Abstract. Methane (CH4) emissions from natural landscapes constitute roughly half of global CH4 contributions to the atmosphere, yet large uncertainties remain in the absolute magnitude and the seasonality of emission quantities and drivers. Eddy covariance (EC) measurements of CH4 flux are ideal for constraining ecosystem-scale CH4 emissions due to quasi-continuous and high-temporal-resolution CH4 flux measurements, coincident carbon dioxide, water, and energy flux measurements, lack of ecosystem disturbance, and increased availability of datasets over the last decade. Here, we (1) describe the newly published dataset, FLUXNET-CH4 Version 1.0, the first open-source global dataset of CH4 EC measurements (available at https://fluxnet.org/data/fluxnet-ch4-community-product/, last access: 7 April 2021). FLUXNET-CH4 includes half-hourly and daily gap-filled and non-gap-filled aggregated CH4 fluxes and meteorological data from 79 sites globally: 42 freshwater wetlands, 6 brackish and saline wetlands, 7 formerly drained ecosystems, 7 rice paddy sites, 2 lakes, and 15 uplands. Then, we (2) evaluate FLUXNET-CH4 representativeness for freshwater wetland coverage globally because the majority of sites in FLUXNET-CH4 Version 1.0 are freshwater wetlands which are a substantial source of total atmospheric CH4 emissions; and (3) we provide the first global estimates of the seasonal variability and seasonality predictors of freshwater wetland CH4 fluxes. Our representativeness analysis suggests that the freshwater wetland sites in the dataset cover global wetland bioclimatic attributes (encompassing energy, moisture, and vegetation-related parameters) in arctic, boreal, and temperate regions but only sparsely cover humid tropical regions. Seasonality metrics of wetland CH4 emissions vary considerably across latitudinal bands. In freshwater wetlands (except those between 20∘ S to 20∘ N) the spring onset of elevated CH4 emissions starts 3 d earlier, and the CH4 emission season lasts 4 d longer, for each degree Celsius increase in mean annual air temperature. On average, the spring onset of increasing CH4 emissions lags behind soil warming by 1 month, with very few sites experiencing increased CH4 emissions prior to the onset of soil warming. In contrast, roughly half of these sites experience the spring onset of rising CH4 emissions prior to the spring increase in gross primary productivity (GPP). The timing of peak summer CH4 emissions does not correlate with the timing for either peak summer temperature or peak GPP. Our results provide seasonality parameters for CH4 modeling and highlight seasonality metrics that cannot be predicted by temperature or GPP (i.e., seasonality of CH4 peak). FLUXNET-CH4 is a powerful new resource for diagnosing and understanding the role of terrestrial ecosystems and climate drivers in the global CH4 cycle, and future additions of sites in tropical ecosystems and site years of data collection will provide added value to this database. All seasonality parameters are available at https://doi.org/10.5281/zenodo.4672601 (Delwiche et al., 2021). Additionally, raw FLUXNET-CH4 data used to extract seasonality parameters can be downloaded from https://fluxnet.org/data/fluxnet-ch4-community-product/ (last access: 7 April 2021), and a complete list of the 79 individual site data DOIs is provided in Table 2 of this paper.

