Three Decades of Wetland Methane Surface Flux Modeling by Earth System Models‐Advances, Applications, and Challenges

Earth System Models (ESMs) simulate the exchange of mass and energy between the land surface and the atmosphere, with a key focus on modeling natural greenhouse gas feedbacks. Methane is the second most important greenhouse gas after carbon dioxide. There are growing concerns over the rapidly increasing methane concentration in the atmosphere, underscoring the need for accurate global modeling of its emissions using ESMs. Of the multitude of sources of methane globally, wetlands are the largest natural emitters for methane, leading to significant efforts targeting their representation in ESMs with a special focus on their methane emissions. In this review, we first provide a historical overview of including wetland‐methane components in ESMs and how methane modeling approaches have evolved over time. Second, we discuss recent modeling advancements that show promise for improvements in methane emissions predictions, namely the coupling of surface and atmospheric modules of ESMs, the representation of microtopography and transport mechanisms, the resolution of microbial processes at different spatial‐temporal scales, and the improved mapping of wetland area extent across the different wetland types. Third, we shed light on the different challenges hindering accurate estimations of wetland‐methane emissions, as shown by the consistent discrepancy between bottom‐up and top‐down models' predictions. Finally, we emphasize that more detailed representation of biogeochemistry and dynamic hydrology while resolving the within‐wetland vegetation heterogeneity should improve model predictions, especially when coupled with expanding ground‐based measurement networks and high‐resolution remote sensing mapping of methane‐relevant variables, such as water elevation, water table depth, and methane concentration.


Introduction
Wetlands are distinctive areas characterized by recurrent or sustained, shallow inundation, or saturation at or near the surface of the substrate (National Research Council, 1995).The hydrology of a wetland, influenced by both climate and basin geomorphology (referred to as hydrogeomorphology), directly alters the physiochemical environment.These factors collectively shape the biota within the wetland, including vegetation, animals, and microbes (Mitsch & Gosselink, 2007).The key diagnostic attributes of wetlands are hydric soils, wetland hydrology, and hydrophytic vegetation (hydrophytes) (Cowardin, 1979;National Research Council, 1995).Wetlands are typically located at the interface of terrestrial (e.g., upland forests and grasslands), and aquatic systems (e.g., rivers, lakes, and oceans).They can also exist in seemingly isolated settings, where they are often connected to the adjacent groundwater aquifer (Mitsch & Gosselink, 2007).While water is a defining feature, the depth and duration of flooding vary significantly within and across wetland types, and from year to year.Water levels fluctuate between seasons and years within the same wetland and wetland type, making it challenging to determine wetland presence and extent based solely on the presence of water at any given time.Depth is also a key characteristic distinguishing wetlands (shallow, typically <10 m over a long-term average, though the exact threshold is arbitrary and varies by application) from lakes (which are deeper).The size of wetlands varies widely, ranging from small patches of a few hectares to expansive areas covering several hundreds of square kilometers.Their locations throughout the landscape also vary greatly, from coastal to inland wetlands, tropical to arctic, and rural to urban regions.These dynamic and variable characteristics of wetlands, including variations in depth, duration of flooding, size, and type present unique challenges in wetland modeling (Mitsch & Gosselink, 2007;Tiner, 2016).
While occupying only 5%-8% of the Earth's surface, wetlands are estimated to store 16%-20% of the global soil carbon pool because of their anaerobic conditions that enhances carbon sequestration (Jackson et al., 2017;M. Lu et al., 2021).Wetlands provide diverse ecosystem services such as flood control, habitat for biodiversity, and water quality regulation (Salimi et al., 2021).However, wetlands and inland waters are also the largest natural emitters of methane and are estimated to contribute about 83% of global natural methane emissions and 22%-48% of total (natural and anthropogenic) emission sources (Saunois et al., 2020).Strictly defined wetlands (i.e., not rivers, lakes, ponds, and reservoirs) contribute about half of that amount (Saunois et al., 2020).Terrestrial methane emissions, mostly from tropical and mid-latitude wetlands, account for the largest component of recent decades' increase in global methane concentrations (Basu et al., 2022).Consistently, wetland methane emissions are identified as the largest source of uncertainty in global methane emission estimations, using either top-down or bottom-up approaches (Saunois et al., 2020).Wetlands are often discussed as a potential nature based-solution, typically emphasizing their hydrological function in flood prevention and improvement of water quality.However, they are very rarely considered for greenhouse gas reduction (for example, Thorslund et al., 2017).This is probably due to the complex role they play as both large stores of carbon and, at the same time, significant sources of secondary greenhouse gases, that is, methane and N 2 O (Mitsch et al., 2013;Novick et al., 2022).Some exceptions, which discuss wetlands as climate change nature-based solutions: reconstruction and conservation of tidal wetlands, with low methane production due to salinity, were proposed as nature-based solution for climate mitigation (Fargione et al., 2018;Kroeger et al., 2017), rewetting of drained peatland to prevent soil carbon loss (for example, Günther et al., 2018); and approaches to reduce methane emissions from flooded agricultural land, such as rice fields (Runkle et al., 2019).
