacitance probes for Site 1a (a), Site 2a (b), and Site 2b (c).Average bubble production rate (i.e., mg CH 4 m −2 days −1 ) are reported for each site.

Table 1
Summary of Sampling Periods Used in This Study That Reports Start and End Date, Average Water Depth, Average Air Temperature, and Average Soil Temperature for Each Sampling Period

Table 2
Site 2 Results From the Bland-Altman Test

Table 3
Coefficient of Determination and Significance for Site 2 Average CH 4 Fluxes From Each Sampling Period in This Study Using Each Measurement Method Versus Sampling Period Means of Water Level (WL), Air Temperature (AT), and Soil Temperature (ST)