Methane (CH4) is a potent greenhouse gas (GHG) with atmospheric concentrations that have nearly tripled since pre‐industrial times. Wetlands account for a large share of global CH4emissions, yet the magnitude and factors controlling CH4fluxes in tidal wetlands remain uncertain. We synthesized CH4flux data from 100 chamber and 9 eddy covariance (EC) sites across tidal marshes in the conterminous United States to assess controlling factors and improve predictions of CH4emissions. This effort included creating an open‐source database of chamber‐based GHG fluxes (
Tidal wetlands play a critical role in global carbon (C) cycling (Bianchi, 2006;Odum, 2002). They have the potential to provide major feedbacks to the Earth's climate system as they exchange greenhouse gasses (GHGs) with the atmosphere, store large soil C pools, and have the potential to sequester C through continuous vertical accretion, allochthonous sediment deposition, and biomass accumulation (Duarte et al., 2013). Low rates of organic matter decomposition in their waterlogged soils promote the preservation of large quantities of soil organic C (also known as blue carbon) for centuries to millennia, contributing to the long-term removal of carbon dioxide (CO 2 ) from the atmosphere (Chmura et al., 2003).
However, the anaerobic conditions that promote soil C storage also lead to microbial methane (CH 4 ) production (Megonigal et al., 2004). Methane, with 32-45 times the warming potential of CO 2 over a 100-year time horizon, is second behind CO 2 in contributing to increases in atmospheric GHGs associated with recent warming (Forster et al., 2021;Neubauer & Megonigal, 2015).
When considered in the timeframe relevant to meeting the Paris Agreement's target of limiting warming to 1.5°C, the CH 4 global warming potential is even higher, reaching 75 times that of CO 2 (Abernethy & Jackson, 2022). A recent compilation of global CH 4 emissions identified wetlands as among the primary natural sources of atmospheric CH 4 , second only to freshwaters, with emissions ranging between 150 and 180 Tg CH 4 year -1 (Saunois et al., 2020) and likely contributing to growing atmospheric CH 4 concentrations as the climate becomes warmer and wetter (Zhang et al., 2023), and anthropogenic wetland modifications increase (Kroeger et al., 2017;Rosentreter, Borges, et al., 2021). Global atmospheric CH 4 contributions from coastal wetlands, like tidal marshes, have been less studied (Rosentreter, Borges, et al., 2021), and recent bottom-up estimates (1-3.5 Tg CH 4 year -1 ; Saunois et al., 2020) are not well constrained due to the lack of systematic observations, data quality (i.e., primarily discrete measurements with few high-frequency continuous measurements), uncertainties associated with coastal wetland area, and the risk of double counting of ecosystem types (e.g., tidal and non-tidal riverine, floodwater or estuarine wetlands) (Rosentreter et al., 2023;Rosentreter, Borges, et al., 2021;Roth et al., 2022;Saunois et al., 2020).
Global-scale controls on wetland CH 4 dynamics are reasonably known, including climatic zones, the presence of permafrost, peat, or mineral soils, groundwater, and surface water inputs in addition to precipitation, and the influence of salinity (Bridgham et al., 2013;Turetsky et al., 2014). These large spatial-scale characteristics subsequently control plant composition and the soil attributes that drive anaerobic C cycling and CH 4 dynamics. However, in tidal wetlands, where the tidal influence can dominate over other global factors, it remains unclear whether CH 4 responses to commonly studied predictors in non-tidal freshwater wetlands also apply. Tidal wetlands experience unique spatial and temporal variation in tides, redox conditions, and the influence of seawater (Cloern & Jassby, 2012;Seyfferth et al., 2020). These variations contribute to spatial gradients in plant species and traits, microbial communities, and processes such as soil accretion, primary production, respiration, and decomposition (Borde et al., 2020;Campbell & Kirchman, 2013;Morris et al., 2002;Watson & Byrne, 2009).
Any or all of these factors substantially influence CH 4 emissions, highlighting the complexity of predicting CH 4 fluxes from these ecosystems.
On a local scale, rates of methanogenesis in wetland soils are primarily governed by the balance of electron donors and terminal electron acceptors. In general, saline tidal wetlands have high concentrations of porewater sulfate, typically leading to low rates of methanogenesis as acetate and hydrogen, primary substrates for methanogens, are utilized by sulfate reducers to decompose organic matter anaerobically (Megonigal et al., 2004). However, other pathways like methylotrophic methanogenesis can also be important in saline environments where non-competitive substrates are degraded to methyl compounds, producing CH 4 even when sulfate reduction co-occurs (Oremland et al., 1982;Seyfferth et al., 2020).
Once CH 4 is produced, it can reach the atmosphere by physical (diffusion and ebullition) and plant-mediated processes. Some plants efficiently vent CH 4 to the atmosphere through aerenchymatous tissue. For example, in species such as Phragmites australis, Typha latifolia, and T. angustifolia, convective gas transport during light conditions allows CH 4 produced in soils to bypass CH 4 oxidation zones and escape to the atmosphere at greater rates (Bendix et al., 1994;Sanders-Demott et al., 2022;Van der Nat & Middelburg, 1998;Vroom et al., 2022).
and tidal height controlled variability at diel and seasonal scales. At the diel scale, GPP preceded temperature in importance for predicting CH 4 flux changes, while the opposite was observed at the seasonal scale. Water levels influenced the timing and pathway of diel CH 4 fluxes, with pulsed releases of stored CH 4 at low to rising tide. This study provides data and methods to improve tidal marsh CH 4 emission estimates, support blue carbon assessments, and refine national and global GHG inventories.
contiguous United States, eddy covariance, flux chamber, methane, open-source database, predictors, synthesis, tidal wetlands Previous syntheses have identified important drivers of CH 4 emissions in tidal wetlands, including salinity and sulfate concentrations (Poffenbarger et al., 2011), temperature, and the quality and quantity of organic matter (Al-Haj & Fulweiler, 2020). Additionally, tidal pumping (Call et al., 2015;Trifunovic et al., 2020), dominant vegetation types and hydrology (Derby et al., 2022), plant phenological phases (Vázquez-Lule & Vargas, 2021), and functional trait composition (Mueller et al., 2020;Tong et al., 2018;Van der Nat & Middelburg, 1998) have been shown to play significant roles. While recent syntheses (Al-Haj & Fulweiler, 2020;Rosentreter, Al-Haj, et al., 2021;Rosentreter, Borges, et al., 2021) have qualitatively discussed these biogeochemical (e.g., salinity, temperature, organic matter) and biotic (e.g., plant-mediated transport) drivers, they have not resolved the relative influence of overlapping predictors in a quantitative fashion. This may be partly attributed to the reliance of syntheses on literature values, which often represent summarized flux averages or temporally upscaled estimates that can mask a wealth of processes and important variations evident in original disaggregated measurements. In contrast, this study focuses on original, disaggregated data and metadata obtained directly from researchers, allowing for a more detailed and in-depth analysis of CH 4 fluxes and their influencing factors.