Abstract. Methane (CH4) emissions from natural landscapes constitute roughly half of global CH4 contributions to the atmosphere, yet large uncertainties remain in the absolute magnitude and the seasonality of emission quantities and drivers. Eddy covariance (EC) measurements of CH4 flux are ideal for constraining ecosystem-scale CH4 emissions due to quasi-continuous and high-temporal-resolution CH4 flux measurements, coincident carbon dioxide, water, and energy flux measurements, lack of ecosystem disturbance, and increased availability of datasets over the last decade. Here, we (1) describe the newly published dataset, FLUXNET-CH4 Version 1.0, the first open-source global dataset of CH4 EC measurements (available at https://fluxnet.org/data/fluxnet-ch4-community-product/, last access: 7 April 2021). FLUXNET-CH4 includes half-hourly and daily gap-filled and non-gap-filled aggregated CH4 fluxes and meteorological data from 79 sites globally: 42 freshwater wetlands, 6 brackish and saline wetlands, 7 formerly drained ecosystems, 7 rice paddy sites, 2 lakes, and 15 uplands. Then, we (2) evaluate FLUXNET-CH4 representativeness for freshwater wetland coverage globally because the majority of sites in FLUXNET-CH4 Version 1.0 are freshwater wetlands which are a substantial source of total atmospheric CH4 emissions; and (3) we provide the first global estimates of the seasonal variability and seasonality predictors of freshwater wetland CH4 fluxes. Our representativeness analysis suggests that the freshwater wetland sites in the dataset cover global wetland bioclimatic attributes (encompassing energy, moisture, and vegetation-related parameters) in arctic, boreal, and temperate regions but only sparsely cover humid tropical regions. Seasonality metrics of wetland CH4 emissions vary considerably across latitudinal bands. In freshwater wetlands (except those between 20∘ S to 20∘ N) the spring onset of elevated CH4 emissions starts 3 d earlier, and the CH4 emission season lasts 4 d longer, for each degree Celsius increase in mean annual air temperature. On average, the spring onset of increasing CH4 emissions lags behind soil warming by 1 month, with very few sites experiencing increased CH4 emissions prior to the onset of soil warming. In contrast, roughly half of these sites experience the spring onset of rising CH4 emissions prior to the spring increase in gross primary productivity (GPP). The timing of peak summer CH4 emissions does not correlate with the timing for either peak summer temperature or peak GPP. Our results provide seasonality parameters for CH4 modeling and highlight seasonality metrics that cannot be predicted by temperature or GPP (i.e., seasonality of CH4 peak). FLUXNET-CH4 is a powerful new resource for diagnosing and understanding the role of terrestrial ecosystems and climate drivers in the global CH4 cycle, and future additions of sites in tropical ecosystems and site years of data collection will provide added value to this database. All seasonality parameters are available at https://doi.org/10.5281/zenodo.4672601 (Delwiche et al., 2021). Additionally, raw FLUXNET-CH4 data used to extract seasonality parameters can be downloaded from https://fluxnet.org/data/fluxnet-ch4-community-product/ (last access: 7 April 2021), and a complete list of the 79 individual site data DOIs is provided in Table 2 of this paper.

• We evaluate methane flux gap-filling methods across 17 boreal-to-tropical wetlands • New methods for generating realistic artificial gaps and uncertainties are proposed • Decision tree algorithms perform slightly better than neural networks on average • Soil temperature and generic seasonality are the most important predictors • Open-source code is released for gap-filling steps and uncertainty evaluation Time series of wetland methane fluxes measured by eddy covariance require gap-filling to estimate daily, seasonal, and annual emissions. Gap-filling methane fluxes is challenging because of high variability and complex responses to multiple drivers. To date, there is no widely established gap-filling standard for wetland methane fluxes, with regards both to the best model algorithms and predictors. This study synthesizes results of different gap-filling methods systematically applied at 17 wetland sites spanning boreal to tropical regions and including all major wetland classes and two rice paddies. Procedures are proposed for: 1) creating realistic artificial gap scenarios, 2) training and evaluating gap-filling models without overstating performance, and 3) predicting half-hourly methane fluxes and annual emissions with realistic uncertainty estimates. Performance is compared between a conventional method (marginal distribution sampling) and four machine learning algorithms. The conventional method achieved similar median performance as the machine learning models but was worse than the best machine learning models and relatively insensitive to predictor choices. Of the machine learning models, decision tree algorithms performed the best in cross-validation experiments, even with a baseline predictor set, and artificial neural networks showed comparable performance when using all predictors. Soil temperature was frequently the most important predictor whilst water table depth was important at sites with substantial water table fluctuations, highlighting the value of data on wetland soil conditions. Raw gap-filling uncertainties from the machine learning models were underestimated and we propose a method to calibrate uncertainties to observations. The python code for model development, evaluation, and uncertainty estimation is publicly available. This study outlines a modular and robust machine learning workflow and makes recommendations for, and evaluates an improved baseline of, methane gap-filling models that can be implemented in multi-site syntheses or standardized products from regional and global flux networks (e.g., FLUXNET).

• We evaluate methane flux gap-filling methods across 17 boreal-to-tropical wetlands • New methods for generating realistic artificial gaps and uncertainties are proposed • Decision tree algorithms perform slightly better than neural networks on average • Soil temperature and generic seasonality are the most important predictors • Open-source code is released for gap-filling steps and uncertainty evaluation Time series of wetland methane fluxes measured by eddy covariance require gap-filling to estimate daily, seasonal, and annual emissions. Gap-filling methane fluxes is challenging because of high variability and complex responses to multiple drivers. To date, there is no widely established gap-filling standard for wetland methane fluxes, with regards both to the best model algorithms and predictors. This study synthesizes results of different gap-filling methods systematically applied at 17 wetland sites spanning boreal to tropical regions and including all major wetland classes and two rice paddies. Procedures are proposed for: 1) creating realistic artificial gap scenarios, 2) training and evaluating gap-filling models without overstating performance, and 3) predicting half-hourly methane fluxes and annual emissions with realistic uncertainty estimates. Performance is compared between a conventional method (marginal distribution sampling) and four machine learning algorithms. The conventional method achieved similar median performance as the machine learning models but was worse than the best machine learning models and relatively insensitive to predictor choices. Of the machine learning models, decision tree algorithms performed the best in cross-validation experiments, even with a baseline predictor set, and artificial neural networks showed comparable performance when using all predictors. Soil temperature was frequently the most important predictor whilst water table depth was important at sites with substantial water table fluctuations, highlighting the value of data on wetland soil conditions. Raw gap-filling uncertainties from the machine learning models were underestimated and we propose a method to calibrate uncertainties to observations. The python code for model development, evaluation, and uncertainty estimation is publicly available. This study outlines a modular and robust machine learning workflow and makes recommendations for, and evaluates an improved baseline of, methane gap-filling models that can be implemented in multi-site syntheses or standardized products from regional and global flux networks (e.g., FLUXNET).

