Rising Water Levels and Vegetation Shifts Drive Substantial Reductions in Methane Emissions and Carbon Dioxide Uptake in a Great Lakes Coastal Freshwater Wetland

ABSTRACT

Coastal freshwater wetlands are critical ecosystems for both local and global carbon cycles, sequestering substantial carbon while also emitting methane (CH₄) due to anoxic conditions. Estuarine freshwater wetlands face unique challenges from fluctuating water levels, which influence water quality, vegetation, and carbon cycling. However, the response of CH₄ fluxes and their drivers to altered hydrology and vegetation remains unclear, hindering mechanistic modeling. To address these knowledge gaps, we studied an estuarine freshwater wetland in the Great Lakes region, where rising water levels led to a vegetation shift from emergent Typha dominance in 2015–2016 to floating‐leaved species in 2020–2022. Using eddy covariance flux measurements during the peak growing season (June–September) of both periods, we observed a 60% decrease in CH₄ emissions, from 81 ± 4 g C m⁻² in 2015–2016 to 31 ± 3 g C m⁻² in 2020–2022. This decline was driven by two main factors: (1) higher water levels, which suppressed ebullitive fluxes via increased hydrostatic pressure and extended CH₄ residence time, enhancing oxidation potential in the water column; and (2) reduced CH₄ conductance through plants. Net carbon dioxide (CO₂) uptake decreased by 90%, from −267 ± 26 g C m⁻² in 2015–2016 to −27 ± 49 g C m⁻² in 2020–2022. Additionally, diel CH₄ flux patterns shifted, with a distinct morning peak observed in 2015–2016 but absent in 2020–2022, suggesting changes in plant‐mediated transport and a potential decoupling from photosynthesis. The dominant factors influencing CH₄ fluxes shifted from water temperature and gross primary productivity in 2015–2016 to atmospheric pressure in 2020–2022, suggesting an increased role of ebullition as a primary transport pathway. Our results demonstrate that changes in water levels and vegetation can substantially alter CH₄ and CO₂ fluxes in coastal freshwater wetlands, underscoring the critical role of hydrological shifts in driving carbon dynamics in these ecosystems.

Coastal freshwater wetlands are dynamic ecosystems located at the interface between land and lakes. These wetlands are among the largest natural sources of methane, yet their response to changes in hydrology, such as lake water levels, remains poorly understood. In this study, we investigated how an increase in water depth and subsequent changes in vegetation composition affected ecosystem scale methane fluxes. We found that both productivity and methane emissions decreased substantially compared to earlier conditions. The primary drivers of the reduction in methane fluxes were enhanced oxidation in the water column and reduced plant‐mediated transport.

1. Introduction

Since the Industrial Revolution around 1760, methane (CH₄) has contributed significantly to global warming, accounting for about 30% of the temperature increase (International Energy Agency ²⁰²⁴). With a high radiative efficiency (5.7 ± 1.4 × 10⁻⁴ W m⁻² ppb⁻¹) and a relatively short lifetime (11.8 ± 1.8 years), CH₄ traps 27–30 times more heat than an equivalent amount of carbon dioxide (CO₂) over a century (Forster et al. 2021). Alternative metrics, such as the sustained flux global warming potential (SGWP), suggest warming impact of CH₄ could be up to 40% higher (Neubauer and Megonigal 2015). Therefore, reducing CH₄ emissions is an effective strategy for mitigating climate change compared to other greenhouse gases (GHGs). Vegetated wetlands are consistently identified as the largest natural sources of CH₄ emissions and a major source of uncertainty in global emissions estimates (Jackson et al. 2020, 2024; Kirschke et al. ²⁰¹³; Saunois et al. ^{2016, 2020}).

Freshwater coastal wetlands are among the highest emitters of CH₄ globally (Chu et al. 2014; Delwiche et al. ²⁰²¹; Rey‐Sanchez et al. ²⁰¹⁸) due to a combination of high productivity, persistent anoxic sediment conditions, and efficient gas transport mechanisms (Bridgham et al. 2013; Chu et al. ²⁰¹⁴; Delwiche et al. ²⁰²¹). Synthesizing data from 23 eddy covariance freshwater wetland sites, Knox et al. (2021) identified dominant physical and biological controls on CH₄ fluxes (F_CH4): temperature dominated at the seasonal scale, while temperature, water table depth, and atmospheric pressure (PA) controlled fluxes at the multiday scale. At the diel scale, latent heat flux (LE), net ecosystem exchange (NEE), and gross primary productivity (GPP) were the key predictors, highlighting the important role of vegetation as a source of labile carbon and a conduit of transport. However, how the importance of these CH₄ drivers changes with hydrological alterations remains unclear.

The intensification of the global water cycle is driving dynamic fluctuations in freshwater levels, shaping wetland configuration and spatial extent (Keddy and Reznicek 1986; Ward et al. ²⁰²⁰). This phenomenon has been observed across diverse climatic zones, from Arctic regions to tropical and temperate wetlands. In Arctic and subarctic regions, hydrological fluctuations have shaped the wetland vegetation communities (Gałka et al. 2023; Sim et al. ²⁰²¹; Zou et al. ²⁰²¹). Similar changes have been documented in tropical wetlands, including those in the Amazon basin (Hess et al. 2003; Junk et al. ²⁰¹¹), the Pantanal (Bergier et al. 2018; Ivory et al. ²⁰¹⁹), and Southeast Asia (Arias et al. 2013), where the hydrological dynamics—such as flood pulses, water storage, and inundation—have driven changes in vegetation composition, productivity, and diversity. In temperate regions, these changes are evident in Mediterranean wetlands, where hydrological shifts have altered the vegetation spatial patterns (Alvarez‐Cobelas et al. ²⁰⁰⁸), and in the Prairie Pothole Region of North America, where water‐level fluctuations have influenced wetland vegetation shifts and succession (van der Valk ^{1994, 2005}). These hydrological changes significantly alter CH₄ fluxes through two primary mechanisms: directly by affecting soil redox conditions on different timescales (e.g., Baur et al. ²⁰²⁴; Cui et al. ²⁰²⁴) and indirectly through long‐term changes in vegetation composition (Keuschnig et al. 2022; Koebsch et al. ²⁰²⁰). Climate projections suggest more frequent extreme events and greater water‐level fluctuations for inland freshwater systems (Parmesan et al. 2023), with the North American Great Lakes projected to experience increases in average annual water levels of 0.19 to 0.44 m by 2040–2049 compared to 2010–2019 (Kayastha et al. 2022). These vegetation shifts, driven by hydrological changes, are expected to intensify under future climate scenarios, potentially altering wetland ecosystems globally.

The Laurentian Great Lakes (LGL), comprising Lakes Superior, Michigan, Huron, Erie, and Ontario, form the world's largest freshwater ecosystem, with about 2,695 coastal freshwater wetlands covering approximately 215,672 ha—estimated to be about half of their original extent (Environmental Protection Agency ²⁰²⁴; Great Lakes Coastal Wetland Consortium ²⁰⁰⁴). These wetlands, hydrologically connected to the Great Lakes, experience periodic lake‐level fluctuations with significant temporal variability but no clear long‐term trend (Keough and Thompson 1999). This variability makes them ideal for investigating CH₄ emissions and their drivers in response to hydrological alterations. Such hydrological changes can significantly affect vegetation community dynamics (Geis 1985; Wilcox ²⁰⁰⁴), likely influencing plant productivity and ecosystem CO₂ fluxes. During high‐water periods, trees, shrubs, and emergent vegetation such as Typha spp. are often replaced by more aquatic species, including floating‐leaved and submerged vegetation (Keddy and Reznicek 1986; Wilcox ²⁰⁰⁴). When water levels recede, exposed mudflats stimulate the re‐establishment of sedges, grasses, and shrubs from the seed bank (Keddy and Reznicek 1986; Wilcox ²⁰⁰⁴).