One important tool in scientists' arsenal to understand these emissions and to bound their uncertainty is the use of Earth System Models (ESMs).ESMs are a powerful and commonly applied computational tool to quantify feedbacks between the carbon cycle and the atmosphere (Friedlingstein et al., 2006).They contain multiple interacting components including atmosphere, marine, and land components.Of these, the land component, referred to formally as Land Surface Models (LSMs), are responsible for simulating the exchange of energy and mass fluxes across the land-atmosphere interface including carbon fluxes associated with wetlands.Originally developed to study the controls of land surface on meteorological processes, LSMs have incorporated increasingly a wide variation of ecological processes (Fisher & Koven, 2020).Despite their importance, wetland processes and dynamics have been relatively late to be incorporated into LSMs, and the representation of the processes involved in the methane cycle has rapidly evolved through the years, leading to a great deal of variation among models (Melton et al., 2013;Wania et al., 2013).Improving the processlevel representation of wetland methane fluxes in ESMs may facilitate the use of such models to predict and resolve the role wetlands play in the global greenhouse gas budget.This improvement can enhance climate predictability and support planning, management, and crediting of nature-based solutions for climate regulation that utilize wetlands.The power of using ESM models in this context lies in the possibility to assess the longevity of the proposed carbon credits in wetland restoration or protection when threatened by climate extremes, which is currently a large challenge (Mack et al., 2022).In this work, we review the integration of wetlands and their corresponding methane dynamics in LSMs and further into ESMs over the last couple of decades, while emphasizing the different modeling scales, input data sets, and mechanistic processes involved in current applications, the barriers to more common implementation in climate simulations, and the cutting edge of development for near-future applications.

Overview of Methane Dynamics in Wetlands
Three main components govern methane dynamics in wetlands: (a) methane production (methanogenesis) and (b) methane consumption (methanotrophy), which are both driven by microbial communities active in the wetland sediments (Lyu et al., 2018;Singleton et al., 2018), and (c) methane transport to the atmosphere.Methanogenesis is considered to be the largest source of biogenic methane in the world, where methane is the end product of microorganismal degradation, known as methanogens, occurring mostly under anaerobic conditions (Thauer et al., 2008).Notably, some recent investigations showed methanogenic activity under oxygenated soils as well (Angle et al., 2017;Smith et al., 2018).Methanogens express the methyl-coenzyme M reductase (Mcr) responsible for methane production.The Mcr reacts with three main substrates: carbon dioxide, acetate, and methylated compounds (A.M. Zhang et al., 2019), thus resulting in three main pathways for methane production in wetlands: hydrogenotrophic, acetoclastic, and methylotrophic (Conrad, 2020).In contrast to methanogenesis, methanotrophy is processed by one class of microorganisms known as methanotrophs through oxidation, which is responsible for the consumption of a large part of the produced methane (Le Mer & Roger, 2001).Oxidation occurs under both aerobic and anaerobic conditions, where oxidation in the aerobic topsoil uses oxygen as the electron acceptor, while anaerobic oxidation (in wetland sediments, and deeper saturated soil layers) relies on other oxidants, most commonly nitrate or sulfate (Segers, 1998;W. Wang et al., 2018).Methane transport to the atmosphere occurs primarily through three different pathways: (a) molecular diffusion through the soil sediments and water column, (b) ebullition representing the release of methane bubbles trapped under the surface and rapidly emitted to the atmosphere while bypassing oxic soils at the top of the surface (Peltola et al., 2018), and (c) transport through plant aerenchyma, which are internal plant tissues that facilitate gas diffusion (Villa et al., 2021).Different environmental and atmospheric variables influence methane fluxes from wetlands at different temporal scales, with water elevation (when water is above the surface) or water table depth (when the inundated fraction is below the soil surface), and soil temperature shown to be major predictors, in addition to atmospheric pressure, vapor pressure deficit, and boundary layer mixing processes (Knox et al., 2021).
A main difference between wetland and upland ecosystems is the strong interaction between ecology and hydrology driving most biogeochemical processes within wetlands (Zhou et al., 2016).The inundation dynamics control aerobic and anaerobic boundaries, thus making hydrological variables, such as water table depth, water surface elevation, inundation fraction, and soil moisture content important factors regulating methane production, oxidation, and transport (Bansal et al., 2016;Le Mer & Roger, 2001;Perryman et al., 2020).Typically, wetter conditions enhance methane production and reduce oxidation resulting in more surface methane emission.However, there are competing effects that complicate the interaction between these processes, such as the methanogenic substrates availability in wet soils, competition between methane oxidation and heterotrophic respiration for available oxygen, and the gas transport, which in turn are functions of inundation (Altor & Mitsch, 2008;Calabrese et al., 2021;Hondula et al., 2021).In addition, lateral flow plays a major role in the wetland's carbon and nutrient budgets where dissolved materials could be imported or exported from the wetland, thus altering the mass balance and corresponding dynamics of carbon, nitrogen, and phosphorus cycles (Bogard et al., 2020;Herbert et al., 2020;Xiao et al., 2019), where it is highly dependent on the wetlands' type and location (Atkinson et al., 2019).For example, lateral flow exports and imports are significantly different between wetlands downstream of agricultural lands (Johnes et al., 2020;Villa et al., 2023), tropical/subtropical and tidal wetlands (Cobb et al., 2020;Yao et al., 2022), and arctic wetlands (Young et al., 2021).
Another difference between wetland and upland ecosystems is vegetation physiology and structure.In order to survive in wet environments, wetland vegetation have developed special tissues known as aerenchyma characterized by low resistance to oxygen, thus providing saturated soil layers with necessary oxygen, while representing at the same time a pathway for methane transport from deep soil to the atmosphere (Guntenspergen et al., 2020;Laanbroek, 2010;Vroom et al., 2022).Furthermore, it has been shown that methane transported through plant tissues could be released through the stomata of wetland vegetation (Barba et al., 2019;Keppler et al., 2006;Nisbet et al., 2009;Villa et al., 2020).In addition, plants are a major contributor to labile carbon compounds that represent the primary substrate for methanogenesis in wetland top-soils (Tittel et al., 2019).On the other hand, radial oxygen loss through the roots of wetland plants alters the amount and depth distribution of oxygen in the soil, thus contributing to the dynamics of methanotrophy in the soil column (Laanbroek, 2010).Due in part to the multitude, diversity, and complexity of these plant processes, high variability exists in the resulting methane flux of the different wetland vegetation types even within the same wetland (Villa et al., 2020).This variability is particularly influenced by differing physiological characteristics of individual plant species as well as their respective phenology (Garnet et al., 2005;Helfter et al., 2022).Furthermore, methane production and oxidation are sensitive to soil chemical factors including pH and the presence of alternative terminal electron acceptors such as sulfate, nitrate, and Fe(III).In addition, methane production is sensitive to the supply of labile organic substrates, particularly derived from vegetation litter and root exudates.Because vegetation types and productivity as well as pH and soil chemistry can differ greatly between wetland types (e.g., fens, bogs, and coastal wetlands) methane emission factors can be greatly different between different wetland types (Turetsky et al., 2014).Even within a single wetland type (e.g., coastal marsh) different plants can have contrasting effects on soil redox state, leading to differences in methane emission factors (Noyce & Megonigal, 2021;Noyce et al., 2023).

