FLUXNET-CH₄: a global, multi-ecosystem dataset and analysis of methane seasonality from freshwater wetlands

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.

1 Introduction

Methane (CH₄) has a global warming potential that is 28 times larger than carbon dioxide (CO₂) on a 100-year timescale (Myhre et al., 2013), and its atmospheric concentration has increased by > 1000 ppb since 1800 (Etheridge et al., 1998). While atmospheric CH₄ concentrations are substantially lower than those of CO₂, CH₄ has contributed 20 %–25 % as much radiative forcing as CO₂ since 1750 (Etminan et al., 2016). Despite its importance to global climate change, natural CH₄ sources and sinks remain poorly constrained and with uncertain attribution to the various biogenic and anthropogenic sources (Saunois et al., 2016, 2020). Bottom-up and top-down estimates differ by 154 Tg yr⁻¹ (745 versus 591 Tg yr⁻¹, respectively); much of this difference arises from natural sources (Saunois et al., 2020). Vegetated wetlands and inland water bodies account for most natural CH₄ emissions, as well as the majority of uncertainty in bottom-up emission estimates (Saunois et al., 2016). Better diagnosis and prediction of terrestrial CH₄ sources to the atmosphere requires high frequency and continuous measurements of CH₄ exchange across a continuum of time (hours to years) and space (meters to kilometers) scales.

Tower-based eddy covariance (EC) measurements providing ecosystem-scale CH₄ fluxes at high temporal resolution across years are coupled with measurements of key CH₄ drivers such as temperature, water, and recent substrate input (inferred from CO₂ flux) and thus help constrain bottom-up CH₄ budgets and improve CH₄ predictions. Although EC towers began measuring CO₂ fluxes in the late 1970s (Desjardins, 1974; Anderson et al., 1984), and some towers began measuring CH₄ in the 1990s (Verma et al., 1992), most CH₄ flux EC measurements began within the last decade (2010s). Given that many EC CH₄ sites are relatively new, the flux community has only recently compiled them for global synthesis efforts (e.g., Chang et al., 2021) and is still working to standardize CH₄ flux measurements and establish gap-filling protocols (Nemitz et al., 2018; Knox et al., 2019). Furthermore, the growth of EC networks for CH₄ fluxes has sometimes taken place in a relatively ad hoc fashion, often at sites that were already measuring CO₂ fluxes or where higher CH₄ fluxes were expected, potentially introducing bias. The representativeness and spatial distribution of CO₂ flux tower networks have been assessed to evaluate their ability to upscale fluxes regionally (Hargrove et al., 2003; Hoffman et al., 2013; Papale et al., 2015; Villarreal et al., 2018, 2019) and globally (Jung et al., 2009, 2020). However, a relatively sparse coverage of CH₄ flux towers prompts the question of how well the current observation network provides a sufficient sampling of global or ecosystem-specific bioclimatic conditions.

Broad-scale wetland CH₄ seasonality estimates, such as when fluxes increase, peak, and decrease and the predictors of seasonality, remain relatively unconstrained across wetlands globally. These key seasonality metrics vary considerably across high-emitting systems such as wetlands and other aquatic systems (Desjardins, 1974; Dise, 1992; Melloh and Crill, 1996; Wik et al., 2013; Zona et al., 2016; Treat et al., 2018). The few continuous CH₄ flux datasets across representative site years make it difficult to establish trends in seasonal dynamics, though monthly or annually aggregated estimates of CH₄ fluxes from different seasons do exist for high latitudes (Zona et al., 2016; Treat et al., 2018). Seasonal variability in freshwater wetland CH₄ fluxes is expected to be driven by changes in air (TA) and soil temperature (ST), soil moisture (including water table dynamics), and recent carbon substrate availability, which influence the rates of CH₄ production and consumption (Lai, 2009; Bridgham et al., 2013; Dean et al., 2018). Temperature has widely been found to strongly affect CH₄ flux (Chu et al., 2014; Yvon-Durocher et al., 2014; Sturtevant et al., 2016), but the relationship is complex (Chang et al., 2020) and varies seasonally (Koebsch et al., 2015; Helbig et al., 2017). CH₄ flux is also driven by inundation depth since anoxic conditions are typically necessary for methanogenesis (Lai, 2009; Bridgham et al., 2013), though CH₄ production under bulk-oxic conditions has been observed (Angle et al., 2017). Substrate availability influences CH₄ production potential and is linked with gross primary productivity (GPP) because recent photosynthate fuels methanogenesis, though this relationship can vary by ecosystem type, plant functional type, and biome (Megonigal et al., 1999; Chanton et al., 2008; Hatala et al., 2012; Lai et al., 2014; Malhotra and Roulet, 2015; Sturtevant et al., 2016). In process models, the seasonality of CH₄ emissions from wetlands globally is primarily constrained by inundation (Poulter et al., 2017) with secondary within-wetland influences from temperature and availability of carbon (C) substrates (Melton et al., 2013; Castro-Morales et al., 2018). Bottom-up and top-down global CH₄ estimates continue to disagree on total CH₄ flux magnitudes and seasonality, including the timing of annual peak emissions (Spahni et al., 2011; Saunois et al., 2020). Thus, the variability in and predictors of wetland CH₄ seasonality globally remain a knowledge gap that high-frequency and long-term EC data can help fill.

Here, we first describe version 1.0 of the FLUXNET-CH₄ dataset (available at https://fluxnet.org/data/fluxnet-ch4-community-product/, last access: 7 April 2021). Version 1.0 of the dataset expands and formalizes the publication of data scattered among regional flux networks as described previously in Knox et al. (2019). 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 upland ecosystems. FLUXNET-CH₄ includes an additional two wetland sites (RU-Vrk and SE-St1), but they are not available under the CC BY 4.0 data policy and thus are excluded from this analysis. Since the majority of sites in FLUXNET-CH₄ Version 1.0 (hereafter referred to solely as “FLUXNET-CH₄”) are freshwater wetlands, which are a substantial source of total atmospheric CH₄ emissions, we use the subset of data from freshwater wetlands to evaluate the representativeness of freshwater wetland coverage in the FLUXNET-CH₄ dataset relative to wetlands globally, and we provide the first assessment of global variability in and predictors of freshwater wetland CH₄ flux seasonality. We quantify a suite of CH₄ seasonality metrics and evaluate temperature and GPP (a proxy for recent substrate input) as predictors of seasonality across four latitudinal bands (northern, temperate, subtropical, and tropical). Due to a lack of high-temporal-resolution water table data at all sites, our analyses are unable to evaluate the critical role of water table on CH₄ seasonality. Here we provide parameters for better understanding and modeling seasonal variability in freshwater wetland CH₄ fluxes and generate new hypotheses and data resources for future syntheses.

2 Methods

2.1 FLUXNET-CH₄ dataset

2.1.1 History and data description

The FLUXNET-CH₄ dataset was initiated by the Global Carbon Project (GCP) in 2017 to better constrain the global CH₄ budget (https://www.globalcarbonproject.org/methanebudget/index.htm, last access: 6 July 2021). Beginning with a kick-off meeting in May 2018 in Washington DC, hosted by Stanford University, we coordinated with the AmeriFlux Management Project, the European Ecosystem Fluxes Database, and the Integrated Carbon Observation System Ecosystem Thematic Centre (ICOS-ETC) to avoid duplication of efforts as most sites are part of different regional networks (albeit with different data products). We collected and standardized data for FLUXNET-CH₄ with assistance from the regional flux networks, AmeriFlux's “Year of Methane”, FLUXNET, the EU's Readiness of ICOS for Necessities of Integrated Global Observations (RINGO) project, and a US Geological Survey Powell Center working group. FLUXNET-CH₄ is a community-led project, so while we developed it with assistance from FLUXNET, we do not necessarily use standard FLUXNET data variables, formats, or methods.

FLUXNET-CH₄ includes gap-filled half-hourly CH₄ fluxes and meteorological variables. Gaps in meteorological variables (TA – air temperature; SW_IN – incoming shortwave radiation; LW_IN – incoming longwave radiation; VPD – vapor pressure deficient; PA – pressure; P – precipitation; WS – wind speed) were filled with the ERA-Interim (ERA-I) reanalysis product (Vuichard and Papale, 2015). We used the REddyProc package (Wutzler et al., 2018) to filter flux values with low friction velocity (u_∗) based on relating nighttime u_∗ to fill gaps in CO₂, latent heat, and sensible heat fluxes and to partition net CO₂ fluxes into gross primary production (GPP) and ecosystem respiration (RECO) using both the daytime (Lasslop et al., 2010) and nighttime (Reichstein et al., 2005) approaches. Data gaps of CH₄ flux were filled using artificial neural network (ANN) methods first described in Knox et al. (2015, 2019) and summarized here in Sect. 2.1.2. Gap-filled data for gaps exceeding 2 months are provided and flagged for quality. Please see Table B1 for variable description and units, as well as quality flag information. For the seasonality analysis in this paper we excluded data from gaps exceeding 2 months, and we encourage future users of FLUXNET-CH₄ to critically evaluate gap-filled values from long data gaps before including them in analyses (Dengel et al., 2013; Kim et al., 2020).