Measuring CH 4 emissions from tidal marshes has primarily depended on chamber methods, especially static chambers sampled manually, due to their low cost, simplicity of application, and ease of deployment in areas without power (Rolston, 1986;Yu et al., 2013). The fine spatial resolution of chamber flux measurements, their generally low detection limits, and their compatibility with concurrent surface and porewater sampling of relevant analytes (salinity, pH, dissolved gas concentrations, or alternate electron acceptors) have made them an invaluable resource for process-based research and for measuring small fluxes and instances of net CH 4 oxidation. However, due to logistical constraints, chamber methods introduce limitations as they encourage sampling at low tide, during daylight hours, and on an intermittent sampling schedule (typically monthly).
As a result, pulse events triggered by diurnal and tidal cycles, changes in atmospheric pressure, or sediment disturbances are often unobserved or filtered out to remove ebullition (Altor & Mitsch, 2006;Morin et al., 2014;Podgrajsek et al., 2014;Rosentreter et al., 2018).
To address the low resolution of daytime static chamber measurements, some studies have employed custom gas exchange models that incorporate continuous air temperature or water table depth, among other factors, to model CH 4 fluxes between sampling events (Krauss et al., 2016;Neubauer, 2013;Sanders-Demott et al., 2022;Schultz et al., 2023;Weston et al., 2014). In the absence of continuous measurements of predictor variables, others have used scaling factors to convert average daily fluxes into annual emissions (Bartlett & Harriss, 1993;Bridgham et al., 2006;Poffenbarger et al., 2011). These factors are based on studies that report both annual and daily rates and represent the ratio of annual CH 4 flux to average daily flux (Bridgham et al., 2006).
On the contrary, the continuous, high-frequency eddy covariance (EC) technique offers promising datasets for understanding tidal marsh CH 4 fluxes, which often involve nonlinear and asynchronous processes across multiple timescales (Reid et al., 2013;Sturtevant et al., 2016). However, EC studies in tidal marshes are less common than chamber studies and are not as widespread as in other ecosystems, such as inland freshwater wetlands, rice paddies, and tundra. Unlike chambers, EC systems are less effective in discerning the influence of factors operating at small spatial scales, such as plant traits and microtopography, and often lack supporting water level (Knox et al., 2019), salinity (Delwiche et al., 2021), et al., 2024). Details on the data acquisition process and database structure can be found in Supporting Information sections 1 and 2.
For the analysis in this study, we selected chamber plots without any experimental treatment, except for those involving salinity. These treatments consisted of slight changes from fresh to oligohaline conditions, resembling the seasonal variations in salinity that tidal marshes may naturally experience. This resulted in a total of 100 marsh sites with discrete CH 4 flux measurements disaggregated by sampling event, spanning from 1 day to 4 years, and accompanied by ancillary data, including porewater biogeochemistry. Analyses of the chamber dataset were complemented with nine independent EC tidal marsh datasets available through FLUXNET-CH 4 (US-LA1, US-LA2, US-Srr), AmeriFlux (US-EDN, US-Dmg, US-StJ), and towersite PIs (US-MRM, US-HPY, and US-PLM) (Table S1), all of which adhere to standards and data QA/QC procedures explained elsewhere (Chu et al., 2023;Delwiche et al., 2021;Knox et al., 2019). Gap-filled fluxes were obtained from site PIs if missing in the original datasets.
Likewise, when absent, water quality parameters, including conductivity or salinity and dissolved nitrate concentrations, were sourced from the USGS National Water Information System or National Estuarine Research Reserves online databases (NOAA, 2023;USGS NWIS 11180770, 2023). See Supporting Information section 1 and Table S1 for details on tower sites, including fraction of gap-filled CH 4 flux data.
Eddy covariance and chamber sites were classified using a multifaceted classification system described in Table 1 and the Supporting Information section 2. The definition of "site" differs between chamber and EC studies. Eddy covariance sites are represented by the flux footprint rather than by gradients in biotic and abiotic factors such as elevation, species composition, or substrate type (channel, mudflat, or vegetated area). In this study, we defined chamber sites as distinct locations within a tidal wetland that could consist of several chamber plots. Adjacent areas, if composed of different wetland vegetation, salinity, elevation, or disturbance classes, were considered different chamber sites in our database. 2). These flags indicate if any thresholds (R 2 and/or p-value) were used to test, modify, or remove flux rates in the original studies. 65% of the submitted CH 4 fluxes were accompanied by R 2 and/or p-values (24% had both reported), and 35% had no associated statistics. Chamber flux measurements across studies in the synthesis varied in methodology. Some were based on syringe headspace samples, while others used continuous gas analyzers, which provide a more accurate assessment of concentration changes over time. To harmonize data quality control procedures across the dataset for analysis in this study, we filtered the data using percentiles (to remove outliers) and flux quality flags (Table 2). For the entire dataset, CH 4 fluxes below and above the 1st and 99th percentiles were excluded if statistics supporting these extreme values were not reported or if the R 2 of the regression was lower than 0.90 (Figure S2a). Furthermore, CH 4 flux rates flagged as "Not Significant" or "Not Tested" with reported statistics were further filtered out based on p-values, R 2 , and the number of sample points (n) used to fit the regression model, all conditional on flux magnitude. Briefly, CH 4 flux rates with a p-value >0.10 were filtered out if they exceeded 10 nmol m -2 s -1 (the mean flux rate estimated for fluxes flagged as not significantly different from zero). If p-values were not provided but R 2 and n were available, we filtered out fluxes exceeding 10 nmol m -2 s -1 , those with R 2 < 0.80 and n < 4, or R 2 < 0.55 and n < 6. Rates <10 nmol m -2 s -1 were retained even if they did not meet the R 2 requirements because samples with low CH 4 concentrations generally have low R 2 values reflecting the low fluxes from those chambers rather than a poor measurement quality. Of 8980 individual chamber CH 4 fluxes, 156 were filtered out because they exceeded the 1st and 99th percentiles with no supporting statistics. An additional 60 measurements were discarded based on p, R 2 , and n values (Figure S2b). The result is a vetted dataset without unsubstantiated outliers of 8764 observations. We acknowledge that fluxes driven by nonlinear processes such as ebullition may be underrepresented.
We estimated annual tidal marsh CH 4 fluxes where there was a full year of tower or chamber data. At EC sites, we calculated annual sums using gap-filled data. At chamber sites, we used published estimates where available. At sites where CH 4 fluxes had been measured across all months but annual estimates had not been reported, we integrated daytime fluxes using linear interpolation between year-round measurements after calculating the median of CH 4 flux rates within replicate chambers.
