Uncertainties in wetland methane‐flux estimates

The wetland at Old Woman Creek National Estuarine Research Reserve, with flux tower US-OWC, which is characterized by high methane fluxes, high spatial heterogeneity, dynamic hydrology and water level fluctuations, and high lateral transport of dissolved organic carbon and nutrients, embodies many of the challenges in modeling methane fluxes.


| U N ITE D S TATE S
Wetlands are the largest emitter of biogenic methane and store the largest amount of carbon per area of all terrestrial biomes. Therefore, understanding the future state of the climate requires accurate prediction of fluxes from wetlands. Unfortunately, wetlands account for the largest source of uncertainty in global methane budget predictions, where substantial divergence exists between bottom-up and top-down model predictions (Saunois et al., 2020).
Bottom-up models simulate surface fluxes based on corresponding sinks and source rates per area source type and up-scale them based on estimates of the area coverage of each source/sink type to get global estimations. Such models use process-based techniques to predict methane flux from different source types and locations. A large portion of their uncertainty is attributed to the lack of explicit representation of some of the mechanisms responsible for specific pathways and responses in methane dynamics. Another source of error for bottom-up estimates is the lack of accurate information about wetland extent and especially small wetlands and ephemeral flood plains. A new generation of water-sensing satellites, including the very recent SWOT mission, intends to resolve this uncertainty.
Top-down models use atmospheric observations of methane concentration combined with inverted atmospheric transport models (i.e., simulating where each air parcel came from) through a mass-budget approach assuming that the presence of any "missing" (positive or negative) concentration indicates the presence of a surface flux that contributed that missing source or sink. Their estimate uncertainties are tied to the different applications of atmospheric transport and optimization methods that may lead to significant discrepancies in flux estimations. Inversion modeling of estimated fluxes is confined to the model resolution of the surface, where all sources within a simulation grid are considered as one single source. Another important factor constraining top-down estimation precision is that the observed concentration used to drive the models are provided through remote sensing approaches, such as tall towers, aircraft, or satellite methane measurements. Inversion model precision is dependent on the uncertainty of these input concentration observations, which may be obstructed due to clouds or aerosols in the measurement pathway, and on their spatial resolution and their revisit time of the same pixel, which are typically low (Qu et al., 2021).
To investigate models' uncertainties, Chang et al. (2023) performed an inter-comparison study of multiple top-down and bottom-up models used for site-level and global methane emissions estimates. The open-source International Land Model Benchmarking (ILAMB) system was used for inter-model evaluation. Such a benchmarking system provides a normalized and consistent platform for ranking models' performance. It was shown that by restricting the analysis to the best performing models, uncertainty within the ensembles of both bottom-up and top-down models was reduced.
However, the discrepancy between top-down and bottom-up estimate from these selective ensembles of best performing models increased. Another key finding of Chang et al. (2023) is the necessity of refining inter-model variability at the grid-cell level. In bottom-up models, increased modeling complexity and resolution did not necessarily lead to better results, thus shedding light on the current poor state of knowledge and representation of biogeochemical processes responsible for methane dynamics in wetlands at high spatial and temporal resolutions. Chang et al. (2023) find that the inclusion of observational constraints would improve bottom-up and top-down models' precision but not necessarily result in higher accuracy or convergence. To bridge that gap between top-down and bottom-up estimates, three key points must be considered in future modeling and data acquisition development: (1) Higher representativeness of wetland sites in global datasets associated with fine spatial and temporal resolutions, (2) more in-depth study and model representation of site-level and patch-level environmental conditions, pathways, and processes driving methane emissions, and (3) improving our understanding and representation of atmospheric methane dynamics.
Challenges in estimating methane emissions lie in the fact that wetlands host a dynamic combination of biogeochemical processes driven by complex interactions between ecological, meteorological, mineral, hydrological, and microbial processes. The complexity of wetland processes is compounded by strong small-scale spatial variability of the ecology and hydrology within wetlands, such that even small wetlands display large variations in flux rates among ecohydrological patches (Morin et al., 2017). Such small-scale spatial and temporal heterogeneity within a wetland is not commonly resolved by global models. Developing an eco-hydrological patch-level resolution in wetland models. Similar approach that is currently used to represent dryland vegetation could address some of these problems.