In addition to half-hourly data, the FLUXNET-CH₄ Version 1.0 release also contains a full set of daily mean values for all parameters except wind direction and precipitation. Daily precipitation is included as the daily sum of the half-hourly data, and daily average wind direction is not included.

2.1.2 Gap-filling methods and uncertainty estimates

As described in Knox et al. (2015, 2019), the ANN routine used to gap-fill the CH₄ data was optimized for generalizability and representativeness. To avoid biasing the ANN toward environmental conditions with typically better data coverage (e.g., summertime and daytime measurements), the explanatory data were divided into a maximum of 15 clusters using a k-means clustering algorithm. Data used to train, test, and validate the ANN were proportionally sampled from these clusters. For generalizability, the simplest ANN architecture with good performance (< 5 % gain in model accuracy for additional increases in architecture complexity) was selected for 20 extractions of the training, test, and validation data. Within each extraction, each tested ANN architecture was reinitialized 10 times, and the initialization with the lowest root-mean-square error was selected to avoid local minima. The median of the 20 predictions was used to fill each gap. A standard set of variables available across all sites was used to gap-fill CH₄ fluxes (Dengel et al., 2013), which included the previously mentioned meteorological variables TA, SW_IN, WS, and PA and sine and cosine functions to represent seasonality. These meteorological variables were selected for their relevance to CH₄ exchange and were gap-filled using the ERA-I reanalysis data. Other variables related to CH₄ flux (e.g., water table depth, WTD, and soil temperature, TS) were not included as explanatory variables as they were not available across all sites or had large gaps that could not be filled using the ERA-I reanalysis data (Knox et al., 2019). The ANN gap-filling was performed using MATLAB (Mathworks 2018a, version 9.4.0).

While the median of the 20 predictions was used to fill each gap, the spread of the predictions was used to provide a measure of uncertainty resulting from the ANN gap-filling procedure. Specifically, the combined annual gap-filling and random uncertainty was calculated from the variance of the cumulative sums of the 20 ANN predictions (Knox et al., 2015; Anderson et al., 2016; Oikawa et al., 2017). The (non-cumulative) variance of the 20 ANN predictions was also used to provide gap-filling uncertainty for each half-hourly gap-filled value. While this output is useful for data–model comparisons, it cannot be used to estimate cumulative annual gap-filling error because gap-filling error is not random, which is why the cumulative sums of the 20 ANN predictions are used to estimate annual gap-filling error.

Random errors in EC fluxes follow a double exponential (Laplace) distribution with the standard deviation varying with flux magnitude (Richardson et al., 2006, 2012). For half-hourly CH₄ flux measurements, random error was estimated using the residuals of the median ANN predictions, providing a conservative “upper limit” estimate of the random flux uncertainty (Moffat et al., 2007; Richardson et al., 2008). The annual cumulative uncertainty at 95 % confidence was estimated by adding the cumulative gap-filling and random measurement uncertainties in quadrature (Richardson and Hollinger, 2007; Anderson et al., 2016). Annual uncertainties in CH₄ flux for individual site years are provided in Table B2. Throughout this paper, we include uncertainties on individual site years when discussing single years of data. In sites with multiple years of data, we report the standard deviation of the multiple years.

2.1.3 Dataset structure and site metadata

FLUXNET-CH₄ contains two comma-separated data files per site at half-hourly and daily resolutions which are available for download at https://fluxnet.org/data/fluxnet-ch4-community-product/ (last access: 7 April 2021), along with a file containing select site metadata. Each site has a unique FLUXNET-CH₄ DOI. All data from the 79 sites used in this analysis are available under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/, last access: 6 July 2021) copyright license (FLUXNET-CH₄ has an additional two sites available under the FLUXNET Tier 2 license (https://fluxnet.org/data/data-policy/, last access: 6 July 2021), though these sites are not included in our analysis).

Metadata (Table B3) include site coordinates, ecosystem classification based on site literature, presence/absence and dominance for specific vegetation types, and DOI link, as well as calculated data such as annual and quarterly CH₄ flux values. FLUXNET-CH₄ Version 1.0 sites were classified based on site-specific literature as fen, bog, swamp, marsh, salt marsh, lake, mangrove, rice paddy field, wet tundra, upland, or drained ecosystems that previously could have been wetlands, seasonally flooded pastures, or agricultural areas. To the extent possible, we followed classification systems of previous wetland CH₄ syntheses (Olefeldt et al., 2013; Turetsky et al., 2014; Treat et al., 2018). Drained systems are former wetlands that have subsequently been drained but may maintain a relatively shallow water table, which can contribute to occasional methane emissions, although we do not have specific water table depth information at all drained sites. Upland ecosystems are further divided into alpine meadows, grasslands, needleleaf forests, mixed forest, crops, tundra, and urban. Freshwater wetland classifications follow hydrological definitions of bog (ombrotrophic), fen (minerotrophic), wet tundra, marshes, and swamps and were designated as per the primary literature on the site. For all sites, vegetation was classified for the presence or absence of brown mosses (all species from the division Bryophyta except those in the class Sphagnopsida), Sphagnum mosses (any species from class Sphagnopsida), ericaceous shrubs, trees (of any height), and aerenchymatous species (mostly order Poales but includes exceptions). These categories closely follow Treat et al. (2018) except that aerenchymatous species had to be expanded beyond Cyperaceae to incorporate wetlands globally. Presence/absence of vegetation groups was designated based on species lists in primary literature from the site. Out of the vegetation groups present, the dominant (most abundant) group is also reported and is based on information provided by lead site investigators.

In addition to the variable description table (Table B1) and the site metadata (Table B3), we provide several more tables to complement our analysis. Table B4 includes the climatic data used in the representativeness analysis. Table B5 provides seasonality parameters for CH₄ flux, air temperature, soil temperature (from the probe closest to the ground surface), and GPP. For sites with multiple soil temperature probes, the full set of soil temperature parameters are in Table B6. Table B7 contains the soil temperature probe depths. Table B2 contains the annual CH₄ flux and uncertainty. All Appendix B tables are also available at https://doi.org/10.5281/zenodo.4672601 (Delwiche et al., 2021).

2.1.4 Annual CH₄ fluxes

Annual CH₄ fluxes were calculated from gap-filled data for site years with data gaps shorter than 2 consecutive months or for sites above 20^∘ N where > 2-month data gaps occurred outside of the highest CH₄-emission months of 1 May through 31 October. Since we did not sum gap-filled values for > 2 month-gaps during the winter, annual sums from these years will be an underestimate since winter fluxes can be important (Zona et al., 2016; Treat et al., 2018). Several sites had less than 1 year of data, and we report gap-filled CH₄ flux annual sums for sites with between 6 months and 1 year of data (BW-Gum = 228 d, CH-Oe2 = 200 d, JP-Swl = 210 d, and US-EDN = 182 d). While these sums will be an underestimate of annual CH₄ flux since they do not span a full year (and we therefore do not use them in the seasonality analysis), their relative magnitude can still be informative. For example, site JP-SWL is a lake site, and even with less than 1 year of data the summed CH₄ flux of 66 g C m⁻² is relatively high (Taoka et al., 2020). In addition to sites with short time series, the annual CH₄ sum for site ID-Pag represents 365 d spanning June 2016 to June 2017.

2.1.5 Subset analysis on freshwater wetland CH₄ flux

In addition to the FLUXNET-CH₄-wide description of site class distributions and annual CH₄ fluxes, we also include a subset analysis on freshwater wetlands given that it is the dominant ecosystem type in our dataset and an important global CH₄ source (Saunois et al., 2016). First, we analyze freshwater wetland representativeness and subsequently the seasonality of their CH₄ emissions. Freshwater wetlands included in the seasonality and representativeness analysis are indicated in Table B3, column “IN_SEASONALITY_ANALYSIS”.

2.2 Wetland representativeness

2.2.1 Principal component analysis

To compare the FLUXNET-CH₄ site distribution to the global wetland distribution, we evaluated their representativeness in the entire global wetland cover along four bioclimatic gradients. Only freshwater wetland sites were included in this analysis. Coastal sites were excluded because salinity, an important control on CH₄ production, could not be evaluated across the tower network due to a lack of global gridded salinity data (Bartlett et al., 1987; Poffenbarger et al., 2011). The four bioclimatic variables used were mean annual air temperature (MAT), latent heat flux (LE), enhanced vegetation index (EVI), and simple ratio water index (SRWI; data sources in Table B4). We use EVI because it is a more direct measurement than GPP from global gridded products and is considered a reasonable proxy for GPP (Sims et al., 2006). Together, these environmental variables account for, or are, proxies for key controls of CH₄ production, oxidation at the surface, and transport (Bridgham et al., 2013). We use a principal components analysis (PCA) to visualize the site distribution across the four environmental drivers at once. For this analysis, we consider the annual average bioclimatic conditions over 2003–2015. In the PCA output, we evaluate the coverage of the 42 freshwater sites over 0.25^∘ grid cells containing > 5 % wetland mean cover in Wetland Area and Dynamics for Methane Modeling (WAD2M; Zhang et al., 2020, 2021) for the same time period.