While these shifts have been documented (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶; Ju and Bohrer ²⁰²²; Whyte, Francko, and Klarer ¹⁹⁹⁷; Wilcox ²⁰⁰⁴), a knowledge gap remains regarding how ecosystem shifts and altered hydrological regimes influence CH₄ fluxes and their drivers. To address this, our study investigates CH₄ fluxes and their drivers in two ecosystem states using available data from the Old Woman Creek National Estuarine Research Reserve: first, from 2015 to 2016, when Typha spp. dominated the ecosystem (Rey‐Sanchez et al. ²⁰¹⁸), and second, from 2020 to 2022, characterized by high water levels (~1 m) and dominance by floating‐leaved species. We investigated the following questions:

• How do diurnal and seasonal carbon fluxes (CO₂ and CH₄) in the floating‐leaved species compare to those in the earlier emergent vegetation state in shallower water?

• Do the environmental predictors of CH₄ exchange differ between these two ecosystem states?

2. Methods and Materials

2.1. Study Site

Old Woman Creek (OWC), located approximately 5 km east of Huron, Ohio, is one of the few remaining intact freshwater estuaries on the southern shore of Lake Erie. Designated in 1980 as both a State Nature Preserve and a National Estuarine Research Reserve (NERR), the 230‐ha reserve is co‐managed by the National Oceanic and Atmospheric Administration (NOAA) and the Ohio Department of Natural Resources (ODNR) (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶). Positioned at the southernmost point of the Great Lakes system, OWC was the first reserve of its kind within the Great Lakes and remains one of only two freshwater estuaries designated as reserves within the NERR system (NERRS).

OWC is a natural freshwater estuary with mineral‐rich sediment, located at the drowned mouth of a small tributary to Lake Erie. It drains a 69 km² watershed, primarily agricultural, originating on a till plain at an elevation of 270 m above sea level and extends 24 km, descending 96 m down to Lake Erie (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶). The sand barrier at the mouth of the wetland, formed by lake waves and littoral drift (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶), undergoes cycles of buildup and breakdown, on average isolating the wetland from Lake Erie for 42% of each year over the past decade (2014–2023). When closed, this natural barrier acts as an effective pollutant filter, trapping runoff from surrounding areas, reducing pollutants, and promoting sediment deposition before they reach Lake Erie. The estuary typically maintains a free connection with the lake, with its water exchange primarily driven by wind tides, seiches in Lake Erie, and stream drainage from the catchment (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶). During closure periods, the creek's inflow raises water levels, which are balanced by slow subsurface seepage through the barrier (Villa et al. 2019). However, during storm flows or surges, water elevation rises rapidly, and in some cases, the barrier is breached, enabling rapid water exchanges with Lake Erie to equalize the water levels until the barrier naturally reforms.

Long‐term monitoring of wetland vegetation shows a tight coupling with dynamic lake levels. Elevated water levels from the 1970s to 1999 maintained OWC wetland as an open water system dominated by Nelumbo lutea (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶; Whyte, Francko, and Klarer ¹⁹⁹⁷). When lake levels declined in 1999, emergent vegetation such as Typha spp. and Phragmites australis replaced Nelumbo spp., increasing the total cover of emergent vegetation from less than 10% to more than 45% by 2000 (Herdendorf, Klarer, and Herdendorf ²⁰⁰⁶; Trexel‐Kroll ²⁰⁰²). In recent years, rising water levels at OWC have again shifted the plant communities, with floating‐leaved species such as Nelumbo spp. and Nymphaea spp. replacing the previously dominant emergent vegetation (Ju and Bohrer 2022). During 2015–2016, the land cover in the wetland predominantly consisted of open water, Typha spp., floating‐leaved vegetation, such as Nelumbo spp. and Nymphaea spp., and mudflats. Specifically, in 2015, the coverage was 47% open water, 41.8% Typha spp., 9.7% floating‐leaved vegetation, and 1.5% mudflats (Morin et al. 2022; Rey‐Sanchez et al. ²⁰¹⁸). As water levels increased, the Typha spp. patches did not extend laterally into other areas due to the steep banks of OWC, which limit the lateral extent of vegetation zones. Instead, they disappeared entirely by 2019, and the wetland has been predominantly covered by floating‐leaved vegetation, with some patches of open water (Ju and Bohrer 2022).

2.2. Measurements of Fluxes, Micrometeorological Variables, and Water Quality

Measurements of methane flux (F_CH4) and carbon dioxide flux (F_CO2) at OWC were conducted at the US‐OWC AmeriFlux tower (Bohrer and Kerns 2024), located at 41°22′45.4″ N, 82°30′44.4″ W (Figure 1a). Sensors were mounted 3 m above the sediment surface in 2015–2016 and 6 m above in 2020–2022. Observations were made following the eddy covariance approach, using an LI‐7700 open‐path CH₄ analyzer and either an LI‐7500A open‐path CO₂/H₂O analyzer (LI‐COR, Lincoln, NE) prior to July 30, 2021, or an EC150 (Campbell Scientific, Logan, UT) from July 30, 2021 onward, coupled to a CSAT3 three‐dimensional sonic anemometer (Campbell Scientific, Logan, UT). Data were logged at 10 Hz using a CR3000 datalogger (Campbell Scientific, Logan, UT). Air temperature (TA) and relative humidity (RH) were measured using an HMP45 probe (Vaisala Inc., Vantaa, Finland), and these data were recorded every minute. Additional meteorological data, including photosynthetic photon flux density (PPFD), TA, RH, wind speed (WS), wind direction (WD), and precipitation were sourced from a NOAA NERRS‐operated weather station (41°22′39″ N, 82°30′29″ W) located about 406 m from the tower (Figure 1a). These meteorological variables were continuously recorded at a sampling frequency of 15 min. All data were averaged over each 30‐min period.

FIGURE 1.

(a) Map of the study site showing the locations of the flux tower, weather station, and two aquatic monitoring stations: lower estuary (OL) and wetland mouth (WM). The site transformed from (b) cattail‐dominated vegetation in 2015–2016 to (c) floating‐leaved species (lotus and water lily) in 2020–2022.

Water quality data, including water temperature (TW), pH, dissolved oxygen (DO), water level, salinity, turbidity, and specific conductivity, were available from an aquatic monitoring station (Lower Estuary) managed by NOAA NERRS, located at 41°22′55″ N, 82°30′51″ W, 333 m downstream from the tower (Figure 1a). These measurements were recorded every 15 min and averaged over 30 min. For this study, we utilized all data measured during the peak growing season from June to September in 2015–2016 and 2020–2022. To more accurately reflect the water level in the area surrounding the tower, the data were adjusted by adding the sonde's position above the sediment (0.26 m in 2015–2016 and 0.45 m in 2020–2022) and accounting for bathymetry differences (0.51 m). Any missing data were gap‐filled using measurements from an adjacent aquatic monitoring station (Wetland Mouth, south of State Route 6) located at 41°22′57″ N, 82°30′53″ W, 77 m from the Lower Estuary station (Figure 1a), by accounting for the differences observed before the gap.

2.3. Data Post‐processing: Gap Filling, NEE Partitioning, and Uncertainty Analysis

Fluxes and meteorological datasets were accessed from the AmeriFlux network (https://ameriflux.lbl.gov/sites/siteinfo/US‐OWC; Bohrer and Kerns ²⁰²⁴). Initial data pre‐processing, including flux calculations, was performed as detailed in Rey‐Sanchez et al. (2018). We further post‐processed the flux data to include quality control measures such as friction velocity (USTAR) filtering, gap filling, net ecosystem CO₂ exchange (NEE) partitioning, and uncertainty analysis. The solar zenith angle was calculated to differentiate between day and night periods. To enhance the quality of gap‐filled data, we used the median absolute deviation (MAD) with a z‐value of 4 to remove residual spikes from the half‐hourly F_CO2, latent heat flux (LE), and F_CH4 (Papale et al. 2006). Additionally, to correct F_CO2 measurement errors during periods of low turbulent nights, we used USTAR thresholds determined by the REddyProc package (Wutzler et al. 2018). These thresholds, calculated for each growing season (June to September), were 0.138, 0.121, 0.083, 0.089, and 0.093 m s⁻¹ for 2015, 2016, 2020, 2021, and 2022, respectively. The variability in thresholds reflects differences in surface roughness and atmospheric stability across measurement years due to changes in vegetation structure and site conditions. Observations falling below these thresholds were filtered out to ensure robust flux estimates.