Overview of Regional/Global Methane Flux Observation Data Sets
Incorporating methane models into ESMs has been a gradual process, largely attributed to the evolving understanding of methane emissions from wetlands over the past few decades.To discuss the modeling techniques knowledgeably, it is essential first to summarize the observational work on methane emissions, which forms the basis of these model frameworks.Observational methane flux data sets form the foundation for model optimization and parameterization, uncertainty evaluation, model validation, and identifying model shortcomings and missing processes through model intercomparisons (for example, WETCHIMP, Wania et al., 2013), and consequently, motivate and facilitate model development.These advancements have paved the way for modern modeling techniques to successfully simulate wetland methane dynamics.
The spatially heterogenous and temporally intermittent nature of methane emissions from wetlands (and other sources) makes these emissions challenging to measure and estimate globally.Traditionally, observations of methane accumulation rates in chambers, conducted manually in multiple locations or repeated automatically in fixed locations were used to estimate methane flux rates.These observations represent point measurements (in space and time) with relatively little replication within the global context.Nonetheless, data sets combining a large number of chamber observation campaigns were used to fit empirical or mechanistic models to determine emission factors per wetland types/biomes.These data sets were combined with spatial estimations of wetland extent from remote sensing (Bohn et al., 2015;Gerlein-Safdi et al., 2021;Kuhn et al., 2021;Melton et al., 2013) to generate regional and global button-up estimates of methane fluxes from wetlands (Bloom et al., 2017;Poulter et al., 2017;Saunois et al., 2016;Saunois et al., 2020).Advancements in remote sensing continuously improve the estimate of wetland extent (Jensen & Mcdonald, 2019), and have recently started estimating water table depth, water elevation, and vegetation type characterization to improve wetland detection and characterization (Burdun et al., 2023;Domeneghetti et al., 2018;Hess et al., 2015;Ju & Bohrer, 2022;Melton et al., 2022;Yazbeck et al., 2024).Recently, the development of fast methane gas analyzers, which can be used as a part of eddy covariance systems, has led to a significant increase in surface flux research related to wetlands.This advancement has notably heightened the interest of micrometeorologists in this field (Morin, 2019).While these studies started as individual efforts led by individual investigators, in the late 2010s, the global FLUXNET network for ecosystemscale methane monitoring using eddy covariance techniques, known as FLUXNET-CH4, was established (Knox et al., 2019).Such eddy covariance observations observe fluxes continuously in time and from a footprint area, much larger than the point observations represented by chambers.However, they cannot easily resolve different flux rates from different vegetation or hydrological patch types mixed within the footprint.It was recently demonstrated that networks of ground-based site-level observations can be up-scaled in space and time to cover large regions over long periods using machine learning approaches (Peltola et al., 2019).Despite its limitations, this eddy covariance flux observations network has provided a valuable database for studying methane cycling, refining methane emissions models, and improving LSMs (McNicol et al., 2023;Z. Zhang et al., 2023a).These approaches, based on scaling of direct ground-based observations of surface fluxes, are considered "bottom-up" approaches.Mechanistic models, such as LSMs, that predict fluxes based on aggregation and scaling of resolved surface fluxes, are also following a bottom-up approach.
Alternatively, "top-down" approaches utilize remote sensing (and ground-based) observations of methane concentrations dynamics in the atmosphere.Remote sensing of methane concentrations in the atmospheric column is available from instruments on satellites such as GOSAT, further interpreted to remove cloud-driven uncertainty and interpret column-integrated observations into vertically detailed data products (Parker et al., 2011;Yoshida et al., 2013).Methane concentration observations can also be based on ground-based measurements, collected into regional and continental observation networks often used to constrain the interpretation of the satellite observations.They then use mechanistic inversion models that trace the changes in methane concentrations upwind and infer surface fluxes by comparing the inversion model-generated expected concentration at a near-surface location with the observed concentration at that location (e.g., NOAA's CarbonTracker-CH4, Oh et al., 2023).See a review of top-down modeling approaches in Houweling et al. (2017).There is a well reported discrepancy between top-down and bottom-up methane emission estimates and an active research effort to understand its causes and reduce global methane flux estimates (Chang et al., 2023;Yazbeck & Bohrer, 2023).Some of the challenges in incorporating top-down estimates as data sources for parameterization/validation of bottom-up models are related to reconciling very different sources of error between uncertainties in the top-down satellite retrievals, and the bottom-up model representation errors (Ma et al., 2021).

History (1993-2023)
The critical mass of decades of observational discoveries have catalyzed recent innovations in wetland methane modeling.However, these advancements are built upon the foundational work of pioneering researchers.In this section, we trace the evolution of wetland and Earth system modeling, detailing the slow evolution to the frameworks that are widely recognized and utilized today.In the last two decades, they have significantly broadened their scope and are increasingly used as a decision tool (Blyth et al., 2021;Fisher & Koven, 2020).Xu et al. (2016) reviewed the development of mechanistic models (n = 40 with <5 being LSMs at the time) with a focus on the process representation.Here we focus on summarizing significant developments for global predictions highlighting the most recent developments and increased coupling within ESMs (Table 1).