2.2.2 Global dissimilarity and constituency analysis

To further identify geographical gaps in the coverage of the FLUXNET-CH₄ Version 1.0 network, we quantified the dissimilarity of global wetlands from the tower network, using a similar approach to that taken for CO₂ flux towers (Meyer and Pebesma, 2020). We calculated the 4-dimensional Euclidean distance from the four bioclimatic variables between every point at the land surface to every tower location at the FLUXNET-CH₄ network. We then divided these distances by the average distance between towers to produce a dissimilarity index. Dissimilarity scores < 1 represent areas whose nearest tower is closer than the average distance among towers, while areas with scores > 1 are more distant. Lastly, we identified the importance of an individual tower in the network by estimating the geographical area to which it is most analogous in bioclimate space. We divided the world's land surface according to closest towers in bioclimatic space. The area to which each tower is nearest is defined as the tower's constituency.

2.3 Wetland CH₄ seasonality

To examine freshwater wetland CH₄ seasonality across the global range of sites in FLUXNET-CH₄, we extracted seasonality parameters for CH₄, temperature, and GPP using TIMESAT, a software package designed to analyze seasonality of environmental systems (Jönsson and Eklundh, 2002, 2004; Eklundh and Jönsson, 2015). TIMESAT calculates several seasonality parameters, including baseline flux, peak flux, and the slope of spring flux increase and fall decrease (Fig. 1). We also calculate parameters such as amplitude (peak flux minus baseline, which is the average of spring and fall baselines; “e” − ((“a” + “b”)$/\mathrm{2}$) in Fig. 1) and relative peak timing ((“g” − “f”)$/$(“h” − “f”) in Fig. 1). TIMESAT uses a double-logistic fitting function to create a series of localized fits centered on data minima and maxima. Localized fits are determined by minimizing a merit function with the Levenberg–Marquardt method (Madsen et al., 2004; Nielsen, 1999). These localized fits are then merged using a global function to create a smooth fit over the full time interval. To fit CH₄ time series in TIMESAT, we used gap-filled data after removing gaps exceeding 2 months. We do not report TIMESAT parameters when large gaps occur during the spring CH₄ emissions' increase, peak, or fall decrease.

Figure 1 TIMESAT parameter description. (a, b) Base values (TIMESAT reports the average of these two values), (c, d) slopes of seasonal curves (lines drawn between 20 % and 80 % of the amplitude), (e) peak value, and day of year (DOY) for the start (f), peak (g), and end (h) of the elevated methane (CH₄) emission season. Data points are the mean daily gap-filled CH₄ fluxes from site JP-Bby in 2015.

We estimate “start of elevated emission season” when CH₄ emissions begin to increase in the spring ( “f” in Fig. 1), and “end of elevated emission season” when the period of elevated CH₄ flux ends in the fall (“h” in Fig. 1), as the intercept between the TIMESAT fitted baseline parameter and shoulder-season slope (similar to Gu et al., 2009). To extract seasonality parameters with TIMESAT, sites need a sufficiently pronounced seasonality, a sufficiently long time period, and minimal data gaps (we note that while TIMESAT is capable of fitting two peaks per year, all the freshwater wetland sites have a single annual peak). We excluded site years in restored wetlands when wetlands were still under construction. Of the 42 freshwater wetland sites in FLUXNET-CH₄ Version 1.0, 36 had sufficient data series to extract seasonality parameters. These 36 wetlands had 141 site years of data in total, which we fit with the double-logistic fitting method which followed site data well (representative examples in Fig. 2). For extratropical sites in the Southern Hemisphere, we shifted all data by 182 d so that maximum solar insolation seasonality would be congruent across the globe.

Figure 2 Examples of TIMESAT fits for two FLUXNET-CH₄ sites, (a) RU-Che and (b) JP-Bby. Methane (CH₄) flux data showing daily average flux tower data, with several high outliers excluded to improve the plot (dark gray), gap-filled values (light gray), and TIMESAT-fitted curve (dark red line) for sites JP-Bby and RU-Che. TIMESAT captures the size and shape of peaks (note different scale on y axes). CH₄= methane.

We also used TIMESAT to extract seasonality metrics for GPP, partitioned using the daytime-based approach (Lasslop et al., 2010) (GPP_DT), air temperature (TA), and soil temperature (TS_1, TS_2, etc). For sites where winter soil temperatures fall significantly below 0 ^∘C, TIMESAT fits a soil temperature “start of elevated season” date to periods when the soil is still frozen. In order for TIMESAT to define the soil temperature seasonality within the thawed season, we converted all negative soil temperatures to zero (simply removing these values results in too many missing values for TIMESAT to fit). Many sites have more than one soil temperature probe, so we extracted separate seasonality metrics from each individual probe (although we used the metrics from the shallowest temperature probe in our analysis). Table B4 contain the TIMESAT seasonality parameters used in the seasonality analysis. We did not include water table depth in the seasonality analysis because many sites either lack water table depth measurements or have sparse data.

We regressed the CH₄ seasonality parameters from TIMESAT against annual temperature, annual water table depth, and TIMESAT seasonality parameters for air temperature, soil temperature, and GPP (proxy for recent carbon input available as substrate) using linear mixed-effect modeling with the lmer command (with site as a random effect) from the R (R Core Team, 2018, version 3.6.2) package lmerTest (Kuznetsova et al., 2017). For these regressions we present the marginal R² outputs from lmer, which represent the variance explained only by the fixed effects. Mixed-effect modeling was necessary to account for the non-independence between measurements taken at the same site during different years (Zona et al., 2016; Treat et al., 2018). We also compared how seasonality metrics varied across latitudinal bands by dividing sites into northern (> 60^∘ N), temperate (between 40 and 60^∘ N), subtropical (absolute value between 20 and 40^∘ N latitude, with site NZ-KOP being the only Southern Hemisphere site), and tropical (absolute value below 20^∘ N). Site-year totals for the northern, temperate, subtropical, and tropical bands were n=57, 36, 39, and 9, respectively. We used the Kruskal–Wallis test to establish whether groups (either across quarters or across latitudes) were from similar distributions and the post hoc multiple comparison “Dwass, Steel, Critchlow, and Fligner” procedure for inter-group comparisons. Kruskal–Wallis and post hoc tests were implemented in Python Version 3.7.4, using stats from scipy for Kruskal–Wallis and posthoc_dscf from scikit_posthocs.

We also compared quarterly CH₄ flux sums by dividing data into quarterly periods: January–February–March (JFM), April–May–June (AMJ), July–August–September (JAS), and October–November–December (OND). For the sake of simplicity, we chose to compare quarterly periods rather than site-specific growing/non-growing season periods so that all time periods would be the same length. Quarterly sums were computed from the gap-filled CH₄ fluxes when the longest continuous data gap within the quarter did not exceed 30 d, leading to site-year counts of 67, 92, 95, and 72 for JFM, AMJ, JAS, and OND, respectively. We compared quarterly CH₄ fluxes across latitudinal bands both for the total CH₄ flux and for the quarterly percentage of the annual CH₄ flux. Quarterly statistics were also conducted with the Kruskal–Wallis test and the post hoc multiple comparison “Dwass, Steel, Critchlow, and Fligner” procedure implemented in Python. Quarterly values are provided in Table B3, and the sum of mean quarterly CH₄ flux does not always equal mean annual CH₄ flux because some quarters either do not have data or have data gaps that exceed 30 d.

3 Results and discussion

3.1 FLUXNET-CH₄ dataset

3.1.1 Dataset description

Version 1.0 of the FLUXNET-CH₄ dataset contains 79 unique sites, 293 total site years of data, and 201 site years with sufficient data to estimate annual CH₄ emissions. A synthesis paper, published prior to the public data release of FLUXNET-CH₄ Version 1.0, had 60 unique sites and 139 site years with annual CH₄ emission estimates (Knox et al., 2019). Freshwater wetlands make up the majority of sites (n=42), and the dataset also includes five salt marshes and one mangrove wetland. Notable additions to FLUXNET-CH₄ from the previous unpublished dataset used in Knox et al. (2019) include six tropical sites (between 20^∘ S and 20^∘ N), including one site in South America, two sites in southern Africa, and three sites in Southeast Asia. The 15 upland sites include six needleleaf forests, three crop sites (excluding rice), two alpine meadows, one grassland, one mixed forest, one tundra, and one urban site. The drained sites represent former wetlands that have been artificially drained for use as grasslands (n=3) or croplands (n=3). FLUXNET-CH₄ sites span the globe, though they are concentrated in North America and Europe (Fig. 3). Table B3 includes characteristics of all sites in the dataset.

Figure 3 Global map of FLUXNET-CH₄ Version 1.0 site locations colored by site type. Panels (a)–(d) show sites that were too closely located to distinguish in the global map.