Half‐hourly fluxes were gap‐filled separately for the two periods, 2015–2016 and 2020–2022, using artificial neural networks (ANNs) through MATLAB's Deep Learning Toolbox and the Statistics and Machine Learning Toolbox (The MathWorks Inc. ²⁰²⁴). There were continuous gaps in observations of about 47 days in 2015 for F_CO2, LE, and F_CH4, a 37‐day gap in 2020 for F_CO2 and F_CH4, a 36‐day gap in 2021 for F_CH4, and a 38‐day gap in 2021 for F_CO2. The gap‐filling routine followed the approach described by Knox et al. (2015, 2016), which includes the selection of representative training data, optimization of ANN architecture, and uncertainty estimation of gap‐filled data. To ensure data representativeness, explanatory variables were initially divided into a maximum of 10 clusters using the k‐means algorithm, with each cluster's data evenly split into training, validation, and testing sets, ensuring representative of the varying conditions across the dataset. This strategy minimized biases associated with unequal data coverage across different environmental conditions. Several ANN architectures were tested, ranging from simpler structure with a single hidden layer containing nodes equal to the number of explanatory variables (N), to more complex structures with two hidden layers configured with 1.5N and 0.75N nodes. Each architecture was initialized 10 times with random starting weights to circumvent potential local minima, and the initialization resulting in the lowest mean sampling error was selected. The simplest architecture that achieved a less than 5% reduction in mean squared error with additional complexity was deemed optimal. This optimized procedure was then applied across 20 random data extractions to robustly estimate the fluxes, and the median of these 20 predictions was used to fill each gap.

To gap‐fill daytime F_CO2, the explanatory variables included TA, PPFD, VPD, WS, USTAR, water level, TW, the sine (sinWD) and cosine (cosWD) of wind direction, decimal day of the year, and seasonal sine and cosine functions. For nighttime F_CO2, the same variables were used except for PPFD and VPD, as they do not have hypothetical support for influencing F_CO2 after sunset. In contrast, LE and F_CH4 were not gap‐filled with separate sets for day and night. Variables used for gap‐filling LE were the same as those for daytime F_CO2. For F_CH4, the variables included LE, TA, PPFD, VPD, WS, USTAR, water level, TW, atmospheric pressure (PA), sinWD, cosWD, decimal day of the year, and seasonal sine and cosine functions. Observed (and gap‐filled) F_CO2 during the nighttime was attributed to ecosystem respiration (ER). The best nighttime F_CO2 model, trained on observed nighttime data for gap‐filling, was used to estimate ER during the daytime. Gross primary productivity (GPP) was then calculated by subtracting the observed F_CO2 (NEE) from the modeled ER (GPP = ER − NEE).

The uncertainty procedure is similar to that described in Knox et al. (2019). ANNs were employed to estimate cumulative random and gap‐filling uncertainties in CH₄ and CO₂ flux measurements. The process began by modeling the random errors in the original measurements using a parameterized Laplace distribution, followed by Monte Carlo simulations to generate multiple realizations of the cumulative fluxes. For each half‐hourly original measured value, the random error [σ(δ)] was estimated using the residuals of the median ANN predictions. These residuals were binned by flux magnitude to parameterize a Laplace distribution, as random errors in eddy covariance fluxes follow this distribution with a standard deviation varying with flux magnitude (Hollinger and Richardson 2005). Once the Laplace distribution was parameterized, it was used in the Monte Carlo simulation to draw 100 random errors for each original measurement, which were added to the original measurements to create 100 different possible realizations of cumulative fluxes. The variance of these 100 cumulative sums provides an estimate of the uncertainty in the cumulative flux due to random errors. For each gap‐filled point, the combined gap‐filling and random uncertainty was calculated from the variance of the 20 ANN predictions. The total uncertainty is obtained by combining both random and gap‐filling uncertainties, adding their variances in quadrature, providing 95% confidence for the cumulative flux estimates.

2.4. Path Analysis

To investigate the environmental predictors of CH₄ exchange for each ecosystem state, we employed path analysis within the framework of structural equation modeling (SEM) using R Version 4.3.1 (R Core Team ²⁰²⁴) to explore the complex relationships and causal pathways. We applied the same hypothesized model for F_CH4 to both time periods, 2015–2016 and 2020–2022, and compared the models to identify differences in the environmental predictors between the two ecosystem states. The analysis was conducted using gap‐filled half‐hourly data. Although some variables were used in both the gap‐filling and the path analysis, gap‐filling was performed using a nonlinear ANN, whereas our SEM focuses on linear relationships, minimizing concerns about circular reasoning. Prior to conducting the path analysis, we performed data screening and tested several assumptions to ensure the validity of our model, following procedures similar to those used in regression analysis. Detailed methods and results are provided in the Supporting Information.

Path analysis in SEM involves five logical stages: model specification, identification, parameter estimation, model evaluation, and model modification (Byrne 2010; Fan et al. ²⁰¹⁶; Kline ²⁰²³).

1. Model specification: We first selected the predictors based on their hypothesized direct influence on CH₄ production, consumption, and transport, as listed in Knox et al. (2021), which are theoretically and empirically supported. These predictors included GPP, TA, TW, water level, USTAR, and PA. We added VPD, which does not directly influence F_CH4 but has a strong effect on vegetation and was identified by Knox et al. (2021) as one of the key predictors of F_CH4 in many wetlands. We also included sinWD and cosWD, which do not directly affect the flux but characterize the flux footprint observed by the tower, and thus, may explain differences in observed fluxes due to site heterogeneity. Given its role as a proxy for redox conditions, we included DO in the model to explore its potential role in influencing the mechanisms driving F_CH4. NEE was added to the model, particularly because it was directly observed and showed a significant reduction over the observation period. Since NEE was partitioned into GPP and ER, we included GPP and ER as the direct effect to NEE and as indirect effects on F_CH4 through NEE to capture the full dynamics of carbon fluxes influencing F_CH4. The hypothesized model is illustrated in Figure 2, with GPP, TA, TW, water level, USTAR, VPD, PA, sinWD, cosWD, and DO hypothesized to directly affect F_CH4, while GPP and ER also indirectly affect F_CH4 through NEE.

2. Model identification: Model identification determines whether the model is overidentified, just‐identified, or underidentified by calculating the degrees of freedom (DF). Model coefficients can only be estimated in just‐identified (DF = 0) or overidentified (DF > 0) models, whereas a model with DF < 0 is underidentified and not estimable. The DF is calculated as the difference between the number of observed variances and covariances and the number of estimated parameters:

   $$\mathrm{DF}=\frac{m\left(m+1\right)}{2}-p$$

   where m is the number of observed variables and p is the number of estimated parameters. The DF for the hypothesized model is 78, indicating it is overidentified.

3. Parameter estimation: We used the “lavaan” package in R (Rosseel 2012) with maximum likelihood estimation and adopted robust (Huber–White) maximum likelihood (MLR) to address issues of non‐normality and heteroscedasticity. Prior to this, we scaled the variables to ensure comparability and improve the accuracy of parameter estimation.

4. Model evaluation: Model evaluation was based on the fit indices for the test of individual path coefficients (i.e., p value and standard error) and the overall model fit, which included the comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). The CFI and TLI are incremental fit indices that compare the fit of the hypothesized model to a baseline model (i.e., a model with the worst fit), with TLI tending to be more conservative (Xia and Yang 2019). RMSEA and SRMR are absolute fit indices that assess how far a hypothesized model is from a perfect model. RMSEA measures the misfit per degree of freedom based on covariance discrepancies, while SRMR focuses on the average residuals between observed and predicted correlations. The cutoff criteria for evaluating these fit measures are summarized in Table 1 (Hu and Bentler 1999; MacCallum, Browne, and Sugawara ¹⁹⁹⁶; Schermelleh‐Engel, Moosbrugger, and Müller ²⁰⁰³).