Already in the late 1980s, significant strides in earth observations culminated in the development of global spatially explicit digital databases that detailed the distribution of wetland types and areas through more comprehensive ecological classification schemes (Aselmann & Crutzen, 1989;Matthews & Fung, 1987).This advancement not only enabled the synthesis of more precise estimates for global and regional methane emission from wetlands (Aselmann & Crutzen, 1989;Bartlett et al., 1990;Bartlett & Harriss, 1993;Matthews & Fung, 1987), but also facilitated the integration of atmospheric data with global three-dimensional atmospheric modeling (Fung et al., 1991).Data collection laid the groundwork for the development of process-based models in the late 1990s (Cao et al., 1996;Frolking & Crill, 1994;Segers & Kengen, 1998).An influential model was developed by Walter et al. (1996), which first combined distinct representation of methane production and oxidation with simulating three transport pathways.These models, grounded in mechanistic process-based principles, provided "bottom-up" estimates for regional and global methane fluxes.This approach stands in contrast with the previously used atmospheric transport models which invert atmospheric methane concentration observations to infer fluxes, known as the "top-down" approach (Fung et al., 1991;Taylor et al., 1991;Wahlen et al., 1989).
In the 2000s, efforts increased to incorporate existing process-based models of wetland dynamics into Global Climate models (Gedney et al., 2004;Shindell et al., 2004) as well as developing represetations in Dynamic Global Vegetation Models (DGVM) (Wania et al., 2009a;Wania et al., 2009bWania et al., , 2010)).These efforts yielded first estimates of climate feedbacks of wetland emissions as well as estimates of future emissions responding to changing environmental conditions (Gedney et al., 2004;Shindell et al., 2004).At the time, DGVMs already existed and facilitated studies of vegetation-climate interactions and carbon cycling on the global scale (Sitch et al., 2003) but despite resolving multiple land cover types and ecosystems, early DGVMs and other LSMs did not explicitly include wetlands.Wania et al. (2009aWania et al. ( , 2009b) ) provide examples for an early implementation of the physical and biological mechanisms specific to wetlands, which are required precursors for dynamic simulations methane emissions (Wania et al., 2010).Similarly, Riley et al. (2011) discussed the need to first describe hydrological patterns and wetland vegetation behavior before focusing on modeling methane emissions in a coupled model.In their case, though, that model does not explicitly includes a wetland land-cover type, but instead activates wetland processes in the fraction of the saturated soil columns of any ecosystem type.At that point in time, 10 modeling groups had developed regional or global wetland methane models and conducted a systematic wetland model comparison (Wania et al., 2013).The Wetland and Wetland CH 4 Inter-comparison of Models Project (WETCHIMP) yielded four major conclusions (Melton et al., 2013): (a) Models showed disagreement in predictions of wetland extent and emissions in space and time, (b) all models showed positive responses to climate change factors like atmospheric CO 2 concentrations, but differed in response to changes in temperature and precipitation, (c) Model validation was severely hampered by lack of flux observations and information on wetland extent, (d) a large range in flux predictions indicated substantial parameter and structural uncertainties in large scale wetland methane models.These findings have spurred increased research on dynamically mapping wetland extent (e.g., A. C. Zhang et al., 2022) and flux data syntheses (Knox et al., 2019;Turetsky et al., 2014) and we now see first systematic model evaluations against these data sets (Z.Zhang et al., 2023b).
However, systematic model intercomparison of wetland methane emissions are still rare (but see the Global Carbon Project wetland methane synthesis, e.g., Poulter et al. (2017), or a recent study focusing on cold season fluxes, Ito et al. ( 2023)).The Coupled Model Intercomparison Project (CMIP), which is part of the IPCC (current version CMIP6 accompanies the IPCC AR6) have not adopted 'dynamic modeling' for wetland methane emissions.In CMIP jargon, dynamic modeling means that methane atmospheric concentration is prescribed, based on target scenario, instead of surface flux, and surface flux (if resolved) does not vary with temperature or other environmental drivers of the methane cycle.As such, CMIP models do not include feedbacks between methane emissions and climate.Arguably, fully coupled ESMs now include biogeochemical processes regulating methane emissions in wetlands, incorporating feedbacks between climate dynamics and the global carbon cycle (Folberth et al., 2022;He et al., 2020;Nazarenko et al., 2022;Nzotungicimpaye et al., 2021).Similar to LSMs in general (Fisher & Koven, 2020), wetland modeling schemes are now becoming more complex, focusing on aspects across all temporal and spatial scales for example, either on incorporating detailed reaction networks (Sulman et al., 2023) or on coupling with the atmosphere (Folberth et al., 2022).