Sites represent a range of ecosystem types, latitudes, median fluxes, and seasonality patterns (Table 1). Across all FLUXNET-CH₄ sites (including non-wetland sites), mean average annual CH₄ flux is positively skewed with a median flux of 9.5 g C m⁻² yr⁻¹, a mean flux of 16.9 g C m⁻² yr⁻¹, and numerous annual CH₄ fluxes exceeding 60 g C m⁻² yr⁻¹. Marshes and swamps have the highest median flux, and upland, salt marsh, and tundra sites have the lowest (Fig. 4). Lake emissions are highly variable due to one high-flux lake site (JP-SWL). Flux data at many sites show strong seasonality in CH₄ emissions, but data coverage is also lower outside the growing season (Table 1). Data coverage is lowest during the JFM quarter (on average 20 % of half-hourly time periods contain flux data), reflecting the predominance of Northern Hemisphere sites and the practical difficulties in maintaining EC tower sites during colder winter months (Table 1). Bogs, fens, and marshes have pronounced seasonality, with fluxes being highest in the AMJ and JAS quarters. In contrast, CH₄ fluxes from uplands, drained sites, and salt marshes are more uniform and low year-round.

Figure 4 Histogram of annual methane fluxes ($F_{{\mathrm{CH}}_{\mathrm{4}}}$, g C m⁻² yr⁻¹) grouped by site type.

Table 1 Summary table of sites grouped by ecosystem class reporting annual mean flux (Ann_Flux) and standard deviation from interannual variability (Ann_Flux_SD), site years of data, percent data cover per quarter, and median (med.) flux across site class. JFM signifies January–February–March, AMJ April–May–June, JAS July–August–September, and OND October–November–December.

3.1.2 Freshwater wetland CH₄ characteristics

The FLUXNET-CH₄ Version 1.0 dataset contains 42 freshwater wetlands that span 37^∘ S to 69^∘ N, including bogs, fens, wet tundra, marshes, and swamps, and a range of annual CH₄ emission rates (Fig. 4). The majority of freshwater wetlands in our dataset emit 0–20 g C m⁻² yr⁻¹, with 10 emitting 20–60 g C m⁻² yr⁻¹, and one more than 60 g C m⁻² yr⁻¹. Differences in annual CH₄ flux among wetland types is partially driven by temperature (which is often linked to site type), with mean annual air temperature explaining 51 % of the variance between sites (Fig. 5, exponential relationship). The global relationship between annual methane emissions and temperature can be described using a Q₁₀ relationship where $Q_{\mathrm{10}}=R\mathrm{2}/R{\mathrm{1}}^{\left(\right(T\mathrm{2}-T\mathrm{1})/\mathrm{10})}$, with R2 and R1 being the CH₄ emission rates at temperatures T2 and T1, respectively (temperature in ^∘C). The Q₁₀ based on Fig. 5 data is 2.57. We also note that annual CH₄ flux from individual biomes may have different relationships with temperature, as previous work has shown biome-specific trends in CH₄ flux with environmental drivers (Abdalla et al., 2016). However, there currently are not enough data points in each biome category to compare relationships between mean annual CH₄ flux and temperature. Annual CH₄ flux is not correlated with mean annual water table depth in FLUXNET-CH₄, unlike in Knox et al. (2019), which used a subset of the FLUXNET-CH₄ sites in which CH₄ flux was correlated with water table depth only for sites with water table below ground for 90 % of measured days (r²=0.31, p< 0.05, n=27 site years). Freshwater wetland seasonality is further described in Sect. 3.3.

Figure 5 Relationship between mean annual wetland methane (CH₄) flux (g C m⁻² yr⁻¹, logarithmic scale) and mean annual air temperature (^∘C) for each freshwater wetland site, with wetland type indicated by symbol. Markers represent individual site means, with vertical error bars representing the standard deviation of interannual variability.

3.1.3 Upland, rice, and urban CH₄ characteristics

Upland agricultural sites are characterized by a lack of seasonal pattern in CH₄ emissions, relatively low flux, and sometimes negative daily flux (i.e., CH₄ uptake) averages. All of the upland non-agricultural sites in FLUXNET-CH₄ Version 1.0 are net (albeit weak) CH₄ sources except for the needleleaf forest site US-Ho1, which has a mean annual CH₄ flux of $-\mathrm{0.1}\pm \mathrm{0.1}$ g C m⁻² yr⁻¹ (see Table B3 for site acronyms and metadata). The average agricultural site emissions are 1.3±0.8 g C m⁻² yr⁻¹, and non-agricultural site emissions are 1.6±1.2 g C m⁻² yr⁻¹ across sites.

Rice sites (n=7) have average annual emissions across all sites of 16.7±7.7 g C m⁻² yr⁻¹ and are characterized by strong seasonal patterns, with either one or more CH₄ emission peaks per year depending on the number of rice seasons and field water management. One peak is typically observed during the reproductive period for the continuously flooded sites with one rice season (i.e., US-HRC, JP-MSE) (Iwata et al., 2018; Runkle et al., 2019; Hwang et al., 2020). For sites with only one rice season but with single or multiple drainage and re-flooding periods, a secondary peak may appear before the reproductive peak (i.e., KR-CRK, IT-Cas, and US-HRA; Meijide et al., 2011; Runkle et al., 2019; Hwang et al., 2020). Two reproductive peaks appear for sites with two rice seasons (i.e., PH-RiF), and each reproductive peak may be accompanied by a secondary peak due to drainage events (Alberto et al., 2015). Even sites with one continuously flooded rice season may experience a second peak if the field is flooded during the fallow season to provide habitat for migrating birds (e.g., US-Twt; Knox et al., 2016).

The dataset has 1 year of urban data from site UK-LBT in London, England. UK-LBT observes CH₄ fluxes from a 190 m tall communications tower in the center of London and has a mean annual CH₄ flux of 46.5±5.6 g C m⁻² yr⁻¹. This flux is more than twice as high as the mean annual CH₄ flux across all FLUXNET-CH₄ sites, 16.9 g C m⁻² yr⁻¹. The London site has higher CH₄ emissions in the winter compared to summer, which is attributed to a seasonal increase in natural gas usage (Helfter et al., 2016.)

3.1.4 Saltwater and mangrove wetland CH₄ characteristics

Three of the five saltwater wetlands in FLUXNET-CH₄ (US-Edn, US-MRM, and US-Srr) have a very low mean annual CH₄ flux (see Table B2 for individual site-year CH₄ flux sums and associated uncertainty) and minimal seasonality. Two other FLUXNET-CH₄ saltwater sites (US-La1 and US-StJ) have significantly higher fluxes, with annual sums of 12.6±0.6 and 9.6±1.0 g C m⁻² yr⁻¹, respectively, while the mangrove site HK-MPM has annual mean fluxes of 11.1±0.5 g C m⁻² yr⁻¹. This range of CH₄ fluxes across different saltwater ecosystems could be valuable for exploring the effect of salinity and different biogeochemical pathways of CH₄ production, oxidation, and transport of CH₄ (Bartlett et al., 1987; Poffenbarger et al., 2011). Saltwater wetlands along the coast have unique CH₄ dynamics attributable to the presence of abundant electron acceptors, most importantly sulfates which inhibit methanogenesis (Pattnaik et al., 2000; Mishra et al., 2003; Weston et al., 2006) but at low concentrations can have no effect (Chambers et al., 2011) or even increase methanogenesis (Weston et al., 2011). In fact, estuarine wetlands with moderate salinity can still be significant sources of CH₄ (Liu et al., 2020). Even under sulfate-rich conditions, high CH₄ production can be found via methylotrophic methanogenesis (Dalcin Martins et al., 2017; Seyfferth et al., 2020,) or because the processes of sulfate reduction and methanogenesis are spatially separated (Koebsch et al., 2019). Consequently, representing the biophysical drivers of ecosystem-scale CH₄ fluxes in non-freshwater wetlands is challenging and may represent a combination of competing or confounding effects (Vazquez-Lule and Vargas, 2021).

3.2 Freshwater wetland representativeness

We evaluated the representativeness of freshwater wetland sites in the FLUXNET-CH₄ Version 1.0 dataset against wetlands globally, based on bioclimatic conditions of our sites. When evaluating bioclimatic variables individually, the distribution of freshwater wetlands across the network was significantly different from the global distribution (alpha > 0.05; two-tailed Kolmogorov–Smirnov tests; see Table B4). We exclude wetlands classified as “salt marsh” in this representativeness analysis and the seasonality analysis below because of the unique CH₄ flux dynamics in saltwater ecosystems (as discussed in Sect. 3.1.4), though we note that some of the coastal wetlands included in the freshwater analysis periodically experience brackish water (i.e., US-Myb, US-Sne).

When considering the four bioclimatic variables, MAT, LE, EVI, and SRWI in a PCA, we found that our tower network generally samples the bioclimatic conditions of global wetland cover, but some noticeable gaps remain (Fig. 6). Three clusters of the world's wetland-dense regions are identified but are not equally sampled by the network. A cluster of low-temperature wetlands is sampled by a large number of high-latitude sites. The other two wetland clusters are not as well sampled: a high-temperature and LE cluster is represented only by two towers (ID-Pag and MY-MLM), while drier and temperate and subtropical wetlands including large swathes of the Sahel in Africa only have a site in Botswana (BW-Npw) as their closest analog tower.