5. Model modification: If the initial model did not achieve an adequate fit, we would modify it by removing nonsignificant coefficients or low estimates, ensuring all changes were theoretically justified.

FIGURE 2.

The hypothesized path analysis model (a priori) for methane flux (F_CH4). Solid lines indicate direct effects, while the dashed lines indicate the indirect effects. cosWD, cosine of wind direction; DO, dissolved oxygen; ER, ecosystem respiration; GPP, gross primary productivity; NEE, net ecosystem CO₂ exchange; PA, atmospheric pressure; sinWD, sine of wind direction; TA, air temperature; TW, water temperature; USTAR, friction velocity; VPD, vapor pressure deficit; WL, water level.

TABLE 1.

Cutoff criteria for absolute fit indices indicating the quality of fit in path analysis.

Fit measure	Good fit	Poor fit
Comparative fit index	≥ 0.95	< 0.90
Tucker–Lewis index	≥ 0.95	< 0.90
Root mean square error of approximation	≤ 0.06	> 0.10
Standardized root mean square residual	≤ 0.08	> 0.10

For consistent comparison, we used the hypothesized model for both periods, as it provided the best overall fit indices for comparing 2020–2022 with 2015–2016. Additionally, a refined model was applied to the 2015–2016 data, excluding sinWD and USTAR, which slightly improved fit indices and provided a more accurate representation of the relationships driving F_CH4 during this period (Figure S6).

2.5. Statistical Hypothesis Testing

To assess the statistical significance of differences between the two periods (2015–2016 and 2020–2022), we performed a permutation t‐test using the perm.t.test function from the “Deducer” package in R (Fellows 2012). We used daily means (or sums) for the variables to avoid the lack of sensitivity in half‐hourly data and the overgeneralization of seasonal aggregates. Permutation tests are robust alternatives to traditional parametric tests and are particularly useful when assumptions of normality or homoscedasticity (equal variances) are not met, or when sample sizes are unequal (Good 2005; Pesarin and Salmaso ²⁰¹⁰). The test resamples the data 10,000 times, randomly permuting the group labels (2015–2016 and 2020–2022) to generate a distribution of mean differences under the null hypothesis of no difference. This method offers a reliable assessment of statistical significance at the 5% level while effectively controlling for type I error rates (Noguchi et al. 2021).

3. Results

3.1. Environmental Conditions

Meteorological conditions were similar between the two study periods (2015–2016 and 2020–2022), except for water level (Figure 3; Table 2). The mean water level more than doubled, increasing from 0.38 ± 0.25 m in 2015–2016 to 0.98 ± 0.07 m in 2020–2022 (p < 0.001) (Figure 3d; Table 2), reflecting higher Lake Erie levels during this period (Ju and Bohrer 2022). The lowest water level was observed in 2016 at 0.20 ± 0.04 m, while the highest was recorded in 2020 at 1.1 ± 0.11 m (Table 2). PPFD followed a similar seasonal pattern in both periods, gradually declining as each growing season progressed (Figure 3a). Cumulative PPFD was higher during 2020–2022 than in 2015–2016 (p = 0.006), while mean TA and TW were comparable between periods (TA: p = 0.87; TW: p = 0.44). VPD decreased significantly (p = 0.01), and mean PA was higher in 2015–2016 compared to 2020–2022 (p < 0.001) (Table 2). WS also decreased from an average of 1.16 ± 0.75 m s⁻¹ in 2015–2016 to 1.07 ± 0.72 m s⁻¹ in 2020–2022 (p < 0.001), with winds predominantly from the southwest in both periods (Figure S4).

FIGURE 3.

Time series of (a) daily sum of photosynthetic photon flux density (PPFD), and daily means of (b) air temperature (TA), (c) water temperature (TW), and (d) water level (WL) during the peak growing season (June–September) in 2015, 2016, 2020, 2021, and 2022.

TABLE 2.

Seasonal sum of photosynthetic photon flux density (PPFD), mean air temperature (TA), vapor pressure deficit (VPD), atmospheric pressure (PA), water temperature (TW), water level (WL), evapotranspiration (ET, calculated from latent heat flux), net ecosystem CO₂ exchange (NEE), ecosystem respiration (ER), gross primary productivity (GPP), and methane flux (FCH₄) during the peak growing season (June–September) in 2015, 2016, 2020, 2021, and 2022. Error bounds indicate one‐sided uncertainty for fluxes and standard deviations for meteorological variables.

Year	PPFD (mol m−2)	TA (°C)	VPD (kPa)	PA (kPa)	TW (°C)	WL (m)	ET (mm)	NEE (g C m−2)	ER (g C m−2)	GPP (g C m−2)	FCH4 (g C m−2)
2015	4569 ± 78	21.0 ± 4.1	0.53 ± 0.45	100.7 ± 1.06	22.5 ± 2.6	0.56 ± 0.20	411 ± 44	−249 ± 36	474 ± 246	723 ± 188	84 ± 22
2016	4489 ± 76	22.6 ± 4.3	0.72 ± 0.48	99.9 ± 0.83	24.9 ± 2.5	0.20 ± 0.04	492 ± 4	−286 ± 15	531 ± 58	817 ± 47	78 ± 6
Mean ± SD	4529 ± 57	21.8 ± 1.1	0.62 ± 0.13	100.3 ± 0.57	23.7 ± 1.7	0.38 ± 0.25	452 ± 57	−267 ± 26	503 ± 41	770 ± 67	81 ± 4
2020	5103 ± 87	21.6 ± 4.5	0.55 ± 0.41	100.0 ± 1.18	23.8 ± 3.0	1.06 ± 0.11	380 ± 9	−65 ± 20	414 ± 36	479 ± 30	33 ± 6
2021	4579 ± 79	22.2 ± 4.0	0.50 ± 0.39	99.7 ± 0.75	23.5 ± 2.5	0.92 ± 0.30	406 ± 7	28 ± 14	460 ± 34	432 ± 28	29 ± 4
2022	5029 ± 84	21.7 ± 4.2	0.66 ± 0.52	99.8 ± 0.43	23.3 ± 2.6	0.95 ± 0.20	595 ± 3	−44 ± 7	402 ± 28	446 ± 20	35 ± 0.5
Mean ± SD	4904 ± 284	21.8 ± 0.3	0.57 ± 0.08	99.8 ± 0.13	23.5 ± 0.3	0.98 ± 0.07	460 ± 118	−27 ± 49	426 ± 30	453 ± 24	32 ± 3

Comparing 2015–2016 to 2020–2022, decreases were observed in all water quality parameters: DO decreased from 5.6 ± 1.6 mg L⁻¹ to 4.1 ± 0.55 mg L⁻¹ (p < 0.001), specific conductivity from 480 ± 35 μS cm⁻¹ to 421 ± 220 μS cm⁻¹ (p < 0.001), pH from 7.5 ± 0.07 to 7.3 ± 0.03 (p < 0.001), salinity from 0.23 ± 0.20 psu to 0.20 ± 0.004 psu (p < 0.001), and turbidity from 40 ± 15 FNU to 34 ± 12 FNU (p = 0.05). Note that these changes may partly reflect the adjusted position of the sonde, which was located approximately 0.2 m higher in 2020–2022 than 2015–2016 (see Section 2.2). These shifts may also suggest a general increase in water level, driven by the rise in Lake Erie water elevation, along with more stable atmospheric conditions in the later period (2020–2022), characterized by lower wind speeds, higher humidity, and lower VPD. This substantial increase in water level, combined with stable conditions, appears to have contributed to further dilution, thereby reducing concentrations of solutes and particles, as reflected in declines in specific conductivity, salinity, and turbidity. Lower DO and pH may have resulted from higher water levels (Figure 3c) and lower GPP in 2020–2022 (Figure 4c), which limited light penetration and decreased photosynthetic oxygen production. The steady decline in DO over the growing season (Figure S5a) likely resulted from a combination of gradually increasing temperatures (Figure 3b,c), which reduced oxygen solubility and stimulated ER (Figure 4b), increasing oxygen consumption. Additionally, oxygen production may have been limited by declining PPFD (Figure 3a) via GPP (Figure 4c). Together, these factors led to sustained oxygen consumption that outpaced production. Toward the end of the growing season, as temperatures and respiration rates decreased (Figures 3b,c and 4b), oxygen solubility increased, DO began to recover—in some years reaching levels comparable to or even higher than the beginning of the season (Figure S5a).