Spatial Structure of LSMs and Its Impact on Incorporating Wetland Dynamics
The complex interaction between atmospheric, hydrological, ecological, and microbial processes happening at a high spatial-temporal resolution, embedded within a strong variability within and between wetlands, pose challenges for modeling wetland dynamics in LSMs (standing alone, at site-level simulations, or as a component of ESMs' global simulations) (Chang et al., 2023).LSMs are designed to simulate land-surface processes from local to global scales and to be directly coupled to atmospheric dynamics models.As a result, the spatial resolution of LSMs is largely driven by the resolution of atmospheric dynamics models, which in most current ESMs is on the order of 0.5-2°(50-200 km).At this scale, lateral hydrological exchanges are considered to be unimportant in most models, and thus LSMs typically conceptualize soil processes as a 2-D array of 1-D vertical columns and typically neglect the lateral exchanges below ground.This requirement raises challenges for representing the fine-scale landscape heterogeneity that is typical of wetland ecosystems.A common solution is dividing large LSM grid-cells into spatially implicit fractions representing sub-grid heterogeneity (Clark et al., 2015).For example, a grid cell could be divided into forest, grassland, cropland, and wetland fractions, each of which are simulated as an individual one-dimensional column (soil with associated vegetation types), each with an associated fractional area of the total grid cell.Carbon, nitrogen, water, and energy cycling calculations are conducted separately for each sub-grid fraction, and the surface-atmosphere exchanges of the fractions are then scaled by fractional area and combined to calculate the land-atmosphere exchanges at the scale of the entire grid cell.Models vary in the structure of this sub-grid tile approach.In some models the vegetation is resolved in sub-grid tiles representing different plant functional types, but they all share a single soil column.Another alternative has been to use a hillslope methodology within a grid cell that allows for 1-D hydrologic redistribution within the cell (Clark et al., 2015).
A key aspect of these approaches is the spatially implicit representation of heterogeneity: Each sub-grid fraction is conceptualized as a representative of that land type, and the 2-D spatial arrangement and horizontal connections between sub-grid fractions are not represented.This approach is scalable and computationally efficient because it minimizes communication of data across sub-grid fractions and does not require detailed information about spatial patterning.However, it also makes it difficult or impossible to represent horizontal flows of water, carbon, and nutrients among upland and wetland land cover types within a grid cell.The omission of horizontal exchange may introduce errors in carbon and water cycling, especially as models move to higher resolution (but see recent attempts to resolve it, R. L. Zhang et al., 2022).
Specifying wetland area fractions within a grid cell is a major challenge for LSMs, because wetland areas emerge dynamically through a complex interplay of hydrology, small-scale and large-scale topography, and subsurface properties (including subsurface water flows) within a landscape.Some models estimate wetland extent dynamically by calculating the inundated fraction of the land surface as a function of hydrological state and landsurface characteristics (e.g., mean slope within a grid cell) (for example, Riley et al., 2011).An alternative approach prescribes wetland extent based on land-surface data sets or remote sensing (for example, Spahni et al., 2011) (see a review of data sources for wetland extent in Fluet-Chouinard et al. ( 2023)).This approach eliminates the need for dynamic calculations of wetland areas within a grid-cell but does not allow models to represent dynamic changes in wetland areas driven by processes such as climate change-derived changes in precipitation intensity, riverine flooding, or sea level rise.Dynamic wetland area fraction calculations allow models to simulate more rapid changes in methane production from landscapes that are intermittently flooded.However, ESMs and LSMs are structured as vertical grid columns and lack watershed-scale lateral transport and hydrological processes (but see recent developments in coupling a watershed hydrology model with biochemical processes and an (Adebayo et al., 2023)), which often leads to errors in the resolved inundated area.Another challenge for LSMs in dealing with rapid changes in wetland area fractions, is that they assume different plant functional types in dry uplands than in wetlands.The rapid (inter and intra-annual) changes in classification between flooded and dry conditions make it very difficult for the models to predict which vegetation functional type should be established in such locations with intermittent flooding.This is because models typically adopt a long-term ecological succession approach to diagnoseg vegetation type in each land unit.In practice, this means that dynamic wetland area calculations are typically used with methane emission models that estimate methane production as a function of upland carbon fluxes (for example, Riley et al., 2011) rather than representing wetlands as separate sub-grid units with persistent properties that differ from uplands.This approach makes it difficult to represent systems like peatlands that develop different properties from uplands (such as large peat carbon pools) over long periods of time.One potential solution for the tradeoff between dynamic wetland areas and process-based wetland representation would be to include sub-grid units separately representing permanently and intermittently flooded areas or representing topographically defined units, such as height above nearest drainage (Chaney et al., 2018).An additional challenge is representing the diversity of wetland types, which include fens, bogs, and swamps with a range of hydroperiods that can drive orders of magnitude differences in methane emissions.This raises a challenge for LSM approaches that classify a single category of wetland area within each grid cell.Over the years, models have advanced in their approach to solving these challenges, through advancements to the numerical structural representation, and to the process-level resolution they allow (Table 2).
In the typical LSM, wetlands are a sub-grid-scale process.The fraction of the grid area that is inundated (i.e., the wetland) is either defined using existing data sets, forced by remote sensing data, or simulated using the hydrological sub-model (Melton et al., 2013).Table 2 summarizes the development of different approaches to resolving wetlands and processes within them.Examples of LSMs following the early approach to wetland representation as a sub-grid process associated with the flooded sub-column of the soil, include: CLM and ELM (Riley et al., 2011), ORCHIDEE-LSM (Ringeval et al., 2010), CLASS-CTEM (Arora et al., 2018), LPJ-WSL (Hodson et al., 2011), SDVGM (Hopcroft et al., 2011), DLEM (Tian et al., 2010) and many others (Melton et al., 2013).Other approaches, such as the one used in ELM-SPRUCE, include multiple resolved patch types within a wetland land unit and include lateral flow between patches, though specifically ELM-SPRUCE, which is developed to resolve bog/fen ecosystems, only allows two patch types (Ricciuto et al., 2021).In typical applications, the parameters that characterize CH 4 and CO 2 dynamics are retrieved from incubation experiments, and in some models, different parameter sets represent different pathways for methanogenesis (hydrogenotrophic, methylotrophic, and acetoclastic) and methanotrophy (aerobic, and anaerobic) (Xu et al., 2015).More detailed modeling approaches could be found in ecosystem models that focus on soil biogeochemistry, such as ecosys (Grant & Roulet, 2002) and DNDC (Li, 2000), which explicitly represent microbial activities in the wetland sediments and represent the mechanistic carbon-nutrient interaction between soil and vegetation, though such models are usually restricted to site-level simulations.