Figure 6 Principal component analysis displaying the distribution of freshwater wetland sites (points) along the two main principal components together accounting for 91.9 % of variance. Tower sites are represented as points with shapes indicating their wetland type and color shade representing the annual methane (CH₄) flux (gray points represent sites for which < 6 months of flux data were available to estimate annual budget). Sites codes are labeled in blue text for selected sites deviating from average conditions. Loading variables are represented by the arrows: mean annual temperature (MAT), simple ratio water index (SRWI), latent heat flux (LE), and enhanced vegetation index (EVI). The background shades of gray are a qualitative representation of the density of global wetland pixels and their distribution in the PCA climate space, with darker color representing higher densities (excluding Greenland and Antarctica). Only grid cells with > 5 % average wetland fraction according to the WAD2M over 2000–2018 are included (Zhang et al., 2020).

Evaluating the bioclimatic dissimilarity of global wetlands to the FLUXNET-CH₄ network shows the least captured regions are in the tropics (Fig. 7a). Sparse coverage in the tropics also means that the few existing towers occupy a critical place in the network, particularly as tropical wetlands are the largest CH₄ emitters (Bloom et al., 2017; Poulter et al., 2017). Highly dissimilar wetlands are limited in extent and distributed across all latitudes, but the average dissimilarity is higher in north temperate (55 to 65^∘) and tropical (−5 to 5^∘) latitudes (Fig. 7b). To evaluate the importance of individual towers in the network, we estimated the geographical area to which it is most analogous in bioclimate space (Fig. 7c). We found that some towers have disproportionately large constituencies (i.e., wetland areas that share the same closest bioclimatic analog tower). Towers in Indonesia (ID-Pag), Brazilian Pantanal (BR-Npw), and Botswana floodplains (BW-Nxr) represent the closest climate analog for much of the tropics (678 000, 300 000, and 284 000 km², respectively), while CA-SCB represents a vast swath (291 000 km²) of boreal and arctic regions (Fig. 7d).

Figure 7 (a) Distance in bioclimatic space between global land surface and the FLUXNET-CH₄ Version 1.0 tower network (gray areas indicate no mapped wetlands). The Euclidean distance was computed on the four bioclimatic variables and was then standardized by the average distance within the network. Most of the land surface has a dissimilarity score lower than 1, meaning these areas are closer than the average tower distance (lower dissimilarity score means a similar bioclimate to that represented by towers in the network). However, this pattern reflects more the sparsity of the tower network than a similarity of the land surface to the network. Areas with < 5 % coverage by wetlands were excluded to focus on wetland-dense regions. (b) Latitudinal distribution of dissimilarity score, (c) map of the four largest tower constituencies, and (d) scatterplot of wetland area in each tower constituency plotted against the average dissimilarity score (point) and ± standard deviation (error bar).

Our assessment of wetland CH₄ tower coverage determines the ability of our dataset to represent global wetland distributions and highlights some clear representation gaps in the network, particularly in tropical and humid regions. Other geographic regions such India, China, and Australia, where towers exist but are not included in the current network, should be prioritized when expanding the network even though they are not among the most distant areas to the current network. Similar representativeness assessments have been developed for CO₂ tower networks to identify gaps and priorities for expansion (Jung et al., 2009). To improve the geographic coverage of the network for representing global-scale fluxes, locations for new tower sites can be targeted to cover bio-climatically distant areas from the current network (Villarreal et al., 2019). Candidate regions for expansion that are both high CH₄ emitting (Saunois et al., 2020) and located in under-sampled climates are the following: African Sahel, Amazon Basin, Congo Basin, and Southeast Asia. Climatic conditions over boreal and arctic biomes are generally better represented (primarily at lower elevations), but there is scope to expand the network in wetland-dense regions like the Hudson Bay Lowlands and North Siberian Lowlands. Moreover, establishing sites in other ecosystem types, especially lakes and reservoirs (see Deemer et al., 2016; Bastviken et al., 2011; Matthews et al., 2020) in most climatic zones would help capture CH₄ fluxes from these ecosystems.

Understanding the representativeness of the network is essential when inferring general patterns of flux magnitude, seasonality, and drivers from the tower data (Villarreal et al., 2018). We produced a first-order representativeness of average bioclimatic conditions, but temporal representativeness (across seasons, climate anomalies, and extreme events) is particularly needed given the episodic nature of CH₄ fluxes (Chu et al., 2017; Mahecha et al., 2017; Göckede et al., 2019).

Assessing representation of wetland CH₄ sites is complicated by the fact that wetlands occupy only a fraction of most landscapes (except wetland-dense regions such as North Siberian Lowlands, Hudson Bay Lowlands, Congo Basin, etc.) and that not all relevant factors affecting CH₄ production and consumption could be considered in our analysis. For instance, our assessment of representation did not consider wetland types as such maps are limited by the inherent difficulties in remotely sensing wetland features (Gallant, 2015). The attribution of representativeness is further complicated by the fact that many EC tower locations are subject to small-scale variability within the field of view, or footprint, of the sensor. Consequently, the individual time steps within EC flux time series may represent a mixture of different wetland types or different fractions of wetland contribution to the total CH₄ flux, varying with wind direction, atmospheric stability, or season (Chu et al., 2021). This further complicates upscaling efforts. Additionally, this representativeness analysis did not apply weights to the drivers to reflect their varying influence on CH₄ flux. Such weights can be included in future versions as they are generated by a cross-validated machine learning approach (Jung et al., 2020). Future efforts could include the dissimilarity index from this analysis as a metric of extrapolation in a CH₄ flux upscaling effort.

3.3 Freshwater wetland flux seasonality

3.3.1 Seasonal flux comparisons by latitudinal bands

CH₄ flux and seasonality varied substantially across latitudinal bands (northern, temperate, subtropical, and tropical) (Fig. 8). Annual CH₄ fluxes for temperate, and subtropical sites were significantly higher than for northern sites (8.7 ± 5.0, 29.7 ± 25.2, 40.1 ± 14.6, and 24.5 ± 20.7 g C m⁻² yr⁻¹ for northern, temperate, subtropical, and tropical, respectively, and p < 0.0001 using Kruskal–Wallis and post hoc comparisons; Fig. 8a), and tropical sites were similar to all other latitudinal bands likely because of their small sample size. The ratio of seasonal amplitude to peak flux provides a measure of the relative seasonal increase in emissions compared with baseline, in which a ratio of 0 indicates no seasonal change in amplitude, a ratio of 1 indicates the off-season flux is zero, and values over 1 mean the off-season baseline CH₄ fluxes were negative (i.e., uptake). Average amplitude to peak flux ratios were similar across all latitudinal bands (0.9 ± 0.1, 0.9 ± 0.1, 0.9 ± 0.1, and 1.0 ± 0.7 for northern, temperate, subtropical, and tropical, respectively; Fig. 8b). The spring increase in CH₄ emissions began later in northern sites compared with temperate and subtropical sites (end of May versus April, respectively, and p=0.001; Fig. 8c), while tropical sites vary widely in elevated emission season start date. Northern sites also had shorter elevated CH₄ flux season lengths (138 ± 24 d) compared to temperate sites (162 ± 32 d), and both were shorter than subtropical sites (209 ± 43 d; p < 0.0001; Fig. 8e). On average, CH₄ flux peaked earlier for temperate sites compared to northern (p=0.008) and subtropical sites (p=0.02; middle to late July compared with early August; Fig. 8f), while tropical sites again vary widely. Given their unique seasonality and low number of site years (n=9), tropical systems are discussed separately in Sect. 3.3.3 and are not included in the comparisons in the remainder of this section. While our results on CH₄ seasonality corroborate expected trends for these latitudinal bands, they provide some of the first estimates of CH₄ seasonality parameters and ranges across a global distribution of sites.

Figure 8 (a) Annual methane (CH₄) flux (g C m⁻² yr⁻¹), (b) ratio of seasonal amplitude to seasonal peak, where values of 0 indicate uniform annual CH₄ flux, values of 1 indicate zero off-season fluxes, and values exceeding 1 indicate negative off-season fluxes, (c) CH₄ flux (FCH₄) elevated emission season start by day of year (DOY), (d) FCH4 elevated emission season end by DOY, (e) length of elevated CH₄ flux season (days), and (f) DOY of peak FCH₄. Northern (dark blue, solid line), temperate (blue, dashed line), sub-tropical (green, dot-dash line), and tropical (light green, solid line) wetlands plotted using the kernel density function. Each panel has lines that represent latitudinal bands as follows: northern (> 60^∘), temperate (between 40 and 60^∘), subtropical (between 20 and 40^∘), and tropical (< 20^∘), though the site-year totals vary between these groups (n=57, n=36, n=39, and n=9 respectively). All total CH₄ flux values and elevated season start values are positive, and the apparent continuation of the data distribution into negative values is an artifact of the kernel density function. Southern Hemisphere sites below 20^∘ S were shifted by 182 d to make summer the middle of the year for comparability with Northern Hemisphere sites.