FIGURE 4.

Time series of daily sums of (a) net ecosystem CO₂ exchange (NEE), (b) ecosystem respiration (ER), (c) gross primary productivity (GPP), and (d) methane flux (F_CH4) during the peak growing season (June–September) in 2015, 2016, 2020, 2021, and 2022. Negative NEE indicates ecosystem carbon uptake, while positive NEE indicates net carbon release to the atmosphere.

3.2. Interannual, Diurnal, and Seasonal Variations in CH₄ and CO₂ Fluxes

Daily carbon fluxes exhibited significant interannual variability, particularly in NEE, GPP, and F_CH4 compared to ecosystem respiration (ER), with more pronounced variability during 2015 and 2016 (Figure 4). In contrast, the 2020–2022 period showed less variability, with a trend toward lower CO₂ uptake (Figure 4c) and CH₄ emission (Figure 4d). Mean NEE showed significantly higher uptake during 2015–2016 (−2.19 ± 1.78 g C m⁻² day⁻¹) compared to 2020–2022 (−0.22 ± 1.04 g C m⁻² day⁻¹) (p < 0.001) (Figure 4a). Mean ER was also higher in 2015–2016 (4.12 ± 1.27 g C m⁻² day⁻¹) than in 2020–2022 (3.49 ± 0.10 g C m⁻² day⁻¹) (p < 0.001), exhibiting less variability than NEE, GPP, and F_CH4 (Figure 4). In contrast to ER, GPP was notably higher in 2015–2016, averaging 6.31 ± 2.3 g C m⁻² day⁻¹, compared to 3.71 ± 1.27 g C m⁻² day⁻¹ in 2020 to 2022 (p < 0.001). Similarly, mean F_CH4 was significantly higher during 2015–2016 at 0.66 ± 0.28 g C m⁻² day⁻¹, compared to 0.26 ± 0.16 g C m⁻² day⁻¹ in 2020 to 2022 (p < 0.001).

The F_CH4 exhibited greater diurnal fluctuations in 2015 and 2016 compared to 2020, 2021, and 2022 (Figure 5a). In 2015, F_CH4 peaked at 0.81 μmol m⁻² s⁻¹ at 8:30 a.m., decreased to between 0.59 and 0.66 μmol m⁻² s⁻¹ during the day, and increased after 7:30 p.m., remaining between 0.57 and 0.72 μmol m⁻² s⁻¹ throughout the night. Similarly, in 2016, F_CH4 peaked at 9 a.m. at 0.73 μmol m⁻² s⁻¹, decreased during the day to around 0.61–0.65 μmol m⁻² s⁻¹, then sharply declined after 5 p.m. with a minimum of 0.52 μmol m⁻² s⁻¹ at 6:30 p.m., remaining below 0.62 μmol m⁻² s⁻¹ throughout the night. In contrast, from 2020 to 2022, F_CH4 showed reduced variability and more consistent patterns across the day (Figure 5a,b). Specifically, in 2020, 2021 and 2022, F_CH4 gradually increased in the afternoon, peaking around 7–8 p.m. at 0.33, 0.29, and 0.33 μmol m⁻² s⁻¹, respectively (Figure 5a).

FIGURE 5.

Diurnal variations of methane flux (F_CH4) for (a) 2015, 2016, 2020, 2021, and 2022 and (b) averaged over 2015–2016 (emergent cattail dominance) and 2020–2022 (floating‐leaved species) during the peak growing season from June to September. Error bars represent the standard error of the mean.

NEE exhibited a distinct, photosynthesis‐driven diurnal pattern, with CO₂ uptake beginning around 7 a.m., reaching a minimum around 10–11 a.m., and remaining relatively constant until 3 p.m., forming a broad U‐shaped curve (Figure 6a,b). This indicates a prolonged period of sustained high CO₂ uptake by photosynthesis. During 2015–2016, NEE showed higher maximum rates during the day, averaging −10.4 μmol m⁻² s⁻¹, which is double that of 2020–2022, where the average was −5.2 μmol m⁻² s⁻¹ (Figure 6b). However, ER (directly observed at nighttime as NEE) was comparable across years, around 4 μmol m⁻² s⁻¹, with 2015–2016 slightly higher than 2020–2022 (Figure 6a,b,e,f). Additionally, mean NEE during 2015–2016 turned positive half an hour later compared to 2020–2022 (7:30 p.m. vs. 7 p.m.) (Figure 6b), indicating the later onset of reduced photosynthesis in the evening.

FIGURE 6.

Diurnal variations of (a, b) net ecosystem CO₂ exchange (NEE), (c, d) gross primary productivity (GPP), and (e, f) ecosystem respiration (ER). (a, c, e) show the variations for 2015, 2016, 2020, 2021, and 2022, while (b, d, f) present the averages for NEE, GPP, and ER, respectively, over 2015–2016 (emergent Typha dominance) and 2020–2022 (floating‐leaved species) during the peak growing season from June to September. Error bars represent the standard error of the mean.

GPP exhibited a similar pattern across all years (Figure 6c), mirroring the inverse of the NEE pattern, with a peak around midday that persisted for several hours (Figure 6a). The diurnal GPP cycle for 2020, 2021, and 2022 was relatively consistent with similar magnitudes (Figure 6c). However, 2016 showed a larger amplitude, with a maximum photosynthetic rate of 15 μmol m⁻² s⁻¹ compared to 13 μmol m⁻² s⁻¹ in 2015. On average, the maximum photosynthetic rate during 2015–2016 (14 μmol m⁻² s⁻¹) was approximately double that of 2020–2022 (8 μmol m⁻² s⁻¹) (Figure 6d). The diurnal pattern of ER across all years was similar to that of NEE, with a minimum during the day (Figure 6e). Generally, the ER in 2015–2016 (3.5–4.3 μmol m⁻² s⁻¹) was slightly higher than that of 2020–2022 (2.8–4.0 μmol m⁻² s⁻¹) (Figure 6f), with the minimum respiration rate was also being higher in 2015–2016 (3.5 μmol m⁻² s⁻¹) compared to 2020–2022 (2.8 μmol m⁻² s⁻¹) (Figure 6f).

The wetland consistently emitted CH₄ throughout the study period, both during 2015–2016 when cattails dominated and during 2020–2022, when only floating‐leaved species were present. Cumulative F_CH4 during the peak growing season from June to September was: 81 ± 4 g C m⁻² in 2015, 78 ± 6 g C m⁻² in 2016, 33 ± 6 g C m⁻² in 2020, 29 ± 4 g C m⁻² in 2021, and 35 ± 0.5 g C m⁻² in 2022 (Table 2; Figure 7a). Over this period, seasonal CH₄ emissions decreased by 60%, from 81 ± 4 g C m⁻² during 2015–2016 (when cattails dominated) to 32 ± 3 g C m⁻² during 2020–2022 (with floating‐leaved species). The cumulative NEE for the study period was: −249 ± 36 g C m⁻² in 2015, −286 ± 15 g C m⁻² in 2016, −65 ± 20 g C m⁻² in 2020, 28 ± 14 g C m⁻² in 2021, and −44 ± 7 g C m⁻² in 2022 (Table 2; Figure 7b). On average, this represents a 90% reduction in net CO₂ uptake, decreasing from −267 ± 26 g C m⁻² during 2015–2016 to −27 ± 49 g C m⁻² during 2020–2022.