Most chamber sites (82 out of 100) did not have a full year of sampling coverage (i.e., one or more monthly measurements were missing). Additionally, two EC sites did not have a full year of data (US-PLM) or lacked gap-filled meteorological and flux variables (US-HPY). For these and the chamber sites lacking annual CH 4 flux measurements, we developed scaling factors to upscale measurements to annual estimates. Scaling factors were calculated from the ratio between annual flux (in units of g CH 4 m -2 year -1 ) and average daily flux (in mg CH 4 m -2 day -1 ) using sites with a full year of sampling coverage (Bridgham et al., 2006). Scaling factors were calculated for both chamber and EC sites. At EC sites, we used non-gap-filled mean daily flux measurements against the annual sum calculated with gap-filled data. If a specific site-year had more than 2 consecutive months of missing data, we excluded it from the computation of scaling factors.
Particularly, we were interested in developing a scaling factor specific to the peak emission period (annual flux/average daily flux between June and August) since chamber studies in temperate marshes do not often include winter sampling when plant productivity is at its lowest (e.g., Bartlett, 1985;Weston et al., 2014) and most sites in the database met the condition of having data during this period. For sites lacking annual CH 4 flux measurements, we provided a first-order estimation of their annual CH 4 flux using the scaling factor derived from June-August mean daily CH 4 fluxes (without gap-filling for EC sites). We refrained from scaling mean daily CH 4 fluxes to annual estimates at chamber sites with a study duration ≤1 day (sites = 4). We averaged annual fluxes when there was more than 1 year of data for a given chamber or EC site. Additionally, we added two sites for which disaggregated data could not be synthesized (i.e., not in the dataset) but measured annual CH 4 fluxes had been published (Neubauer et al., 2000;Segarra et al., 2013).
We treated annual CH 4 flux estimates from chambers and EC as comparable, making no distinction between the two when estimat- TA B L E 1 Attribute categories used to classify chamber and EC sites in according to wetland type, vegetation, salinity, relative elevation, and disturbance conditions.
Wetland type Palustrine tidal Wetlands dominated by trees, shrubs, or emergents that occur in tidal areas where salinity is <0.5 psu (Cowardin et al., 1979) Estuarine intertidal Tidal wetlands usually semi-enclosed by land but have open, partly obstructed, or sporadic access to the open ocean, and in which ocean water is diluted by freshwater runoff from the land. Salinities >0.5 psu (Cowardin et al., 1979) Vegetation Class Mudflat Describes unvegetated areas exposed and flooded by tides Emergent Describes wetlands dominated by persistent emergent vascular plants Scrub/Shrub Describes wetlands dominated by woody vegetation ≤5 m in height Forested Describes wetlands dominated by woody vegetation >5 m in height Salinity class Fresh <0.5 psu Oligohaline 0.5-5 psu Mesohaline 5-18 psu Polyhaline 18-30 psu Mixoeuhaline 30-40 psu Elevation Class High Elevation above the Mean Highwater mark (MHW), inundated infrequently. Could be defined by vegetation communities (e.g., Spartina patens, Distichlis spicata, Salicornia sp., Juncus sp., and bulrush species) Mid Elevation in the relative middle of the tidal frame, frequently inundated, typically defined by vegetation communities (e.g., Spartina patens) Low Elevation relatively low in the tidal frame, frequently inundated, typically defined by vegetation communities (e.g., Spartina alterniflora) Levee Study-specific definition of a relatively high elevation zone built up on the edge of a river, creek, or channel Back Study-specific definition of a relatively low elevation zone behind a levee Disturbance class Undisturbed No disturbance or management has occurred on the site Tidally restored Tidal flow has been restored by removing an artificial obstruction Tidally restricted Tidal flow is muted or blocked by built structures. Includes impoundment Ditched Tidal hydrology is altered because artificial ditches have been cut to promote tidal flooding and drainage Species invasion Establishment of non-native species that compete with, displace, or even eliminate native species Submergence-Salinization Caused by sea level rise and saltwater intrusion Storm disturbance Major storms, including unusually high precipitation and/or wind events Removal of invasive plants Natural plant communities have been restored by actively removing invasive plant species Revegetation Wetland vegetation has been reintroduced by replanting on unvegetated surfaces Wetland construction Constructed wetland using sediments such as dredge spoils or other sediment source Scaling factors and CH 4 flux data had non-normal distributions; thus, we focused on comparing medians and used non-parametric tests such as paired-sample Signed test (S) for paired comparisons (i.e., year-round vs. June-August scaling factors), Mann-Whitney test (U) for two-group comparisons (e.g., scaling factors between EC and chambers), or a Kruskal-Wallis Dunn's test (H) using the Benjamini-Hochberg correction for multiple comparisons (i.e., fluxes between salinity, elevation, and disturbance classes). All statistical analyses were done at a level of significance of α < 0.05.
To evaluate the predictors of annual CH 4 fluxes across sites in CONUS, we combined chamber and EC annual CH 4 flux estimates and used qualitative (e.g., salinity and elevation class) and long-term climatological data (i.e., climate normals), which were broadly available at all sites. These data were fit to classification and regression trees such as Conditional Inference Trees (CTree) (Hothorn et al., 2006) and Random Forests (RF) (Breiman, 2001). CTrees use a significance test procedure to select variables at each split to reduce overfitting and selection bias. The stopping criterion is implemented when the global null hypothesis of independence between the response and any of the covariates cannot be rejected at a nominal level α, set at 0.05. The tree was constructed using the function "ctree" in the R package "Partykit" (Hothorn & Zeileis, 2015), limiting the tree depth to five levels. CTrees provide direct visualization of the splits at decision nodes, helping the interpretability of predictors, and are similar to binary partitioning methods used for analyzing soil CO 2 efflux measurements in terrestrial ecosystems (Vargas et al., 2010). However, they may suffer from issues associated with single trees, such as overfitting, high variance, or bias toward dominant classes.
For this reason, we added RF to our analysis. We trained a RF algorithm for annual CH 4 fluxes using R's "caret" package (Kuhn, 2008).
Similar to CTrees, the RF model was trained on all available data (i.e., we did not create training and test data splits) since our objective was to determine the hierarchy of predictor importance of CH 4 fluxes in tidal wetlands rather than to identify a predictive model that can generalize to new conditions (Knox et al., 2021). Hyperparameter tuning was performed for mtry (number of predictors randomly sampled at each decision node, selected at 6), and the number of trees was set to 400. For CTrees and RF models, we provide out-of-bag model fit metrics (coefficient of determination, mean absolute error, and root mean squared error) to further evaluate relative confidence in results.
Long-term average normals included in CTree and RF analyses were mean annual temperature and precipitation (MAT and MAP), mean daily maximum annual temperature and vapor pressure deficit (MATmax and VPDmax), and mean total daily shortwave solar radiation (Soltotal), which were extracted from PRISM using specific site coordinates (PRISM Climate Group, Oregon State University, https:// prism. orego nstate. edu, data generated November 10, 2022).