We found that latitudinal groups showed strong differences in absolute CH₄ flux across quarters and narrower differences in percentage of annual CH₄ flux (Fig. 9a versus 9b). Thus, the AMJ quarter had a similar relative contribution to the annual CH₄ flux across latitudes regardless of the absolute annual CH₄ flux. CH₄ fluxes (Fig. 9a) were highest during JAS for northern, temperate, and subtropical sites and highest in AMJ and JAS for temperate sites (p < 0.01). Though CH₄ fluxes in northern sites are most commonly measured during warm summer months (Sachs et al., 2010; Parmentier et al., 2011), fluxes in JFM and OND (50 % of the yearly duration) on average make up 18.1 ± 3.6 %, 15.3 ± 0.1 %, and 31.2 ± 0.1 % (northern, temperate, and subtropical, respectively) of annual emissions. This pattern indicates that a substantial fraction of annual CH₄ fluxes occur during cooler months. The contribution of non-growing season CH₄ emissions to annual CH₄ fluxes has previously been described for arctic and boreal regions (Zona et al., 2016; Treat et al., 2018), and our analysis suggests comparable contributions in temperate and subtropical systems for the same quarterly periods.

Figure 9 (a) Quarterly contribution to total annual CH₄ flux (in g C m⁻²) and (b) percentage of annual CH₄ flux. Sites were divided into northern (> 60^∘ N), temperate (40–60^∘ N), and subtropical (20–40^∘ N). Quarters with continuous data gaps exceeding 30 d were excluded. We used the following quarterly periods: January–February–March (JFM), April–May–June (AMJ), July–August–September (JAS), and October–November–December (OND). Tropical sites are discussed separately in Sect. 3.3.3 because of their unique seasonality and low number of sites.

3.3.2 Predictors of CH₄ flux phenology

The start of the elevated CH₄ flux season, and how long the elevated flux season lasts, correlated strongly with mean annual air temperature (Fig. 10; p < 0.0001 for each). Methane flux began to increase roughly 2 months earlier in the warmest systems (mean annual temperature > 20 ^∘C) compared to the coldest (mean annual temperature near −10 ^∘C), though several of the warmer sites had high variability. Our data suggest that the CH₄ season started 2.8 ± 0.5 d earlier for every degree Celsius increase in mean annual temperature (Fig. 10a). In contrast, the end of the CH₄ emission season was not correlated with mean annual temperature, but a positive trend existed despite high variability at the warmest and coldest sites (Fig. 10b). The high variability seen in the end of the CH₄ season at northern sites is important to note and would likely be better resolved by incorporating other seasonality or phenological characteristics, such as moisture, active layer depth, and plant community composition (e.g., Kittler et al., 2017). Plants with aerenchymatous tissue, for example, influence the timing of plant-mediated CH₄ flux and are a key source of uncertainty when predicting CH₄ seasonality for northern wetlands (Xu et al., 2016; Kwon et al., 2017). Despite the relative lack of trend with season end date, the season length was still positively correlated with mean annual temperature, with the warmest sites having roughly 3 more months of seasonally elevated CH₄ emissions than the coldest sites (Fig. 10c). CH₄ season length increased 3.6 ± 0.6 d for every degree Celsius increase in mean annual temperature (note that these relationships are correlations, and we cannot disentangle causality with this analysis). Temperature is highly correlated with other parameters (i.e., radiation, days of snow cover, etc.), so CH₄ flux is also likely to correlate with other environmental parameters.

Figure 10 The (a) start of the elevated methane (CH₄) emission season (y=-2.8x+130, with “x” in ^∘C and “y” in day of year, DOY), (b) the end of the elevated emission season in DOY, and (c) the length of the emission season with mean annual site air temperature ($y=\mathrm{3.6}x+\mathrm{176.6}$, with “x” in ^∘C and “y” in days). Each point represents a site year of data, and all reported r² values are significant to p< 0.0001. Tropical sites are discussed separately in Sect. 3.3.3.

Although the spring onset of increasing CH₄ emissions correlated with mean annual air temperature, on average it lagged behind the spring increase in the shallowest soil temperatures by 31 ± 40 d (Fig. 11; lag is significantly different than zero, and p < 0.001), with very few instances of CH₄ emissions beginning before seasonal soil temperatures increase (and by 20 ± 50 d for the deepest temperature probes). In contrast, for roughly half of the sites, CH₄ emission increased prior to seasonal GPP (a proxy for fresh substrate availability) increases. This suggests that the initiation of increased CH₄ fluxes at the beginning of the season was not limited by availability of substrate derived from recent photosynthates. Additionally, the onset of CH₄ fluxes tended to occur closer to the onset of soil temperature increase for cooler temperature sites (sites with later start dates tend to be cooler; Fig. 11a). This result is likely attributable to the direct influence of increased temperature on microbial processes (Chadburn et al., 2020), as well as the indirect influences of snowmelt, both via release of CH₄ from the snowpack and a higher water table leading to more CH₄ production (Hargreaves et al., 2001; Tagesson et al., 2012; Mastepanov et al., 2013; Helbig et al., 2017). These observed trends hold for the entire temperature or GPP range of freshwater wetland sites but are not necessarily applicable within individual latitudinal bands.

Figure 11 Relationship between the onset of the methane (CH₄) emission season to (a) the beginning of the air warming by day of year (DOY), (b) soil warming at the shallowest probe depth per site by DOY, and (c) gross primary productivity (GPP) increase for the subset of sites with soil temperature data by DOY. Each point represents a site year of data. Dashed lines represent a 1:1 relationship, and solid lines are significant (p < 0.05) regression fits. On average, the CH₄ emission season lags behind the soil temperature increase by 31 ± 40 d and is more synchronous with GPP.

In contrast with the CH₄ season start timing, the timing of the CH₄ peak did not correlate with the timing of either the soil temperature peak or the GPP peak (Fig. A1). For 63 % of the sites, the average timing of peak CH₄ emissions lagged behind the soil temperature peak, and at 83 % of the sites average peak CH₄ lagged behind peak GPP (Fig. A1). Although there was no simple relationship between absolute CH₄ peak timing and the environmental drivers we investigated, there was a correlation (p=0.0005) between the relative timing of peak CH₄ compared to season onset (calculated as described in Sect. 2.3) and mean annual air temperature (Fig. 12a). For cooler sites, the peak of seasonal CH₄ emissions occurred closer to the onset of the CH₄ emission season than the end of the season, resulting in an asymmetrical seasonal CH₄ flux shape that is illustrated in Fig. 2a. Soil temperature also peaked earlier in the season for cooler wetlands, though the relationship is not as pronounced (p=0.009; Fig. 12b). In contrast, GPP peaked later in the season for cooler wetlands (p=0.009, Fig. 12c). Previous work on Arctic sites (sites US-Ivo, US-Beo, US-Atq, US-Bes, and RU-CH2) highlighted the asymmetrical annual CH₄ peak, with higher fall emissions being attributed to the “zero curtain” period when soil below the surface remains thawed for an extended period of time due to snow insulation (Zona et al., 2016; Kittler et al., 2017). Furthermore, soils can stay above the “zero curtain” range for an extended time into the fall and winter (Helbig et al., 2017), which may also be caused by snow insulation. The rapid onset of emissions in the spring following snowmelt could be attributed to the release of accumulated CH₄ (Friborg et al., 1997), and other high latitude sites have seen similarly sharp increases in CH₄ emissions at snowmelt (Dise, 1992; Windsor, 1992). However, not all studies in high latitudes have observed asymmetrical CH₄ emission peaks, pointing to the inherent complexity of these ecosystems (Rinne et al., 2007; Tagesson et al., 2012).

Figure 12 Site-year peak methane (CH₄) emissions (a) and peak soil temperature (b) occur earlier in the season for sites with lower mean annual temperatures. (c) Gross primary productivity (GPP) tends to peak earlier in the season for warmer sites, though the trend is weak. All r² values are significant at p < 0.001. Each point represents a site year of data.

Table 2 Site identification (SITE_ID), data DOI, and DOI reference for each FLUXNET-CH₄ site.

3.3.3 Uniqueness of tropical wetlands

Tropical wetlands typically do not experience the large swings in temperature and GPP that contribute to CH₄ flux seasonality in temperate and northern sites. Indeed, the relatively constant high temperatures and high GPP in tropical ecosystems may lead to the lower ratio between seasonal amplitude and peak CH₄ flux compared with temperate and northern sites (Fig. 8b). Tropical flux sites have historically been under-studied, leading to a lack of synthesized information about these ecosystems. FLUXNET-CH₄ has five tropical wetland sites (latitudes between 20^∘ S and 20^∘ N) and one tropical rice site, representing 13 site years of data. These sites are especially insightful as they provide the first estimates of CH₄ fluxes from large, tropical, seasonal floodplain systems.