FIGURE 7.

Cumulative sum of (a) methane flux (F_CH4) and (b) net ecosystem CO₂ exchange (NEE) for 2015, 2016, 2020, 2021, and 2022 during the peak growing season from June to September at Old Woman Creek, a Great Lakes coastal freshwater wetland in Ohio. Shaded areas represent the 95% confidence intervals of the cumulative sums.

3.3. Path Analysis of CH₄ Flux

To understand the changes in factors influencing F_CH4, we compared the path analysis models for 2015–2016 and 2020–2022 (Figure 8). In the refined 2015–2016 model after excluding sinWD and USTAR (Figure S6), cosWD had the largest effect on F_CH4 (β = −0.43), followed by positive effects from TW (β = 0.29) and GPP (β = 0.26). Water level also had a positive effect on F_CH4 (β = 0.21), while other variables, such as TA (β = 0.14) and VPD (β = −0.12), exerted relatively weaker effects. Compared to the initial hypothesized model (Figure 8a), excluding sinWD and USTAR in this refined model led to a slight improvement in fit indices, with the effect of cosWD becoming stronger (from −0.35 to −0.43), while other variables exhibited minimal changes (Figure S6).

FIGURE 8.

Path diagram illustrating the effects of environmental variables and CO₂ fluxes on methane flux (F_CH4) using half‐hourly gap‐filled observations during the peak growing season (June–September) for (a) 2015–2016 and (b) 2020–2022. Blue lines indicate positive effects, while red lines indicate negative effects. Solid and dashed lines with single‐headed arrows represent direct and indirect causal paths, respectively, with all standardized path coefficients being significant (p < 0.05). Model fit indices are shown below the path diagram: comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). cosWD, cosine of wind direction; DO, dissolved oxygen; ER, ecosystem respiration; GPP, gross primary productivity; NEE, net ecosystem CO₂ exchange; PA, atmospheric pressure; sinWD, sine of wind direction; TA, air temperature; TW, water temperature; USTAR, friction velocity; VPD, vapor pressure deficit; WL, water level.

In the 2020–2022 model (Figure 8b), PA exerted the strongest (and negative) influence on F_CH4 (β = −0.55), followed by cosWD (β = −0.24) and water level (β = −0.19). Positive effects on F_CH4 were attributed to VPD (β = 0.18) and TW (β = 0.16). Comparing the two periods, we observed several significant changes in factors influencing F_CH4: PA's negative effect intensified substantially, from β = −0.09 in 2015–2016 to β = −0.55 in 2020–2022, while water level shifted from a positive effect (β = 0.21) in 2015–2016 to a negative effect (β = −0.19) in 2020–2022. Additionally, GPP decreased markedly from β = 0.2 to β = 0.01, and VPD shifted from a negative effect (β = −0.12) to a positive effect (β = 0.18). cosWD consistently exerted a substantial negative impact on F_CH4 in both periods, indicating consistently higher F_CH4 from the relatively shallow area of the tower footprint.

4. Discussion

4.1. Mechanisms Causing CH₄ Emission Decline Between Ecosystem States

Ecosystem‐scale CH₄ fluxes are regulated by microbial processes that control CH₄ production and oxidation, as well as by transport mechanisms, all of which can vary with hydrological changes. Here, we argue that microbial communities and processes in the sediment likely remained stable between the two investigated ecosystem states. Previous research at our site revealed stable porewater CH₄ concentration profiles across different vegetation types, indicating no clear differences in CH₄ production and oxidation in the sediment (Villa et al. 2020). Field campaigns conducted between 2013 and 2018 further showed that while flooding altered soil redox conditions, it did not significantly affect the transcriptional profiles of methane‐cycling microorganisms (Oliverio et al. 2024). Methanogens remained stable across both surface and deeper layers, suggesting steady CH₄ production despite fluctuations in water levels. Similarly, no major changes in methanotroph activity were observed on the sediment surface, where methanotrophs were predominantly clustered (Oliverio et al. 2024).

Assuming stable CH₄ production in the sediment across both periods, the observed decline in CH₄ emissions during 2020–2022 can be attributed to increased residence time for CH₄ oxidation in the water column due to higher water levels. The extended residence time allowed methanotrophs more opportunity to consume CH₄ before it reached the water surface (Xiao et al. 2013; Zhao, Wu, and Zeng ²⁰¹³), despite lower dissolved oxygen (DO) observed in 2020–2022 compared to 2015–2016. We hypothesize that beyond a certain threshold—quickly reached even under shallow inundation—DO levels do not significantly affect CH₄ production rates. This is supported by findings showing substantial CH₄ production in the shallower, mildly aerobic soil layers at our site (i.e., the CH₄ paradox; Angle et al. ²⁰¹⁷). In addition, Hassett et al. (2024) reported a switch‐like behavior in a shallow upstream cove of Old Woman Creek (OWC), where CH₄ flux was low when water table was below ground, but increased markedly with soil inundation, regardless of water depth or DO levels. As hydrostatic pressure rose, CH₄ bubbles released from the sediment were more likely to dissolve in the water column rather than escaping directly to the atmosphere (Iwata et al. 2020; McGinnis et al. ²⁰⁰⁶). This suppression of ebullition further limited CH₄ emissions, particularly in deeper waters where bubbles have longer ascent times and greater opportunities to dissolve (Bansal et al. 2020; Villa et al. ²⁰²¹) and become oxidized by methanotrophs. This aligns with previous research at the same site in 2018 (between our measurement periods of 2015–2016 and 2020–2022, a transitional phase from cattail to open‐water dominance in OWC), which found that CH₄ fluxes (diffusive and ebullitive) were higher in emergent vegetation patches (Typha spp.) than in floating‐leaved vegetation, but ebullition nonetheless dominated throughout the wetland (Villa et al. 2021).

Another change in the CH₄ emissions and drivers between the two ecosystem states likely stems from changes in plant‐mediated CH₄ transport. Floating‐leaved species, such as lotus and water lily, exhibited higher CH₄ conductance per leaf area compared to Typha spp., but they have much lower leaf area per unit ground area (Villa et al. 2020). Thus, floating‐leaved species are, overall, less efficient at transporting CH₄ to the atmosphere per unit ground area, particularly as CH₄ fluxes, Typha leaf area, and CH₄ conductance peaked later in the season (Villa et al. 2020). This pattern indicates the effectiveness of Typha spp. in the total CH₄ flux, as diel F_CH4 showed a pronounced morning peak in 2015–2016, which was not evident in 2020–2022 (Figure 5), as also observed in Villa et al. (2020). Emergent species such as Typha spp. depend more on pressurized flow and stomatal regulation, leading to higher but more variable CH₄ emissions, especially under fluctuating environmental conditions (Ge et al. 2024; Vroom et al. ²⁰²²). In contrast, floating‐leaved species rely on large aerenchyma spaces that facilitate a more consistent CH₄ transport pathway throughout the day. This difference in CH₄ transport mechanisms underscores how plant structure and physiological traits between emergent and floating‐leaved species shape CH₄ dynamics in wetlands. As Typha patches disappeared, CH₄ conductance through plants declined, and the relative importance of CH₄ transport by bubbling and diffusion increased.

In addition to the vertical transport processes discussed earlier, lateral carbon inputs from the wetland's hydrological regime could also influence CH₄ emissions patterns. Although the periodic hydrological decoupling of the wetland from Lake Erie limits the lateral carbon exchanges between the wetland and lake, changes in the lateral movement of carbon—particularly through the inflow of dissolved organic carbon (DOC) from nearby river systems—could influence the wetland's overall carbon balance. However, Onyango (2024) found that while DOC concentrations at the OWC wetland's inflow increased with river stage, elevated DOC levels within the wetland did not significantly contribute to CH₄ emissions. This suggests that vertical processes such as plant‐mediated transport and methanotrophic oxidation dominate over lateral transport in controlling the temporal and spatial variability of CH₄ emissions. However, long‐term (seasonal to annual) changes in DOC load may influence the interannual variation in CH₄ flux. The interaction between lateral carbon movement and vertical fluxes may be more influential for long‐term carbon sequestration and export dynamics. Overall, the substantial decline in CH₄ emissions during 2020–2022 can be mainly attributed to the combined effects of reduced aerenchyma transport, lower ebullition fluxes, and extended time for CH₄ oxidation in the water column (Figure 9).