For these analyses, back and levee elevation classes (Table 1) were grouped within the low and high elevation classes, respectively, due to their low representation (less than three sites each) across the dataset. Salinity class was converted to numeric format to run the RF algorithm. Mean annual surface water or porewater salinity was calculated at sites with available data, while the midpoint of the salinity class range was used at sites with no salinity data.
To evaluate predictors of CH 4 flux across sites using chamberdisaggregated CH 4 fluxes by sampling event and time-specific environmental parameters, we employed generalized additive models (GAMs) (Hastie & Tibshirani, 1990). We chose GAMs over regression trees like RF because we prioritized interpretability was used as an estimate of precision and was applied to weight flux rates so that studies in which statistics had supported CH 4 fluxes were given a higher weight (0.8 vs. 0.2). All GAMs were implemented using the R "mgcv" package (Wood, 2011).
Leveraging nonlinear quantile regression, we established quantitative relationships between CH 4 emissions and the topranked predictor variables identified by GAMs, and estimated thresholds above which CH 4 fluxes were negligible. Quantile regression is valuable when assumptions like normality are not met.
Additionally, this approach is adept at resisting the influence of outliers and is well suited for handling heteroscedasticity (i.e.,
where the variance of the CH 4 fluxes varies across different levels of a predictor, e.g., salinity) (Koenker, 2005). It allows for more accurate modeling of the varying spread of CH 4 fluxes by estimating multiple slopes that describe the relationships between specific quantiles of the CH 4 flux distribution and a predictor that regression methods, focused solely on predicting mean values, would otherwise overlook (Cade & Noon, 2003). Quantile regressions were fitted for the 0.1, 0.5, and 0.9 quantiles of CH 4 fluxes using the nlrq() function within the R package "quantreg" (Koenker, 2023) due to the observed nonlinear relationships between CH 4 fluxes and the tested predictor variables. The slopes of the fitted conditional quantile regressions were used to estimate the predictor level required to decrease CH 4 fluxes by half, based on an exponential decay relationship (i.e., X 1/2 = ln(2)/slope). Subsequently, threshold values were estimated as seven times X 1/2 , representing a 99% reduction of CH 4 fluxes through interactions with increasing predictor levels. We calculated these thresholds for the 50th and 90th percentiles of the conditional distribution of CH 4 fluxes, representing the predictor thresholds below which the 50% and 90% of the highest CH 4 fluxes occur, respectively.
Half-hourly EC datasets were used to assess CH 4 flux magnitude fluctuations at diel to seasonal scales and their controlling factors.
Additionally, they provided insights into the significance of predictor variables not often evaluated in chamber studies, such as plant activity through GPP, net ecosystem exchange, or latent heat, as well as the effects of tidal pulsing on modulating CH 4 exchange.
We employed wavelet time series decomposition to identify major timescales of variation within the continuous CH 4 flux time series. Then, we used mutual information (I) to find the relative importance of each predictor variable and identify both synchronous and asynchronous interactions (Ruddell et al., 2013). Mutual information (I) measures the amount of information shared by two variables, X and Y, or the reduction in uncertainty of one variable given the knowledge of the other variable (Fraser & Swinney, 1986). The degree of mutual information between X and Y is increased by adding a time lag (positive or negative) in series Y relative to X, thereby allowing the identification of both synchro-
The final database contained 44 contributed datasets with 109 (100 chamber and 9 EC) tidal marsh sites, with measurements from 1980 to 2022. The dataset was dominated by CH 4 fluxes from tidal mesohaline wetlands with emergent vegetation, predominantly located on the US East Coast (Figure 1). Similar proportions (~30%-40%) of high, mid, and low tidal marsh environments were represented, with fewer sites located in environments affected by banks, berms, or levees. Half of the sites corresponded to undisturbed tidal marshes, followed by tidally restored (20%), ditched (8%), and other disturbance classes with a minority representation (<5%), including tidally restricted sites and species invasion (4%). Chamber-based data were unequally distributed throughout the year, with more observations concentrated from June to August (Figure S3). Five of the nine EC tower sites had paired chamber CH 4 flux measurements. Carbon dioxide and N 2 O fluxes were also compiled alongside CH 4 and are available in 52% and 30% of the sites, respectively, but they were not the focus of this synthesis.
A total of 29 site-years (11 from EC, 18 from chambers) had data covering all months of the year; hence, they were used to com- June-August daily average fluxes (0.21, IQR = 0.06) (S = 3, z = 3.97, P < 0.001) (Figure 2a). Scaling factors estimated using year-round daily averages were significantly lower for EC (s.f. = 0.32) than for chamber measurements (s.f. = 0.39) (U(N EC = 11, N chamber = 18) = 42, z = -2.54, p = .01), but no significant differences in the scaling factor were observed between methods when using daily averages from June to August (U(N EC = 10, N chamber = 18) = 69, z = -0.98, p = .33). For the 29 site-years (EC and chamber combined), the annual CH 4 fluxes estimated using the June-August scaling factor closely agreed with the measured values (Figure 2b). We did not find significant differences in annual scaling factors between S2-S5).
Using available full site-year data, published fluxes, and scaling factors at sites where the sampling coverage was shorter than a year, we estimated annual CH 4 fluxes from tidal marshes in CONUS. From a total of 108 sites including chamber and EC datasets, mean ± SD annual CH 4 fluxes were 26 ± 53 g CH 4 m -2 year -1 , but the central tendency of the estimates represented by the median and the geometric mean were seven to five times lower, at 3.9 and 5.4 g CH 4 m -2 year -1 , respectively (note that the geometric mean excludes sites with negative or zero annual CH 4 fluxes)
(Figure 3a). Median CH 4 fluxes were significantly higher at freshwater sites than at mesohaline, polyhaline, and mixoeuhaline marshes (H(4) = 28.06, p < .001) (Figure 3b). Statistical overlap of annual CH 4 fluxes existed between fresh and oligohaline, oligohaline and mesohaline, and polyhaline and mixoeuhaline conditions. However, when using chamber flux data disaggregated by sampling event, differences between salinity classes were more pronounced, with only fresh and oligohaline conditions showing statistical overlap (z = -1.93, p = .27) (Figure 3c). The variance of CH 4 fluxes at fresh and oligohaline sites was 27, 700, and >5000 times larger than at mesohaline, polyhaline, and mixoeuhaline sites, respectively. No statistical differences were observed between the distributions of chamber-and EC-derived annual fluxes across CONUS or when aggregated by salinity class (Table S6).