Process Representation
One notable simplification in ecosystem models is the omission of explicit representations of microbial populations.Instead, these models frequently employ various shortcuts to bypass the need for detailed microbial simulations, and the degree of process-level detail by which they represent such processes have advanced over the years (Table 2).One approach used for use in the Community Land Model (CLM) involved using a fraction of the total carbon uptake of an ecosystem as a proxy to estimate the maximum methane production rate within a specific grid cell (Riley et al., 2011).This, however, has also been a point of advancement in the field over time.Alongside the ESM context focused on scaling models regionally or globally, there has been a significant evolution in the methodologies for modeling the subsurface microbial dynamics responsible for soil carbon decomposition (Chandel et al., 2023), as well as similar considerations for microbial processes relating to methane cycling and transport (Grant, 1998;Ricciuto et al., 2021).In terrestrial systems the focus and motivation lie primarily in improving the temperature sensitivity of soil decomposition, a factor that directly impacts prediction of CO 2 release with global warming.This seems to be achieved better in 'microbial' soil decomposition modules, which incorporate substrate limitation of decomposition, than 'traditional' ones relating CO 2 production to carbon pool size (Todd-Brown et al., 2012).The same is true for decomposition under anaerobic conditions favoring methane release (Chang et al., 2020).However, simulation of net methane emissions is impacted by more than one microbial process, which all have their respective temperature sensitivities and substrate limitations.For example, methane oxidation rates are a function of temperature, methane, and respective oxidant concentrations • More details can be found in Dalva et al. (2001), Frolking and Crill (1994), Moore and Dalva (1993), and Moore and Roulet (1993) Process-based models • Models methane concentration in the soil • Methane production is function of NPP or heterotrophic respiration • Resolves typically three methane transport pathways: ⚬ Ebullition ⚬ Diffusion ⚬ Plant-mediated transport • Some add more details to processes: competition for oxygen and redox potential, inundation lag, and temperature • Inundation is typically forced or modeled and not derived from hydrology sub-modules • Extensive list of models with their corresponding methane processes can be found in Xu et al. (2016); and Melton et al. ( 2013) • Details on inundation dynamics can be found in Wania et al. (2013) Process-based models resolving microbial populations • Explicitly represent microbial activities in the wetland sediments • Represent the mechanistic carbon-nutrient interaction between soil and vegetation • Restricted to site-level simulations • Examples can be found in Grant and Roulet (2002) and Li (2000) Journal of Geophysical Research: Biogeosciences 10.1029/2023JG007915 FORBRICH ET AL. (Segers & Kengen, 1998).More explicit representations of microbial kinetics have been developed through the incorporation of microbial functional type groups.These groups are characterized by their specific growth and death rates, and their dynamics are regulated based on substrate availability and competition with other microbial groups for resources (Grant, 1998;Segers & Kengen, 1998;van Bodegom et al., 2001).Additionally, the application of Equilibrium Chemistry Approximation (ECA) kinetics has been pioneered (J.Y. Tang & Riley, 2013), which also offers a more realistic representation of microbial dynamics, while allowing for numerical accuracy and still representing broad microbial networks.In addition to employing these generalized and simplified Monod kinetics framework networks, ecosystem models typically modulate the maximum production rates of key reactions in response to critical environmental factors.These adjustments account for the temperature sensitivity of processes, hydrologic responses, and pH variations, reflecting the observed sensitivity of methane production to these factors (Chadburn et al., 2020;Segers, 1998;Whalen, 2005).This is commonly accomplished using response curves that may be based on physics (e.g., Arrhenius or Q10 curves to model microbial activity response to temperature, for instance) or empirical relationships.
Detailed reaction networks are only very recently being incorporated into LSMs (Ricciuto et al., 2021;J. Wang et al., 2024).Methane production is the final step in carbon oxidation, and earlier steps in the decomposition process have been less frequently included.Notably, the processes responsible for creating the substrates used by methanogens, such as acetate or a combination of CO 2 and H 2 , are also commonly represented (Grant, 1998;Lovley & Klug, 1986;Segers & Kengen, 1998;van Bodegom et al., 2001).
Additionally, the role of competing terminal electron acceptors, such as iron, manganese, and sulfur, is recognized as significant in wetland methane cycling (Segers & Kengen, 1998).Despite their importance, these elements are not yet comprehensively incorporated into mainstream ESM-based wetland models (but see Sulman et al. (2023)).
While in terrestrial models CO 2 release is simulated as diffusion, simulating wetland methane transport is more complex (see above).Influenced by early process-based models (Walter et al., 1996), LSM parametrizations typically represent all three transport pathways.All of them need to be linked to the microbial processes controlling gas concentrations.Most LSMs simulate soil processes in vertical layers, which allows for vertical gas transport of CO 2 and methane.Transport via bubbles is more dynamic and depends typically on thresholds of either concentrations, pressure or gas volume (Peltola et al., 2018).Plant-mediated transport of methane is approximated as a function of prescribed root density and stem diameter and density for a specific plant functional type (Ricciuto et al., 2021;Riley et al., 2011), while aerenchyma transport of oxygen is typically not considered.How to incorporate more detailed representation of different plant structure and function in LSMs is currently being debated (LaFond-Hudson & Sulman, 2023).

Remaining Challenges
While ESMs are generally effective in elucidating global fluxes, there is a widespread recognition of their limitations and substantial opportunities for improvement that could enhance their performance.For example, a major remaining challenge for both modeling and observation networks is the consistent discrepancy between bottom up and top-down models (Chang et al., 2023).As long as this significant problem persists, there remains ample scope to innovate and refine ESMs for elucidating methane emissions from wetlands.In this section, we outline specific areas recognized by the field as needing ongoing research and development.