• Large-scale eddy-covariance flux datasets need to be used with footprint-awareness • Using a fixed-extent target area across sites can bias model-data integration • Most sites do not represent the dominant land-cover type at a larger spatial extent • A representativeness index provides general guidance for site selection and data use Large datasets of greenhouse gas and energy surface-atmosphere fluxes measured with the eddy-covariance technique (e.g., FLUXNET2015, AmeriFlux BASE) are widely used to benchmark models and remote-sensing products. This study addresses one of the major challenges facing model-data integration: To what spatial extent do flux measurements taken at individual eddy-covariance sites reflect model- or satellite-based grid cells? We evaluate flux footprints—the temporally dynamic source areas that contribute to measured fluxes—and the representativeness of these footprints for target areas (e.g., within 250–3000 m radii around flux towers) that are often used in flux-data synthesis and modeling studies. We examine the land-cover composition and vegetation characteristics, represented here by the Enhanced Vegetation Index (EVI), in the flux footprints and target areas across 214 AmeriFlux sites, and evaluate potential biases as a consequence of the footprint-to-target-area mismatch. Monthly 80% footprint climatologies vary across sites and through time ranging four orders of magnitude from 10 3 to 10 7 m 2 due to the measurement heights, underlying vegetation- and ground-surface characteristics, wind directions, and turbulent state of the atmosphere. Few eddy-covariance sites are located in a truly homogeneous landscape. Thus, the common model-data integration approaches that use a fixed-extent target area across sites introduce biases on the order of 4%–20% for EVI and 6%–20% for the dominant land cover percentage. These biases are site-specific functions of measurement heights, target area extents, and land-surface characteristics. We advocate that flux datasets need to be used with footprint awareness, especially in research and applications that benchmark against models and data products with explicit spatial information. We propose a simple representativeness index based on our evaluations that can be used as a guide to identify site-periods suitable for specific applications and to provide general guidance for data use.

• Large-scale eddy-covariance flux datasets need to be used with footprint-awareness • Using a fixed-extent target area across sites can bias model-data integration • Most sites do not represent the dominant land-cover type at a larger spatial extent • A representativeness index provides general guidance for site selection and data use Large datasets of greenhouse gas and energy surface-atmosphere fluxes measured with the eddy-covariance technique (e.g., FLUXNET2015, AmeriFlux BASE) are widely used to benchmark models and remote-sensing products. This study addresses one of the major challenges facing model-data integration: To what spatial extent do flux measurements taken at individual eddy-covariance sites reflect model- or satellite-based grid cells? We evaluate flux footprints—the temporally dynamic source areas that contribute to measured fluxes—and the representativeness of these footprints for target areas (e.g., within 250–3000 m radii around flux towers) that are often used in flux-data synthesis and modeling studies. We examine the land-cover composition and vegetation characteristics, represented here by the Enhanced Vegetation Index (EVI), in the flux footprints and target areas across 214 AmeriFlux sites, and evaluate potential biases as a consequence of the footprint-to-target-area mismatch. Monthly 80% footprint climatologies vary across sites and through time ranging four orders of magnitude from 10 3 to 10 7 m 2 due to the measurement heights, underlying vegetation- and ground-surface characteristics, wind directions, and turbulent state of the atmosphere. Few eddy-covariance sites are located in a truly homogeneous landscape. Thus, the common model-data integration approaches that use a fixed-extent target area across sites introduce biases on the order of 4%–20% for EVI and 6%–20% for the dominant land cover percentage. These biases are site-specific functions of measurement heights, target area extents, and land-surface characteristics. We advocate that flux datasets need to be used with footprint awareness, especially in research and applications that benchmark against models and data products with explicit spatial information. We propose a simple representativeness index based on our evaluations that can be used as a guide to identify site-periods suitable for specific applications and to provide general guidance for data use.