The wide numerical range and the right-skewed nature of the CH 4 flux data were also observed in annual CH 4 fluxes when separated by salinity class, which additionally showed a slight bimodal distribution (Figure S4). In fresh-oligohaline and mesohaline conditions, there was a trend of higher CH 4 fluxes in low than high marsh environments, but this pattern was not observed in more saline sites (Figure S5). The assessment of disparities in Results of both the CTree and RF methods were similar, ranking salinity as the most important predictor of the magnitude of annual CH 4 fluxes, followed by mean daily maximum annual temperature (MATmax) and, to a minor extent, elevation class and mean daily maximum annual vapor pressure deficit (Figure 4; Figure S7). These rankings were obtained using EC and chamber-based annual estimates combined. Average model performance was highest for RF with an out-of-bag R 2 of 0.94, MAE (mean absolute error) of 7.9 g CH 4 m -2 year -1 , and RMSE of 14 g CH 4 m -2 year -1 (Figure S7). The best resulting CTree (R 2 = 0.66, MAE = 17, RMSE = 31 g CH 4 m -2 year -1 ) was achieved when fresh and oligohaline, and polyhaline and mixoeuhaline classes were combined. This CTree was composed of 11 decision nodes, with salinity at the root of the tree, followed by MATmax and elevation class. The highest annual CH 4 fluxes were observed at fresh and oligohaline sites with MATmax above 25.6°C, followed by frequently inundated low and mid-fresh-oligohaline marshes with MATmax ≤25.6°C, and mesohaline sites with MATmax above 19°C.
Methane fluxes at sites with salinities >18 psu were consistently low regardless of MATmax or elevation class (Figure 4). The representation of elevation classes was similar among fresh-oligohaline systems with MATmax ≤25.6°C, mesohaline sites with MATmax ≤19°C, and polyhaline and mixoeuhaline marshes (Figure S8). However, elevation could not be assessed as a predictor of annual CH 4 fluxes at warmer fresh-oligohaline (>25.6°C) and mesohaline (>19°C) sites due to the lack of variation in elevation classes within these categories.
While the mean difference between predicted and observed values was roughly equal to the average of annual CH 4 fluxes across the dataset, CTree binary partitioning enabled a breakdown of annual CH 4 fluxes across marsh categories based on salinity and daily maximum annual temperature, with additional consideration of elevation class specifically within the fresh-oligohaline category (Table 3).
We used chamber-based disaggregated data to understand CH 4 flux variability across sites and to establish quantitative relationships between CH 4 fluxes and predictor variables, particularly porewater biogeochemistry. Results from GAM models showed that porewater CH 4 , porewater sulfate concentrations, and porewater salinity explained the highest percentage of the variance (27%-16%), followed by surface water salinity and air temperature (13%-8%) (Table 4).
Similar results were obtained if CH 4 flux data were filtered to consider emissions only (i.e., CH 4 fluxes >0) (Table S7). Surface and porewater nitrate concentrations and porewater temperature also explained some percentage of the deviance in CH 4 fluxes (26%, 10%, and 18%, respectively); however, these results were based on a limited number of studies (n < 5) (Table S7). Adding study ID as a random ) effect increased the model performance between 6 and 32%, with the lowest increase observed for porewater salinity and the highest increase observed for surface water salinity. This highlights the intrinsic site-specific variability of CH 4 fluxes when predicted using surface water salinity. It also emphasizes the disconnect between surface and porewater salinity. Multivariate GAM models combining porewater CH 4 and sulfate concentrations, or surface salinity and air temperature, achieved the best performance with the highest deviance explained, 73% and 67%, respectively. However, these models were representative of only five studies (Table 4).
To estimate effect sizes, we examined the individual relationships between CH 4 fluxes and porewater concentrations of CH 4 , sulfate, and salinity. All these variables, except for porewater CH 4 , had significant effects on the magnitude of CH 4 fluxes, as shown by nonlinear quantile regression fits (Figure 5; Table S8). We observed a significant exponential decrease in CH 4 fluxes as surface and porewater salinity or sulfate concentrations increased. Moreover, as salinity and sulfate concentrations increased, the range of CH 4 fluxes decreased, suggesting that the effects of salinity and sulfate on CH 4 emissions were not consistent across all study sites. At the median quantile, CH 4 fluxes were significantly reduced when porewater sulfate and porewater salinity exceeded 2.8 ± 0.5 mM and 9.6 ± 1.1 psu, respectively. A 90% response, indicated by a significant reduction of the 0.9 quantile of the CH 4 fluxes, was achieved when sulfate concentrations reached 4.7 ± 0.6 mM, and porewater and surface water salinity reached 21 ± 2 and 15 ± 3 psu, respectively (Figure 5). These values represent the mean estimated cutoff points below which either the 50% or 90% of the highest CH 4 fluxes occur.
Analysis of the quantile regression model residuals revealed that CH 4 fluxes were not dependent on only one predictor variable.
Several environmental covariates could explain variability in model residuals of the porewater or surface water salinity-CH 4 relationship. Primarily, the residual variance in CH 4 fluxes modeled from surface water salinity was notably influenced by porewater sulfate and CH 4 concentrations. Likewise, sulfate concentrations and porewater temperature influenced the remaining variance in fluxes modeled TA B L E 3 Summary of annual CH 4 fluxes (g CH 4 m -2 year -1 ) grouped by salinity class, MATmax, and elevation class from the Conditional Inference Tree in Figure 4. Salinity class MAT max Elevation class N Distribution Mean SD SE Median Geom. mean geoSD Fresh-oligohaline >25.6 7 Normal 171.5 79.4 30 204 153 1.7 Fresh-oligohaline ≤25.6 Low, mid 20 Square root-normal 54.9 56 12.5 41.5 15 10.5 Fresh-oligohaline ≤25.6 High 10 Log-normal 5.3 6.3 2 3.2 3.8 a 2.7 Mesohaline >19 8 Log-normal 21.5 19.5 6.9 19.2 15.8 2.3 Mesohaline ≤19 39 Log-normal 6.9 10.3 1.7 3.1 3.4 a 3.6 Poly-mixoeuhaline 24 Log-normal 1.8 2.3 0.5 0.7 1.2 a 2.9 Note: N is number of sites, SD and SE are standard deviation and error, respectively, geom.mean refers to geometric mean, and geoSD is the standard deviation of the latter. a Negative flux values (n = 1 high fresh-oligohaline ≤25.6; n = 1 mesohaline ≤19°C) and fluxes equal to zero (n = 1; mesohaline ≤19°C; n = 3 polymixoeuhaline) were removed to compute the geometric mean. TA B L E 4 General additive model (GAM) results for chamber-disaggregated CH 4 fluxes against time-specific predictor variables. Variable R 2 Deviance explained (%) d.f. AIC # studies Porewater CH 4 0.27 [0.33] 27 [34] 879 8420 [8336] 11 Porewater SO 4 2-0.21 [0.27] 22 [29] 310 4200 [4177] 6 Porewater salinity 0.16 [0.20] 16 [22] 1162 16,734 [16,680] 15 Surface water salinity 0.12 [0.46] 13 [46] 760 9463 [9088] 6 Air temperature 0.08 [0.26] 8 [26] 6729 88,747 [87,337] 28 cosHour 0.05 [0.32] 5.1 [32] 6229 [83305] 18 Porewater SO 4 2-and porewater CH 4 0.67 [0.71] 69 [73] 239 2097 [2065] 5 Surface water salinity and air temperature 0.60 [0.66] 61 [67] 654 7632 [7514] 5 Porewater salinity and porewater CH 4 0.32 [0.36] 34 [39] 347 3594 [3576] 7 Notes: In brackets are GAM results with Study ID as a random effect. Results shown include relationships that explain >5% of the deviance in CH 4 fluxes, focusing on variables available in at least five studies. Table S7 contains GAM results for CH 4 emissions (i.e., >0) against all time-specific predictor variables with a significant relationship (p < .05). d.f. stands for degrees of freedom, AIC is the Akaike information criterion, and # studies indicate the number of studies that recorded each predictor variable (the total number of studies is 35).