One particular area that has garnered considerable attention is the growing recognition of the complexity within the microbial consortium responsible for methane emissions.In the past couple decades, continual discoveries of new microbial lineages are reshaping our understanding of the methane cycle, revealing impacts that were previously unrecognized.For example, Chadburn et al. (2020) demonstrated that the temperature response of methane fluxes cannot be predicted without representation of the different microbial processes that contribute to methane generation and oxidation.The discovery of high affinity methanotrophs suggests that methane emissions from thawing permafrost might be significantly lower than earlier estimates (Oh et al., 2020).Similarly, sulfur cycling in freshwater systems plays a crucial role in methane regulation (Gwak et al., 2022), suggesting the need for more complex microbial networks in modeling contexts.However, the integration of these new kinetic pathways into models is a time-consuming and comprehensive process, presenting significant challenges.This is particularly true given the rapid pace of discovery facilitated by advanced omics-based microbial techniques.The rate at which new pathways are being uncovered often outpaces the ability of modelers to test and evaluate them for inclusion in existing models.
Similarly, structural issues to better simulate wetland dynamics remain, ranging from more dynamic hydrology interacting with more detailed biogeochemistry (e.g.,, the interaction between precipitation, temperature, permafrost dynamics and methane generation, Neumann et al., 2019) to vegetation dynamics.Moving forward, more detailed spatial information on hydrology will become available to simulate the spatial extent of wetlands (see above).In the same context, there is continuing effort to dynamically simulate surface and subsurface water flow across grid cells (Bisht et al., 2018;Felfelani et al., 2021;Xu et al., 2023), which should also help to improve simulation of dynamic wetland extent.However, with the improved hydrology on large scales, it will be increasingly important to separate wetlands from other freshwater aquatic systems that display different patterns and mechanisms in their methane emissions (Lauerwald et al., 2023;Rosentreter et al., 2021).While many empirical studies show differences in carbon flux responses to hydrology between wetland types, such as fens and bogs (Sulman et al., 2010), or other freshwater wetland types (Knox et al., 2021), most LSMs do not distinguish between different wetland types, and amongst the few that do, there is a large variation in the level of detail of wetland type representation.A frequent argument against such sub-type representation is the need for parametrizing several new processes without expanding uncertainties related to these parametrizations (Müller et al., 2015).One possible way forward here is to aggregate wetland-specific traits related to methane cycling in the development of plant functional types (for example, LaFond-Hudson & Sulman, 2023).This concept is traditionally used to describe important ecosystem processes, like photosynthesis and biomass production and allocation, on adequate scales in LSMs (Sulman et al., 2021;Wullschleger et al., 2014).Wetland-specific functional types, which parallel the model representation of dryland vegetation variability through plant functional types (PFTs), have been developed for northern wetlands, where there is a need to characterize moss function as well as flood-tolerant graminoids (Ricciuto et al., 2021;Wania et al., 2009a).However, to simulate wetland vegetation in other climate zones, it seems likely that the number of functional types will have to increase (Yazbeck & Bohrer, 2023), including woody vegetation (Pangala et al., 2017).As for the terrestrial biosphere in general, it is difficult to include human activities and their impact on biogeochemical processes.In the case of wetlands, we need to consider rice agriculture (McDermid et al., 2017)

Modeling
Fortunately, numerous active modeling groups are diligently striving to enhance methane representation in these model families.Their efforts are directed not only towards the areas of improvement we have identified but also towards rectifying clear inaccuracies in the current depiction of methane fluxes more broadly.While ESMs vary in their treatment of wetlands and the methane cycle, there are overarching themes in these representations.
Consequently, certain research directions broadly enhance modeling capabilities across the field.In this context, we spotlight key developments in individual models that have universally advanced the prediction of methane fluxes from wetlands.Globally, one major motivation to simulating wetland methane emissions accurately is to predict atmospheric methane concentrations, an important model development is to dynamically couple the surface module with atmospheric chemistry (Blyth et al., 2021;Folberth et al., 2022;Heimann et al., 2020).A fully coupled ESM should be able to predict fluctuations in atmospheric methane concentration in the past (Kleinen et al., 2023).In combination with more global data sets for validation (see below), it becomes possible to identify global hotspots as well as regions where large discrepancies occur between model and observations (Parker et al., 2022).Currently, no land model distinguishes between wetland types such as fen, bog, marsh, or swamp, neither with regard to hydrology nor with regard to nutrient status and vegetation community.However, first steps are being taken to establish a representation of microtopography and varying water level (Shi et al., 2015), realizing that particularly in northern wetlands methane emissions vary on area basis following hummockhollow patterns.There are other model improvements with regard to hydrology, that will help modeling wetland dynamics in the future, e.g., specifying surface water transport (Getirana et al., 2021;X. Lu et al., 2016;Parker et al., 2022;Ricciuto et al., 2021), or parametrizing floodplain dynamics (Schrapffer et al., 2020), and coastal processes (LaFond-Hudson & Sulman, 2023;O'Meara et al., 2021O'Meara et al., , 2024;;Ward et al., 2020).With more experiments and measurements (see below), improvements have been made regarding the representation of microbial processes and transport mechanisms.For example, Xu et al. (2015) separately simulate two methane production pathways as well as aerobic and anaerobic methane oxidation.Furthermore, explicit representation of soil redox conditions allow simulating alternative anaerobic decomposition and methane oxidation pathways which have the potential to reduce methane production and release (Sulman et al., 2022).In addition, lateral flow (Shi et al., 2015) can be extended to include lateral transport of carbon compounds, including methane (Ricciuto et al., 2021).All of this progress in process detail has to be balanced by increased parameter uncertainty for some of the processes.Finally, temporal dynamics of wetland methane emissions vary strongly across wetland types and climate zone (Knox et al., 2021).Data availability varies greatly across climate zones, with less observational representation in the tropics and arctic regions.This fact compounds with the functional differences between wetlands in different climate zone.For example, process level representation of arctic peatlands must account for permafrost dynamics, while sap flow, xylem transport in tree stems, and canopy turbulence and light dynamics must be represented in tropical swamps.These functional differences lead to the development of biomespecific model solutions.For example, Ricciuto et al. ( 2021) was designed specifically for northern peatlands (albite, without permafrost).Most models can realistically simulate monthly and annual flux behavior in northern wetlands, while emissions are less predictable in temperate and tropical wetlands (Z.Zhang et al., 2023b).However, model parametrization based on northern wetlands does not work well in tropical settings and for different plant functional types (Yuan et al., 2023).This is further indication towards the need to better represent different wetland types and their climatic responses in LSMs.