| Representativeness of wetland sites
Observation location bias is another key barrier to understanding global fluxes. Observations datasets need to be more geographically balanced, where some regions need to be better sampled and represented, including the African Sahel, the tropics, and Australia, where towers are rare or might exist but are not integrated in global networks of eddy-covariance (EC) sites such as Fluxnet . There is a further need for improving our understanding of the mechanisms of microbial, hydrological, meteorological, and ecological interactions at high spatiotemporal resolution (Ganesan et al., 2015). Such investigations should include better monitoring for the spatiotemporal variation of methane flux drivers among ecohydrological patch types and generally within wetlands, whose total fluxes are observed from EC towers. Process-level observations, such as direct measurements of the conductivity of plants to transport methane should be done at the patch level (Villa et al., 2020), to inform models. Such patch-level flux and process-parameter observations are necessary for advancing model developments toward refined, eco-hydrological patch-level, mechanistic representation of methane dynamics in wetlands and their corresponding drivers while improving the coupling between biogeochemical, hydrological, and ecological submodules (Chang et al., 2021).

| Model representation of spatial variability, pathways, and processes driving methane emissions
EC flux observations are "vertically biased," that is, they only include the flux upward from the local footprint to the atmosphere.
They cannot measure lateral fluxes of dissolved methane that leave the site and flow downstream, to be released to the atmosphere farther off-site. Similarly, bottom-up models function in vertically resolved columns. Most models do not resolve the lateral transport of dissolved methane or the lateral import of dissolved carbon that feeds the methanogenesis (but see Bisht et al., 2018). Accounting for aquatic lateral transport may be critical to improving model accuracy, especially in areas where low water quality is indicative of high rates of lateral transport of dissolved organic carbon.

| Improving our understanding and representation of atmospheric methane dynamics
Improved mapping of methane concentration is required for improved accuracy and precision of top-down estimates. This could be achieved by increasing remote sensing resolution and by reducing the corresponding satellite return time (Qu et al., 2021). Such alternatives would provide models with more robust data used in the inversion analysis, which in turn needs to consider finer resolution in their transport and atmospheric chemistry submodules. The prior knowledge used in inversion models needs to be improved to reduce its corresponding uncertainty, where the special focus needs to be given for inundation extent, temperature dependence, and carbon decomposition (Bloom et al., 2017). Atmospheric chemistry and decay of methane introduce additional uncertainty and may be improved by added observations of reactive gases, such as hydroxyl radicals (Saunois et al., 2020). Improving the partitioning of methane sources is another important factor in tuning top-down model estimations, where isotopic measurements can be integrated within inversion models. In addition, gridded products of natural sources of methane need to be updated and integrated into models, while making use of the current and future satellite missions with highresolution measurements.
Advancements in site observations and model development need to account for wetland-type heterogeneity. Chang et al. (2023) showed that benchmarking was sensitive to wetland types (swamp, fen, wet tundra, bog, etc.). The strength and impact of site-specific methane flux drivers, such as tidal inundation, saltwater intrusion, lateral water flow, snow effect, and ice thaw vary strongly from site to site. The sensitivity of methane processes to these drivers should be investigated using site-level and remote sensing observations and integrated into models while allowing the models to represent the existing mosaic of wetland type and their associated parameters . The choice of benchmarking datasets should be more representative of wetland-site diversity and corresponding environmental conditions and future methane models need to account for these variabilities.
The main takeaway of the study presented by Chang et al. (2023) is that improving the observational constraints did not reduce the divergence between bottom-up and top-down estimates. Improvement of the underlying formulation and functionality of models, such as the representation of within-wetland heterogeneity in bottom-up emission and improved representation of atmospheric chemistry during transport may provide key steps toward closing the gap.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing not applicable to this article as no data were generated or analysed for this commentary.