We found a broad range of annual CH₄ fluxes across tropical sites in FLUXNET-CH₄ Version 1.0. Annual CH₄ flux emissions from two Southeast Asian flooded peat forests were relatively low, 0.01 ± 0.1 and 9.5 ± 0.6 g C m⁻² yr⁻¹ for ID-PAG and MY-MLM, respectively, which is consistent with annual CH₄ fluxes measured at another peat forest in Indonesia (Deshmukh et al., 2020). In contrast, mean annual CH₄ flux for a seasonally flooded swamp in the Brazilian Pantanal region (BR-NPW) was over twice as high as MY-MLM, at 19.2 ± 2.5 g C m⁻² yr⁻¹. Similarly high annual CH₄ fluxes were observed at the two Botswana swamp sites in the Okavango Delta (51.7 ± 10.6 and 47.3 ± 3.7 g C m⁻² yr⁻¹ for BW-GUM and BW-NXR, respectively), one of which is seasonally inundated and surrounded by grassland (BW-NXR) and the other a permanently flooded lagoon covered in a floating papyrus mat (BW-GUM). The relatively low fluxes found at the two Southeast Asian peat forest sites indicate that these ecosystems may be smaller CH₄ sources than expected given their location in the humid tropics. Even the higher-emitting tropical sites in Brazil and Botswana are still well within the range of annual CH₄ flux typical in cooler latitudes (Fig. 1).

In addition to having highly variable CH₄ flux magnitudes, the tropical sites differ from each other in their seasonality. CH₄ flux hit a minimum around July for two sites (BW-GUM, latitude 18.965^∘ S, and MY-MLM, latitude 1.46^∘ N), while CH₄ flux increased through July and the subsequent months for the other Botswana site, BW-NXR (latitude 19.548^∘ S). Site ID-Pag (latitude 2.32^∘ S) had minimal seasonality, whereas the flooded forest site in Brazil (BR-NPW, latitude 16.49^∘ S) had near-zero fluxes from approximately July to January and consistently high fluxes for the remainder of the year. The rice site PH-RiF (latitude 14.14^∘ N) had two annual CH₄ flux peaks, which is consistent with some other rice sites and likely reflects management practices. Baseline CH₄ flux values also differed, with the two Botswana sites having the highest off-season fluxes (29 and 133 nmol m⁻² s⁻¹ for BW-NXR and BW-GUM, respectively, estimated by TIMESAT), MY-MLM having an intermediate baseline CH₄ flux (16 nmol m⁻² s⁻¹, estimated by TIMESAT), and the remainder of the sites having essentially zero flux at baseline. While more tropical wetland data will be needed to extract broad-scale conclusions about these ecosystems, the six tropical sites in FLUXNET-CH₄ provide an important starting point for synthesis studies and highlight tropical wetland CH₄ variability.

4 Data availability

Half-hourly and daily aggregations are available for download at https://fluxnet.org/data/fluxnet-ch4-community-product/ (for citations, please cite this study), along with a table containing site metadata compiled from Table B3. Variable descriptions and units are provided in Table B1 and at https://fluxnet.org/data/fluxnet-ch4-community-product/ (last access: 7 April 2021). Each site has a unique FLUXNET-CH₄ DOI as listed in Table B3. All site data used in this analysis are available under the CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/, last access: 6 July 2021) copyright policy (two additional sites in FLUXNET-CH₄ are available under the more restrictive Tier 2 data policy, https://fluxnet.org/data/data-policy/ (last access: 6 July 2021); these sites are not used in our analysis). The individual site DOIs are provided below in Table 2. All seasonality parameters used in these analyses are available at https://doi.org/10.5281/zenodo.4672601 (Delwiche et al., 2021).

5 Conclusions

The breadth and scope of CH₄ flux data in the FLUXNET-CH₄ dataset make it possible to study the global patterns of CH₄ fluxes, particularly for global freshwater wetlands which release a substantial fraction of atmospheric CH₄. To help data users understand seasonal patterns within the dataset, we provide the first global estimates of CH₄ flux patterns and predictors in CH₄ seasonality using freshwater wetland data. In the seasonality analysis, we find that, on average, the seasonal increase in CH₄ emissions begins about 3 months earlier and lasts about 4 months longer at the warmest sites compared with the coolest sites. We also find that the beginning of the CH₄ emission season lags behind the beginning of seasonal soil warming by approximately 1 month with almost no instances of CH₄ emissions increasing before temperature increases. Additionally, roughly half the sites have CH₄ emissions increasing prior to GPP increase, highlighting the importance of substrate versus temperature limitations on wetland CH₄ emissions. Furthermore, relative to warmer climates, wetland CH₄ emissions in cooler climates increase faster in the warming season and decrease slower in the cooling season. This phenomenon has previously been noted on a regional scale, and we show that it persists at the global scale. Constraining the seasonality of CH₄ fluxes on a global scale can help improve the accuracy of global wetland models. FLUXNET-CH₄ is an important new resource for the research community, but critical data gaps and opportunities remain. The current FLUXNET-CH₄ dataset is biased towards sites in boreal and temperate regions, which influence the relationships presented in our analyses. Tropical ecosystems are estimated to account for 64 % of potential natural CH₄ emissions (< 30^∘ N; Saunois et al., 2020) but only account for 13 % of the FLUXNET-CH₄ sites in the dataset. Unsurprisingly, tropical sites in our network do not represent the range of bioclimatic wetland conditions present in the tropics. Therefore, while maintaining flux towers in tropical ecosystems is challenging, it is necessary to further constrain the global CH₄ cycle. Coastal wetlands are also poorly represented in FLUXNET-CH₄ even though there is evidence of substantial CH₄ emissions from these ecosystems, and so better representation across salinity gradients is warranted. Lastly, the average time series for FLUXNET-CH₄ Version 1.0 is relatively short, only 3.7 site years on average compared with 7.2 for CO₂ sites in FLUXNET (Pastorello et al., 2020). Adding additional site years of data from existing sites, as a complement to adding new sites, will increase the community's ability to explain interannual variability in CH₄ emissions and seasonality. Nevertheless, FLUXNET-CH₄ is an important and unprecedented resource with which to diagnose and understand drivers of the global CH₄ cycle.

Appendix A

Figure A1 Peak methane (CH₄) flux timing versus peak gross primary productivity (GPP) timing (a) and peak soil temperature timing by day of year (b). Points represent site average, and error bars represent standard deviations. Dotted line represents 1:1 relationship.

Appendix B: FLUXNET-CH₄ data variables

This web page describes data variables and file formatting for the FLUXNET-CH₄ Community Product.

B1 Data variable: base names

Base names indicate fundamental quantities that are either measured or calculated/derived. They can also indicate quantified quality information.

B2 Data variable: qualifiers

Qualifiers are suffixes appended to variable base names that provide additional information about the variable. For example, the _DT qualifier in the variable label GPP_DT indicates that gross primary production (GPP) has been partitioned using the flux partitioning method from Lasslop et al. (2010).

Multiple qualifiers can be added, and they must follow the order in which they are presented here.

B2.1 Qualifiers: general

General qualifiers indicate additional information about a variable.

• _F: variable has been gap-filled by the FLUXNET-CH₄ team. Gaps in meteorological variables, including air temperature (TA), incoming shortwave (SW_IN) and longwave (LW_IN) radiation, vapor pressure deficit (VPD), pressure (PA), precipitation (P), and wind speed (WS), were filled with ERA-Interim (ERA-I) reanalysis data (Vuichard and Papale, 2015). Other variables were filled using the multidimensional scaling (MDS) approach in REddyProc (see Delwiche et al., 2021, for more details).

• _DT: variable is acquired using the flux partitioning method from Lasslop et al. (2010), with values estimated by fitting the light-response curve.

• _NT: variable is acquired using the flux partitioning method from Reichstein et al. (2005), with values estimated from nighttime data and extrapolated to daytime.

• _RANDUNC: random uncertainty is introduced from several different sources including errors associated with the flux measurement system (gas analyzer, sonic anemometer, data acquisition system, flux calculations), errors associated with turbulent transport, and statistical errors relating to the location and activity of the sites of flux exchange (“footprint heterogeneity”) (Hollinger and Richardson, 2005).

• _ANNOPTLM: gap-filled variable uses an artificial neural net routine from Matlab with the Levenberg–Marquardt algorithm as the training function and parameters optimized across runs (more detail in Knox et al., 2016, 2019).

• _UNC: uncertainty is introduced from ANNOPTLM gap-filling routine, as described in Knox et al. (2016, 2019).

• _QC: this reports quality checks on FCH4 gap-filled data (_ANNOPTLM) based on length of data gap: 1 signifies data gap shorter than 2 months, and 3 signifies data gap exceeding 2 months which could lead to poor-quality gap-filled data.

B2.2 Qualifiers: positional (_V)

Positional qualifiers are used to indicate relative positions of observations at the site. For FLUXNET-CH₄, positional qualifiers are used to distinguish soil temperature probes for sites with more than one probe. Probe depths for each positional qualifier per site are included in the metadata file included with data download and also in Table B7 of Delwiche et al. (2021). For sites where the original database file release in AmeriFlux, AsiaFlux, or EuroFlux contains multiple probes at the same _V depth, we average values and report only the average for each _V position. The one exception to this is site US-UAF where the original positional qualifier from the data we downloaded from AmeriFlux had different depths for the same qualifier. We still averaged the probe data, so _V qualifiers from US-UAF represent an average of more than one depth.