from porewater salinity (Figures S9 and S10). In contrast, no predictor variable was found to explain the variability in model residuals of the sulfate-CH 4 flux relationship.
Mutual Information analysis using wavelet decomposed EC datasets revealed that CH 4 responses to environmental covariates exhibited nonlinearity and were characterized by asynchronous interactions, particularly at the multiday and seasonal scales (compare maximum I R heatmaps in Figure 6 with those of synchronous I R in Figure S11). the diel scale, the main predictors of CH 4 flux were GPP, net ecosystem exchange, and soil temperature. Latent heat and water table depth ranked 4th and 5th in importance. On a multiday scale, the hierarchy of predictors was led by air temperature, water table depth, and atmospheric pressure. At the seasonal scale, water and air temperature were among the top predictors, followed by GPP and incoming shortwave radiation. Vapor pressure deficit and salinity ranked 5th and 6th, respectively, while water table depth was last in importance. At the seasonal scale, the dominance of water and air temperature, followed by GPP was apparent in most sites (Figure 6c). In some cases, lows in dissolved oxygen (at US-StJ) or high precipitation, bringing pulses of fresh water (US-LA1), manifested as key controls of CH 4 flux (Figure 7g,i).
While the influence of temperature (air, soil, or water) and effects of plant activity (through GPP, net ecosystem exchange or latent heat)
on CH 4 fluxes were apparent across timescales and sites, the presence of a diel cycle marked by plant activity was site-specific. Heatmaps in F I G U R E 7 Examples of diel and seasonal variation in the wavelet detail reconstruction for CH 4 flux and predictor variables. Note that the mean is removed in wavelet detail reconstructions; therefore, the y-axes are relative rather than absolute. Panels (a) through (c) illustrate diel wavelet details, panels (d) and (e) show examples of multiday wavelet details, and panels (f) through (i) show seasonal wavelet details. Predictor variable abbreviations are those introduced in Figure 6. DO stands for dissolved oxygen. Wavelet details for CH 4 flux, water table depth, and salinity at the diel and seasonal scales across all eddy covariance sites are in Figures S13 and S14, respectively.
0 4 8 12 16 20 24 -3 0 3 6 FCH4 GPP Hour US-Srr (a) US-Dmg US-Stj US-Dmg US-LA2 US-LA2 -10 -5 0 5 10 15 0 4 8 12 16 20 24 -80 0 80 160 FCH4 WTD Hour CH 4 flux diel detail (nmol m -2 year -1 ) (b) -0.6 -0.4 -0.2 0 0.2 0.4 0 4 8 12 16 20 24 -40 0 40 80 FCH4 WTD Hour (c) -0.8 -0.4 0 0.4 0.8 Driver diel detail 2012-06-09 2012-07-07 2012-08-04 -80 -40 0 40 80 FCH4 WTD CH 4 flux multiday detail (nmol m -2 s -1 ) (d) -0.2 0 0.2 2012-06-09 2012-07-07 2012-08-04 -80 -40 0 40 80 FCH4 PA (e) -0.8 -0.4 0 0.4 0 60 120 180 240 300 360 -40 0 40 80 120 FCH4 TW CH 4 flux seasonal detail (f) -8 -4 0 4 8 Driver seasonal detail 0 60 120 180 240 300 360 -40 0 40 80 FCH4 P DOY CH 4 flux seasonal detail (nmol m -2 s -1 ) US-LA1 (g) US-StJ US-StJ -0.1 -0.05 0 0.05 0.1 0.15 0 60 120 180 240 300 360 -20 0 20 40 FCH4 GPP DOY (h) -4 -2 0 2 4 6 8 0 60 120 180 240 300 360 -25 0 25 50 FCH4 DO DOY (i) -6 -4 -2 0 2 4 Driver seasonal detail
Our primary goal was to improve predictions of CH 4 emissions from tidal marshes in CONUS. This discussion focuses on four research questions that align with this goal. Firstly, we identify the primary predictors of CH 4 flux in tidal marshes across sites using broadly available data. Secondly, using EC sites with continuous CH 4 flux data, we assess the effects of predictors across varying timescales.
Third, we discuss the application of scaling factors to annualize short-term static chamber measurements and their limitations, considering daily and seasonal flux variations. Lastly, we highlight how the identified relationships and gained information can be applied to improve monitoring and predictions of CH 4 emissions in tidal marshes, supporting blue carbon assessments and advancing our ability to constrain estimates of GHG emissions across diverse tidal wetlands.
In our analysis of tidal marsh CH 4 fluxes across CONUS, we observed a wide range of flux magnitudes, varying from -150 to 4120 nmol m -2 s -1 , with an average of 82 nmol m -2 s -1 and a median significantly lower at 5.8 nmol m -2 s -1 . When compared to available estimates from a recent synthesis, the median instantaneous CH 4 flux per unit area aligns with values reported for North American
saltmarshes, yet it is approximately half of that observed for mangroves, and about 10 times higher than figures reported for seagrasses (Rosentreter et al., 2023). Annual CH 4 fluxes showed a positive skew with a mean of 26 ± 53 g CH 4 m -2 year -1 and median and geometric means significantly lower, a reflection of the predominant data from mesohaline tidal marshes as well as the notoriously variable behavior of CH 4 , with hotspots of activity. This raises a question about the reliability of arithmetic, median, or geometric mean values as estimates for representing the regional scale magnitude of annual CH 4 emissions in tidal marshes or their emission factors globally. The adequacy of these metrics will largely depend on the actual distribution and proportion of various marsh salinity classes across different climatic zones (e.g., Table 3). However, at finer spatial and temporal scales, factors such as hydrology, plant activity, and porewater biogeochemistry may become more influential than salinity in determining tidal marsh CH 4 fluxes. fluxes and salinity. Sites with salinities above 18 psu had significantly lower CH 4 fluxes than less saline marshes, but this relationship was found to be less predictive at sites with salinities fresher than 18 psu (Poffenbarger et al., 2011;Windham-Myers et al., 2018). This agrees with our results when data were summarized at the annual level.