FIGURE 9.

Conceptual diagram illustrating the transition between two ecosystem states in a coastal freshwater wetland in the Great Lakes region: cattail‐dominated (Typha spp.) in 2015–2016 and floating leaved species (Nelumbo spp. and Nymphaea spp.) in 2020–2022. The upper section depicts the four dominant predictors of methane flux (F_CH4) identified by the path analysis for each period, excluding footprint drivers (Figure 8). These predictors include water temperature (TW), gross primary productivity (GPP), and water level (WL), and vapor pressure deficit (VPD) in 2015–2016, and atmospheric pressure (PA), WL, WT, and VPD in 2020–2022. Positive and negative signs indicate the direction of each predictor's effects on F_CH4. The lower section summarizes the key mechanisms driving high CH₄ emissions in 2015–2016 and their reduction in 2020–2022, including changes in WL, CH₄ oxidation in the water column, ebullition, and plant‐mediated CH₄ transport.

4.2. Shifting Controls on F_CH4 Between Ecosystem States

The shift from Typha‐dominated wetlands in 2015–2016 to floating‐ leaved species in 2020–2022 significantly altered the factors controlling methane fluxes (F_CH4). Path analysis revealed that, in 2015–2016, cosine of wind direction (cosWD) was the strongest predictor of F_CH4 (β = −0.43). Southerly winds (cosWD = −1), originating from Typha patches located to the south (Rey‐Sanchez et al. ²⁰¹⁸), brought higher F_CH4 to the tower, while northerly winds (cosWD = 1), primarily from open water to the north, resulted in lower observed emissions (Figure 8). In addition to wind direction, GPP had a significant positive effect on F_CH4 (β = 0.26). This influence likely manifests either as a direct effect—by supplying carbon substrates for methanogens (e.g., root exudates, root mortality, plant residues) that fuel methanogenesis (Hatala et al. 2012; Knox et al. ²⁰²¹; Määttä and Malhotra ²⁰²⁴; Whiting and Chanton ¹⁹⁹³)—or as an indirect indicator, reflecting denser Typha population and, therefore, greater CH₄ transport through aerenchyma (Knox et al. 2021). The spatial heterogeneity in land cover, as reflected in the influence of wind direction, may have also strengthened the GPP–F_CH4 correlation by creating localized zones of higher emissions within the footprint. Water temperature (β = 0.29) positively influenced F_CH4 by enhancing microbial CH₄ production (Turetsky et al. 2014; Yvon‐Durocher et al. ²⁰¹⁴) and increasing CH₄ diffusion and ebullition rates (DelSontro et al. ²⁰¹⁶; Xun et al. ²⁰²⁴). Water level also had a positive effect on F_CH4 in 2015–2016 (β = 0.21), when the site was shallow, likely by reducing the spatial extent of locations where the soil was exposed and where CH₄ production was very low (Hassett et al. 2024). As reported by Zhao et al. (2023), greater inundation depths relative to very shallow states promoted taller plant growth in emergent species, facilitating more efficient CH₄ transport through larger aerenchyma channels. However, this plant‐mediated CH₄ emissions effect plateaued or slightly declined at water depths above 30 cm.

In 2020–2022, the controls on F_CH4 shifted toward atmospheric conditions and transport mechanisms, with atmospheric pressure (PA) emerging as the strongest predictor (β = −0.55), suggesting that ebullition became the dominant CH₄ transport mechanism (Sachs et al. 2008; Tokida, Miyazaki, and Mizoguchi ²⁰⁰⁵; Tokida et al. ²⁰⁰⁷; Zhao et al. ²⁰²¹). The cosWD remained a significant negative predictor (β = −0.24), though its effect was weaker compared to 2015–2016, reflecting inherent site heterogeneity. This suggests that, despite a more homogenous vegetation cover, areas within the wetland with relatively deeper water (north of the tower) exhibited lower F_CH4, likely due to an increased potential for CH₄ oxidation. Unlike the earlier period, water level had a negative effect on F_CH4 (β = −0.19), suggesting that higher water level reduced CH₄ emissions, likely by extending CH₄ retention time in the water column, providing greater potential for microbial consumption (Lei et al. 2019; Xiao et al. ²⁰¹³; Zhao, Wu, and Zeng ²⁰¹³). Additionally, periods of lower water level, such as during breaches in the barrier beach connecting the wetland to Lake Erie, may have reduced hydrostatic pressure, resulting in higher F_CH4 due to increased ebullitive fluxes. VPD also became a positive predictor (β = 0.18), suggesting that drier conditions or humidity gradients increased CH₄ transport through convective flow in floating‐leaved species, driven by external humidity gradients (Dacey 1981; Grosse ¹⁹⁹⁶). In contrast, VPD had a negative relationship with F_CH4 during the Typha‐dominated period (β = −0.12), likely due to stomatal closure limiting CH₄ transport through pressurized flow (Vroom et al. 2022), which may explain the observed morning peak in CH₄ emissions (Figure 5). Notably, GPP had a minimal effect in 2020–2022 (β = 0.01), unlike its stronger effect in the earlier period, highlighting a diminished linkage between macrophyte productivity and CH₄ emissions. This likely resulted from a reduced correlation between vegetation heterogeneity and F_CH4 heterogeneity within the observation footprint. In the earlier years, high emitting patches were dominated by Typha spp., which had higher GPP than other patch types. In the later years with a deeper wetland, floating leaves interspersed with open water were more evenly spread throughout, resulting in a more spatially uniform GPP footprint (also apparent in the observed reduction in interdaily variability of GPP and NEE). Meanwhile, the spatial heterogeneity of F_CH4 remained associated with depth, causing the F_CH4 footprint to become decoupled from the vegetation.

4.3. CH₄ and CO₂ Emissions Across Ecosystem States

The CH₄ emissions at Old Woman Creek (OWC) during 2015–2016 were the highest recorded in the FLUXNET‐CH4 synthesis studies (Delwiche et al. 2021; Knox et al. ²⁰²¹), likely due to the influence of agriculture runoff (Rey‐Sanchez et al. ²⁰¹⁸). During the growing season of 2015–2016, CH₄ emissions at OWC (81 ± 4 g C m⁻²) were 1.5–8 times higher than the annual CH₄ emissions reported in other studies of Typha‐dominated, restored or constructed wetlands (Franz et al. 2016; Knox et al. ²⁰¹⁵; Morin et al. ²⁰¹⁴) (Table 3).

TABLE 3.

Comparison of seasonal (S) or annual (A) methane (CH₄) emissions and net ecosystem CO₂ exchange (NEE) using the eddy covariance technique from other studies with similar vegetation types. Wetlands are listed by vegetation type, starting with emergent vegetation followed by floating‐leaved vegetation. Within each vegetation type, sites are ranked in ascending order based on CH₄ emissions.