Data for Forcing and Validation
On regional and global scales, increased effort is being put into mapping wetland area extent of different wetland types (Fluet-Chouinard et al., 2023;Gerlein-Safdi et al., 2021;Gumbricht et al., 2017;Hess et al., 2015;Jensen & Mcdonald, 2019;Murray et al., 2022;Prigent et al., 2020).For a long time, northern latitudes were vastly underobserved in terms of inundation extent, existing land cover types, and methane fluxes.Recent research efforts for sampling in the Arctic, combined with machine learning algorithms are starting to fill in information over the Arctic (Melton et al., 2022;Xi et al., 2022).One improvement is the separation of wetlands from other freshwater aquatic systems (E.F. Zhang et al., 2021) to distinguish these two important natural methane sources (Rosentreter et al., 2021).With targeted remote sensing missions (e.g., NASA's new Surface Ocean and Water Topography (SWOT) mission (Papa & Frappart, 2021)), high resolution data becomes available to improve mapping of transitional/seasonal wetlands, particularly in the tropics (Denbina et al., 2019;Morris et al., 2019).For example, assimilating soil moisture observations from the Soil Moisture Active Passive (SMAP) satellite mission into an LSM improved flux predictions in wetlands, primarily by increasing wetland extent (A. C. Zhang et al., 2022).It remains challenging to separate wetlands from other freshwater ecosystems, though that remains a necessity to avoid 'double-counting' of methane emissions during up-scaling.
At the same time, improvements in data needed to inform and parameterize models has dramatically improved in recent years due to concerned efforts from the micrometeorological community.Data synthesis efforts have led to standardized regional and global flux data sets, such as FLUXNET-CH4, and CarbonTracker-CH 4 (Delwiche et al., 2021;Knox et al., 2019;Kuhn et al., 2021;Murguia-Flores et al., 2023;Oh et al., 2023;Turetsky et al., 2014), which can be used for model validation.An increased focus lies on extending measurements into the tropics.Ecosystem-level flux studies are more established in northern latitudes but are relatively lacking in the tropics (Sakabe et al., 2018;A. C. I. Tang et al., 2018).

Recommendations for Future Work
Despite the ongoing, commendable efforts to enhance methane representation in ESMs and the current rapid pace of improvement, there are several principle-based areas that likely need more attention to further refine methane representations.Below, we propose a non-exhaustive list of key areas that should be prioritized for future enhancements in these model frameworks.
• Wetland morphological type/typology accounting: Emphasizing the integration of diverse wetland categories like bogs, fens, and coastal wetlands is recommended.This involves leveraging advanced remote sensing, refining biogeochemical representation (as discussed below) and addressing dynamic hydrology considerations.• More detailed process representations: The development of models and integration of microbial processes and transport mechanisms into process-based models is recommended to capture the complexities of the methane cycle to ensure accurate representation of methane production, consumption, and emissions.Thus, it is important to support research in enhancing the mechanistic understanding of biogeochemical processes in wetlands at fine scales.• Increased data availability and integration, with a particular emphasis on remote sensing: Addressing data limitations requires support for initiatives aimed at improving the accuracy of data sets used in model validation.Additionally, there should be emphasis on advancing satellite missions and remote sensing technologies to enhance the mapping of wetland areas, especially in tropical regions.
• Increased emphasis on model intercomparison and ensemble-based approaches: Continued support for systematic intercomparisons, exemplified by initiatives, such as WETCHIMP (Bohn et al., 2015;Melton et al., 2013) and the Global Carbon Project wetland methane synthesis (Poulter et al., 2017;Saunois et al., 2020) is imperative for identifying and reducing uncertainties in wetland methane modeling.Additionally, the establishment of advanced intercomparison protocols that include dynamic methane flux predictions by ESMs is essential for diagnosing and addressing errors in both model structure and parameterization.

Summary
With the continued urgency of climate change and the uncertainty regarding the rates of increasing methane concentrations in the atmosphere, accurate predictions of methane surface fluxes are essential, especially from biogenic sources where a lot of uncertainty lies due to the complex interaction of the biogeochemical, hydrological, microbial, and ecological processes behind methane production, consumption, and transport.Over the last decades, significant efforts have focused on representing wetland-methane dynamics in ESMs, though discrepancies persist between different models' predictions, whether aggregated by type of model or temporal or spatial resolution.Current models' advancement are targeting these multiple issues by more detailed representation of methane dynamics, such as coupling microbiology, microtopography, and hydrology.Concurrently, advancements in observations are made through expanding ground-based measurements networks, such as FLUXNET-CH4, and high resolution remote sensing mapping of relevant variables, such as water levels and methane concentration.Such advancements provide more robust data sets for parameterizing and validating wetland-resolving models, and consequently improving their corresponding performance.

Table 1
Timeline of Key Events and Developments in Wetland Methane Studies From the 1970s to the 2020s

Table 2
Methane Models Summary Within LSMs