When examining the data at the disaggregated level, CH 4 fluxes from freshwater and oligohaline marshes (0-5 psu) were the only groups showing statistical overlap (Figure 3c). This suggests that the (~204 g CH 4 m -2 year -1 ). These were followed by frequently inundated low and mid-fresh-oligohaline marshes with MATmax ≤25.6°C
(~42 g CH 4 m -2 year -1 ), and mesohaline sites with MATmax above 19°C (~19 g CH 4 m -2 year -1 ) (Table 3). The fact that elevation class emerged as an important explanatory variable for CH 4 fluxes at low salinity levels may reflect the decrease in soil oxygen availability as flooding increases from high to low elevations (Kirwan et al., 2013), enhancing methanogenesis and suppressing methanotrophy. This pattern, however, did not emerge in saline systems, possibly due to the overriding influence of sulfate availability compared to fresh sites (DeLaune et al., 1983;Martens & Berner, 1974). In saline environments, CH 4 fluxes could remain low regardless of inundation, as sulfate reduction precedes methanogenesis. However sample size limitations and high variability within groups could have also played a role.
Synthesizing chamber-disaggregated flux and ancillary variables by sampling event allowed us to evaluate further the effects of salinity, temperature, and porewater biogeochemistry on the magnitude and variability of CH 4 fluxes in tidal wetlands. As per the well-established CH 4 response to salinity observed in other studies (Bartlett et al., 1987;DeLaune et al., 1983;Poffenbarger et al., 2011;Sanders-Demott et al., 2022;Schultz et al., 2023;Windham-Myers et al., 2018), chamber flux data disaggregated by sampling event also showed a significant exponential decrease with increasing salinity, for both surface water and porewater (Figure 5c,d). Across the range of sites, fitted median CH 4 fluxes fell below 1.6 ± 1.3 nmol m -2 s -1 at porewater salinities above 9.6 ± 1.1 psu. However, some instances of high CH 4 emissions (50-170 nmol m -2 s -1 ) were still evident above this threshold. A more conservative threshold was identified at porewater salinities of 21 ± 2 psu, incorporating the 90th percentile of the CH 4 fluxes. The latter threshold was 15 ± 3 psu if surface water salinity was considered instead. Surface water and porewater salinity are not always well correlated at a site, as surface water salinity is often more variable than porewater salinity (Wilson et al., 2015).
Indeed, the salinity measured on the surface may not accurately reflect the salinity conditions experienced by methanogens or, more precisely, their competition with sulfate-reducing bacteria.
Surface inputs of salts are modified by plant transpiration, while sulfate inputs are modified by the balance between consumption by sulfate-reducing bacteria and production by sulfide oxidation.
Such process may explain the reduced predictability of CH 4 emissions when using surface water salinity as a standalone predictor;
thus, porewater salinity is preferred. Methane emissions significantly decreased with increasing porewater salinity across all conditional quantile levels, whereas surface salinity only explained changes in CH 4 emissions at the 90th percentile of the flux distribution (Table S8; Figure 5).
Causes of CH 4 flux variation at low salinities have previously been explained by microsite spatial and temporal variability in the presence of alternate electron acceptors and the sensitivity of methanogens to variations in their availability (Brooker et al., 2014;Galand et al., 2003). Because of the high abundance of sulfate in seawater, salinity has been used as a proxy of sulfate availability in tidal and salt marsh studies. However, we observed considerable variability in sulfate concentrations at a given salinity level, particularly when surface salinity was used to infer porewater sulfate concentrations (Figure S15a,b). Sulfate concentrations can change independently of salinity due to local sulfate depletion, leading to high CH 4 emissions despite high salinity levels, as observed in Wilson et al. (2015), and at high (>10 psu) porewater salinities in this synthesis (Figure 5d).
The general inverse relationship between sulfate concentration and CH 4 emissions is well-supported and upheld by studies through time (Bartlett et al., 1987;DeLaune et al., 1983;Martens & Berner, 1974;Poffenbarger et al., 2011). However, it is not well defined if there is a specific sulfate concentration above which tidal marsh CH 4 emissions are negligible.
Poffenbarger et al. (2011) found that porewater CH 4 concentrations were negligible at sulfate concentrations >4 mM in marsh soils. However, they did not relate porewater sulfate concentration to CH 4 emissions due to the lack of paired data to evaluate such a relationship. Our analysis suggests that median CH 4 emissions were low across sites at porewater sulfate concentrations >2.8 ± 0.5 mM. However, considering the large spread of the response of CH 4 emissions to sulfate at low (<5 mM) sulfate concentrations, a more stringent threshold of 4.7 ± 0.6 mM delineated CH 4 emissions in the lowest 90th percentile of the data (Figure 5b). CH 4 emissions displayed significant responses to sulfate concentrations at all quantile levels and exhibited a third of the residual variance observed in CH 4 flux responses to porewater salinity. Residual analysis revealed that the modeled salinity-CH 4 relationship tended to overestimate CH 4 emissions as porewater sulfate concentrations decreased, particularly within 0 to 5 mM, indicating a greater sensitivity to increasing sulfate availability than to increasing salinity. A similar trend was observed for residuals fitted to porewater CH 4 concentrations, particularly within the range 0-100 μM CH 4 ,
highlighting the activity of sulfate reducers as a primary control on CH 4 production in surface soils (0-30 cm).
We did not find a consistent significant relationship between porewater CH 4 concentrations and CH 4 fluxes (Figure 5a; Table S8).
The discrepancy between CH 4 production and emissions may be attributed to several factors. These include bacterial-mediated aerobic CH 4 oxidation near the sediment surface during low tide or facilitated by plant rhizosphere oxygenation (Megonigal & Schlesinger, 2002;Van der Nat & Middelburg, 1998). Additionally, non-diffusive emission pathways like ebullition and plant-mediated transport (Hill & Vargas, 2022) are not always accounted for in chamber studies, as chambers may not include tall vegetation or headspace concentrations may be filtered to remove pulsed or erratic emissions. Furthermore, tidal flooding could drive lateral transport of dissolved porewater CH 4 from the marsh, which would be missed by chamber and eddy flux methods (Kelley et al., 1995;Tong et al., 2010;Trifunovic et al., 2020).
Porewater nitrate concentrations are typically low in most coastal marshes (Valiela & Teal, 1974), but surface water concentrations can be notably high at some sites due to agricultural intensification and other anthropogenic sources (Galloway et al., 2003;Pardo et al., 2011) (Figure S15). Nitrate is a thermodynamically favorable electron acceptor used by anaerobic bacteria shown to enhance soil organic matter decomposition (Bulseco et al., 2019).