Site name, country	Wetland type	Study period	Water level (m)	Dominant vegetation	CH4 (g C m−2)	NEE (g C m−2)	References
Olentangy River Wetland Research Park, USA	Constructed urban wetland	April 2011–April 2013	Continuously or intermittently flooded	Typha spp.	8.0 (A) 5.1 (S)	−310 (A)	Morin et al. (2014)
Polder Zarnekow, Germany	Restored peatland/rewetted fen	May 2013–May 2014	0.36–0.77	Typha latifolia	9.9 (A)	204 (A)	Franz et al. (2016)
Twitchell Wetland West Pond, USA	Restored wetland	August 2012–August 2013	0.26	Schoenoplectus acutus, Typha spp.	38 (A)	−397 (A)	Knox et al. (2015)
Mayberry Wetland, USA	Restored wetland	March 2012–March 2013	1.1	Schoenoplectus acutus, Typha spp.	53 (A)	−368 (A)	Knox et al. (2015)
Old Woman Creek, USA	Freshwater marsh	2015–2016 (June–September)	0.38 ± 0.25	Typha spp.	81 (S)	−267 (S)	This study
Old Woman Creek, USA	Freshwater marsh	2020–2022 (June–September)	0.98 ± 0.07	Nelumbo spp. and Nymphaea spp.	32 (S)	−27 (S)	This study
Polder Zarnekow, Germany	Restored peatland/rewetted fen	May 2013–May 2014	0.36–0.77	Ceratophyllum demersum, Lemna minor , Spirodela polyrhiza (submerged and floating macrophytes)	39 (A)	43 (A)	Franz et al. (2016)
Winous Point North Marsh, USA	Freshwater marsh	March 2011–March 2013	0.2–0.6	Nymphaea odorata, Nelumbo lutea	43 (S) 50 (A)	−164 (S) −21 (A)	Chu et al. (2014)

By 2020–2022, a shift from emergent cattails (Typha spp.) to floating‐leaved species (Nelumbo spp. and Nymphaea spp.) driven by higher water levels resulted in a 60% reduction in CH₄ emissions. The seasonal CH₄ emissions during this period (32 ± 3 g C m⁻²) were comparable to those observed at Winous Point North Marsh (43 ± 8 g C m⁻²) in nearby Lake Erie's Sandusky Bay, which was also dominated by similar floating‐leaved plants (Nelumbo spp. and Nymphaea spp.) (Chu et al. 2014), and to the annual CH₄ emissions from an open water area in Polder Zarnekow in Germany (39 ± 0.1 g C m⁻²), which was dominated by submerged and floating macrophytes (a different vegetation type) (Table 3). Nevertheless, CH₄ emissions during this period remained relatively high compared to other wetland types such as rice paddies, fens, bogs, and swamps (Delwiche et al. 2021; Knox et al. ²⁰²¹).

Despite the substantial CH₄ emissions, OWC sequestered carbon in its sediments, with long‐term storage ranging from 24.1 to 85.4 g C m⁻² year⁻¹, averaging 42.3 ± 2.7 g C m⁻² year⁻¹ (Villa et al. 2023). While CO₂ uptake via GPP decreased by 41%, CO₂ efflux (ER) remained high and stable (2020–2022: 426 ± 30 g C m⁻² vs. 2015–2016: 503 ± 41 g C m⁻²), resulting in a 90% reduction in net CO₂ uptake (NEE), making the system nearly CO₂ neutral (−27 ± 49 g C m⁻²). This is somewhat similar to the findings at the Winous Point North Marsh, a site that became a net C source under specific climatic and hydrological conditions (Chu et al. 2015). In deeply inundated wetlands dominated by floating vegetation, GPP and ER were nearly balanced (this study; Chu et al. ²⁰¹⁵), indicating that external carbon inputs may be necessary to sustain ER and CH₄ emissions while continuing carbon sequestration (Chu et al. 2015). The high rates of F_CH4 in OWC were likely driven by significant inputs of dissolved and/or particulate organic carbon and other nutrients from the upstream agriculture‐dominated watershed (Rey‐Sanchez et al. ²⁰¹⁸). These findings underscore the influence of hydrological shifts and potential external carbon sources on carbon emissions at OWC, highlighting the need for further research to confirm these drivers.

4.4. Implications for Wetland Modeling

Land surface models (LSMs) continue to face various challenges in accurately predicting CH₄ fluxes across wetlands (Forbrich et al. 2024; Yazbeck and Bohrer ²⁰²³). These challenges primarily stem from the inherent complexity of wetland ecosystems, where interactions between hydrology, plant functional types (PFTs), and CH₄ transport mechanisms create spatial and temporal variability that is difficult to accurately capture in models (Bridgham et al. 2013; Mitsch and Gosselink ²⁰¹⁵). In coastal wetlands, particularly those around Lake Erie, these difficulties are further compounded by rapid vegetation shifts in response to fluctuating water levels (Keough and Thompson 1999; Wilcox ²⁰⁰⁴).

Most current LSMs lack systematic differentiation between wetland types, leading to significant variations in how wetland‐specific traits are represented (Forbrich et al. 2024). While models of northern wetlands have included certain plant functional types such as mosses and graminoids to simulate peatland carbon dynamics (Frolking et al. 2010; Qiu et al. ²⁰¹⁹; Wania et al. ²⁰¹³), the systematic representation of PFTs and microbial communities across diverse wetland types remains underdeveloped. The dynamic shifts in PFTs and associated C cycle processes at our site highlight the need for LSMs to not only include PFTs but also incorporate criteria to model their relative occurrence and impacts on carbon fluxes under changing hydrological conditions (LaFond‐Hudson and Sulman ²⁰²³).

High‐resolution remote sensing imagery has shown potential for tracking vegetation transitions, though it primarily serves as an external tool for prescribing these transitions (Yazbeck et al. 2024). However, integrating inundation dynamics into LSMs to capture fine‐scale variability remains challenging, limiting the model's ability to accurately simulate real‐time changes in CH₄ fluxes (Forbrich et al. 2024; Yazbeck et al. ²⁰²⁴). Another promising approach is to develop empirical relationships between water levels and species occurrence, with parameterization dependent on data from intensively studied sites like OWC NERR (Morin et al. 2022). Ultimately, given the dynamic lake‐level fluctuations, thresholds need to be defined for both drier and wetter conditions (Morin et al. 2022).

5. Conclusions

This study highlights a significant reduction in methane (CH₄) emissions and carbon dioxide (CO₂) uptake in the estuarine freshwater wetland at Old Woman Creek, located in the Great Lakes region, following an increase in water level that caused a shift from Typha‐dominated vegetation to floating‐leaved species. Between 2020 and 2022, CH₄ emissions decreased by 60%, while net CO₂ uptake declined by 90% compared to 2015–2016. The reduction in CH₄ emissions was primarily attributed to decreased plant‐mediated transport due to changes in plant structure and mechanisms, along with extended residence time for CH₄ oxidation and reduced ebullition as water levels rose. Future studies should focus on direct measurements of CH₄ production and oxidation to better understand their relative contributions.

Overall, our study provides empirical evidence and a comprehensive understanding of how altered hydrology and vegetation transitions—particularly, increases in water elevation and the shift from emergent Typha to floating‐leaved species—can lead to substantial changes in both CH₄ and CO₂ fluxes. CH₄ transport pathways, including plant‐mediated processes and ebullition, along with CH₄ oxidation in the water column, play crucial roles in regulating CH₄ emissions. However, these processes remain underrepresented in current land surface models (LSMs), limiting their ability to fully simulate CH₄ dynamics from wetlands. This underscores the need for LSMs to account for hydrology, vegetation dynamics, and microbial processes to improve the accuracy of carbon flux simulations, particularly in coastal wetlands where climatic conditions drive rapid changes in water elevation and vegetation composition.

Author Contributions

Angela Che Ing Tang: conceptualization, formal analysis, methodology, software, validation, visualization, writing – original draft, writing – review and editing. Gil Bohrer: data curation, funding acquisition, investigation, methodology, writing – review and editing. Avni Malhotra: writing – review and editing. Justine Missik: data curation, writing – review and editing. Fausto Machado‐Silva: writing – review and editing. Inke Forbrich: conceptualization, funding acquisition, project administration, resources, software, supervision, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Data Availability Statement

Fluxes and meteorological datasets are available through the AmeriFlux site (site ID: US‐OWC) at https://doi.org/10.17190/AMF/1418679. Gap‐filled flux data are archived at ESS‐DIVE and accessible at https://data.ess‐dive.lbl.gov/datasets/doi:10.15485/2500238. Water elevation, water quality, and photosynthetic photon flux density (PPFD) data for Old Woman Creek were obtained through NOAA's National Estuarine Research Reserve System (NERRS) Centralized Data Management Office (CDMO) interface at https://cdmo.baruch.sc.edu/.