Similarly to sulfate, methanogenesis can be inhibited in the presence of nitrate. In this synthesis, signals of this process might be observed from the decrease in EC and chamber-based CH 4 fluxes with increasing dissolved nitrate concentrations (Figure S15c,d). However, the extent of CH 4 emission reduction was less pronounced for nitrate than for porewater sulfate. This pattern emerged despite a small dataset (n = 3 studies), suggesting that nitrate inhibition of methanogenesis by nitrate reducers (denitrifiers) is a very strong biogeochemical signal that should be explored further given the fact that the process may also produce nitrous oxide (N 2 O), a potent GHG. Our dataset lacked contemporaneous measurements of N 2 O emissions at these sites. Higher daytime CH 4 fluxes could stem from several factors, increased soil and air temperatures (Bansal et al., 2018), enhanced root exudation with GPP stimulating methanogens and microbial priming (Bridgham et al., 2013), or CH 4 transport mediated by plants.
This plant-mediated transport could either be driven by a pressure gradient that intensifies during daylight and coincides with active photosynthesis (Bansal et al., 2020;Dacey & Klug, 1979;Vroom et al., 2022) or by stomatal control of diffusive transport (Garnet et al., 2005).
The former might be particularly relevant in Phragmitesdominated marshes (van den Berg et al., 2020; Van der Nat & Middelburg, 1998). Evidence from local studies focusing on individual site locations suggested that both impoundment and Phragmites invasion, separately or combined, could markedly increase CH 4 fluxes (Martin & Moseman-Valtierra, 2015; Mueller et al., 2016; Sanders-Demott et al., 2022). The dominance of the diel scale over the seasonal scale in CH 4 flux variance is not limited to EC sites included in this synthesis; it has also been documented in other tidal ecosystems, such as in a subtropical estuarine mangrove (Liu et al., 2022). Chamber studies that conducted measurements over 12-24 h periods or consisted of automated chambers also noted diel CH 4 fluctuations of varying frequency and nature. For instance, Kelley et al. (1995) noted that CH 4 fluxes were higher when tidal waters were closest to the soil surface. Diefenderfer et al. (2018) observed that CH 4 flux was greater at night, likely due to the effects of hydrostatic pressure on diffusion and ebullition processes influenced by water surface elevation dynamics. Capooci and Vargas (2022) found that confluences of peak daily temperatures and low to rising tides could cause CH 4 pulses throughout the day. Tong et al. (2013) reported that plant activity controlled CH 4 emissions during neap tide days when sites were exposed; when sites were flooded, other factors dominated. These findings suggest that while diel CH 4 flux variability is expected in tidal marshes, the presence and strength of a daily cycle are site-specific depending on species, temperature, and tidal forcings. Microtidal sites in Louisiana do not experience large tidal amplitudes nor the sinusoidal pattern with tides; thus, peaks in wavelet analysis were not observed at diel scales related to semidiurnal tides or fortnightly neap-spring tidal cycles (Figure S13). Sites in Louisiana are wind-and river-influenced, and atmospheric pressure changes and freshwater pulses driven by synoptic and mesoscale weather played a more significant role in CH 4 flux variability. These results suggest that CH 4 fluxes and their drivers at microtidal sites may align more closely with those of non-tidal wetlands found elsewhere (Knox et al., 2021; Sturtevant et al., 2016). In contrast, where tides are significant, they may strongly modulate the timing and pathway of CH 4 emissions. At the seasonal scale, water and air temperatures were the primary predictors of CH 4 flux, followed by GPP. Salinity ranked 6th in importance, demonstrating greater relevance at this scale compared with other timescales. This may reflect the limitations of using surface water salinity at EC sites to explain diel CH 4 flux responses. It may also suggest that seasonal, rather than shorter-term diel changes in surface water salinity are required to affect CH 4 emissions significantly. Seasonal salinity changes, driven by reduced freshwater flow and increased seawater intrusion, typically elevate salinity during the summer and growing season. However, higher temperatures and enhanced GPP during these periods could offset potential CH 4 flux inhibition due to increased sulfate availability. A distinct observation was that CH 4 fluxes were largely enhanced when reduced salinity aligned with peak growing season conditions (Figure 7; Figure S14).
Only a few chamber studies have shown the significant variability of CH 4 fluxes at daily scales (Capooci & Vargas, 2022;Diefenderfer et al., 2018;Kelley et al., 1995;Tong et al., 2013), highlighting the limitations of temporally upscaling CH 4 fluxes from discrete static chamber measurements (Hill & Vargas, 2022;Vargas & Le, 2023).
Static chambers, predominantly deployed at low tide, are limited in their sensitivity to sudden pulse events and may not effectively cap- , 2016), disparities appeared to be site-specific and difficult to reconcile at the regional level.
Building on the scaling factors discussed earlier, Bridgham et al. (2006) found a similar ratio between annual and mean daily CH 4 fluxes from chamber measurements in non-tidal freshwater (s.f. = 0.36) and estuarine (s.f. = 0.34) wetlands across North America, suggesting there is a ~ 360-day emission season. Seasonality in CH 4 fluxes is evident in many tidal wetland datasets (Figure S14a) (e.g., Derby et al., 2022;Neubauer, 2013;Vázquez-Lule & Vargas, 2021), Most static chamber studies in temperate marshes typically exclude winter sampling. Traditionally, a common approach was to annualize the average daily fluxes from the growing season, applying a 150-day emissions season and assuming negligible winter emissions (Bartlett & Harriss, 1993;Magenheimer et al., 1996). More recently, scaling factors, as described in Bridgham et al. (2006), have been used to calculate annual CH 4 fluxes from mean daily fluxes sampled during the growing season (Bridgham et al., 2006;Poffenbarger et al., 2011). However, this approach may lead to an apparent overestimation of annual CH 4 fluxes of up to ~50% due to the relative seasonal increase in CH 4 fluxes compared with baseline (Figure S17).
According to our results, an emission season of 210 days (or a s.f. = 0.21) may be more accurate to consider if fluxes are sampled during the peak emission period (June-August) (Figure 2). The variance of the June-August scaling factor across sites was lower than that observed for the ratio of annual flux: mean daily flux averaged year-round. Additionally, no significant differences were detected in the June-August scaling factors between chamber and EC sites. This might be due to the lower variability in within-site CH 4 fluxes during the peak emission season compared with that observed year-round.
Consequently, the ratio of annual flux to June-August mean daily flux may be less affected by how well the length and amplitude of the seasonal cycle are captured by discrete chamber measurements.
Results of the present study could improve predictions of CH 4 emissions in tidal marshes at the local and regional levels in vari- Likewise, the hydrologic setting, often reported qualitatively as either "high" or "low" marsh environments, could be made a stronger