B3 Missing data

Missing data are reported using −9999. Data for all days in a leap year are reported.

Table B1 Data variable names, descriptions, and units.

Table B2 Annual methane flux sum and uncertainty, annual mean soil temperature, and annual mean water table depth. Column headers are explained after the table.

^* Data from ID-Pag spans 365 d from June 2016 to June 2017. Annual methane flux for each year is the sum of these 365 d, with uncertainty being calculated separately for each year.

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
Year	Data year
Ann_Flux_g_C_m−2	Total annual methane flux (g C m−2)
Ann_Flux_Uncertainty_g_C_m−2	Gap-filling and random uncertainty associated with annual flux (g C m−2)
Mean_Soil_Temp_C	Annual mean soil temperature (degree C). For sites with multiple probes, we use the probe closest to the surface.
Mean_Water_Table_Depth_m	Annual mean water table depth (m)

Table B3 Metadata and select data for FLUXNET-CH₄ sites.

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
SITE_NAME	Site name determined by site personnel
SITE_PERSONNEL	People associated with site FLUXNET-CH4 data
COUNTRY	Site country
LAT	Latitude
LON	Longitude
DATA_DOI	DOI link for site FLUXNET-CH4 data
YEAR_START	Year data begin
YEAR_END	Year data end
UTC_OFFSET	Site data offset from coordinated universal time (in hours)
ORIGINAL_DATA_SOURCE	Regional network hosting the site's methane data that were incorporated into FLUXNET-CH4
SITE_CLASSIFICATION	Site classification based on the literature description of sites
UPLAND_CLASS	For upland sites, category of upland type
IGBP	International Geosphere–Biosphere Programme (IGBP) ecosystem surface classification
KOPPEN	Koppen climate zone abbreviation
MEAN_ANNUAL_TEMP_C_WORLDCLIM	Mean annual temperature from WorldClim2 Global Climate Data
MEAN_ANNUAL_PRECIP_MM_WORLDCLIM	Mean annual precipitation from WorldClim2 Global Climate Data
MOSS_BROWN	Presence/absence (1/0) brown moss. Presence/absence designated by Avni Malhotra using site literature
MOSS_SPHAGNUM	Presence/absence (1/0) sphagnum moss. Presence/absence designated by Avni Malhotra using site literature
AERENCHYMATOUS	Presence/absence (1/0) aerenchymatous vegetation. Presence/absence designated by Avni Malhotra using site literature
ERI_SHRUB	Presence/absence (1/0) ericaceous shrubs. Presence/absence designated by Avni Malhotra using site literature
TREE	Presence/absence (1/0) trees. Presence/absence designated by Avni Malhotra using site literature
DOM_VEG	Dominant vegetation type in tower footprint. Dom_veg provided to Avni Malhotra by site personnel via survey, except 15 sites where principal investigators did not answer and Avni Malhotra estimated dominant vegetation type based on site literature
IN_SEASONALITY_ANALYSIS	Is site in freshwater wetland seasonality analysis? 1 = yes, 0 = no.
Mean_Air_Temp_C	Mean annual air temperature (C)
Mean_Air_Temp_stdev_C	Standard deviation of annual air temperature (C)
Ann_Flux_g_CH4-C_m−2	Mean annual methane flux (g CH4-C m−2 yr1)
Ann_Flux_stdev_g_CH4-C_m−2	Standard deviation of annual methane flux (g CH4-C m−2 yr−1)
JFM_flux_g_CH4-C_m−2	Mean methane flux in January, February, March (g CH4-C m−2 yr−1)
JFM_flux_stdev_g_CH4-C_m−2	Standard deviation of methane flux in January, February, March (g CH4-C m−2 yr−1)
AMJ_flux_g_CH4-C_m−2	Mean methane flux in April, May, June (g CH4-C m−2 yr−1)
AMJ_flux_stdev_g_CH4-C_m−2	Standard deviation of methane flux in April, May, June (g CH4-C m−2 yr−1)
JAS_flux_g_CH4-C_m−2	Mean methane flux in July, August, September (g CH4-C m−2 yr−1)
JAS_flux_stdev_g_CH4-C_m−2	Standard deviation of methane flux in July, August, September (g CH4-C m−2 yr−1)
OND_flux_g_CH4-C_m−2	Mean methane flux in October, November, December (g CH4-C m−2 yr−1)
OND_flux_stdev_g_CH4-C_m−2	Standard deviation of methane flux in October, November, December (g CH4-C m−2 yr−1)
SOIL_TEMP_PROBE_DEPTHS	Depth of soil temperature probe (m), with negative values being under the surface

Table B4 Table of bioclimatic predictor data used in the principal component analysis (PCA) of Fig. 6.

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
Enhanced_Vegetation_Index_(EVI)	Enhanced vegetation index (unitless) from MOD13A3 (Didan, 2015), 2001–2018 monthly data
Wong_Simple_Ratio_Water_Index_(SRWI)	Simple ratio water index (unitless) from MOD09A1 (Vermote, 2015), ∼ 2001–2018 monthly data
Latent_Heat_(LE)	Latent heat (in W m−2) from FLUXCOM (Jung et al., 2019), 2003–2013 monthly data
Mean_Annual_Temperature_(MAT)	Mean annual temperature (C) from BioClim (Fick and Hijman, 2017), 2001–2018 monthly data

Table B5 Seasonality parameters estimated using TIMESAT software for methane flux (FCH4), gross primary productivity (GPP), air temperature (TA), and soil temperature (TS, for shallowest probe at each site).

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
Year	Data year
Start_FCH4_(DOY)	Season start for elevated methane fluxes (DOY), point “f” in Fig. 1
End_FCH4_(DOY)	Season end for elevated methane fluxes (DOY), point “h” in Fig. 1
Base_value_FCH4_(nmolCH4/m2/s)	Baseline methane flux during non-elevated season (nmol CH4 m−2 s−1), average of points “a” and “b” in Fig. 1
Ampl_FCH4_(nmolCH4/m2/s)	Amplitude of methane flux during elevated flux season (nmol CH4 m−2 s−1), difference between point “e” in Fig. 1 and Base_value_FCH4
Peak_FCH4_(DOY)	Day of maximum elevated methane flux (DOY), point “g” in Fig. 1
Peak_value_FCH4_(nmolCH4/m2/s)	Maximum value of methane flux (nmol CH4 m−2 s−1), point “e” in Fig. 1
Start_GPP_DT_(DOY)	Season start for elevated GPP_DT (DOY), point “f” in Fig. 1
End_GPP_DT_(DOY)	Season end for elevated GPP_DT fluxes (DOY), point “h” in Fig. 1
Base_value_GPP_DT_(µmolCO2/m2/s)	Baseline GPP_DT flux during non-elevated season (µmol CO2 m−2 s−1), average of points “a” and “b” in Fig. 1
Ampl_GPP_DT_(µmolCO2/m2/s)	Amplitude of GPP_DT flux during elevated flux season (µmol CO2 m−2 s−1), difference between point “e” in Fig. 1 and Base_value_GPP_DT
Peak_GPP_DT_(DOY)	Day of maximum elevated GPP_DT flux (DOY), point “g” in Fig. 1
Peak_value_GPP_DT_(µmolCO2/m2/s)	Maximum value of GPP_DT flux (µmol CO2 m−2 s−1), point “e” in Fig. 1
Probe_name	Temperature probe name as given in data files
Soil_temp_depth_m	Depth of soil temperature probe (m), with negative values being under the surface
Start_TS_(DOY)	Season start for elevated TS (DOY), point “f” in Fig. 1
End_TS_(DOY)	Season end for elevated TS (DOY), point “h” in Fig. 1
Base_value_TS_(C)	Baseline TS during non-elevated season (C), average of points “a” and “b” in Fig. 1
Ampl_TS_(C)	Amplitude of TS during elevated temperature season (C), difference between point “e” in Fig. 1 and Base_value_TS
Peak_TS_(DOY)	Day of maximum elevated TS (DOY), point “g” in Fig. 1
Peak_value_TS_(C)	Maximum value of TS (C), point “e” in Fig. 1
Start_TA_(DOY)	Season start for elevated TA (DOY), point “f” in Fig. 1
End_TA_(DOY)	Season end for elevated TA (DOY), point “h” in Fig. 1
Base_value_TA_(C)	Baseline TA during non-elevated season (C), average of points “a” and “b” in Fig. 1
Ampl_TA_(C)	Amplitude of TA during elevated temperature season (C), difference between point “e” in Fig. 1 and Base_value_TA
Peak_TA_(DOY)	Day of maximum elevated TA (DOY), point “g” in Fig. 1
Peak_value_TA_(C)	Maximum value of TA (C), point “e” in Fig. 1

Table B6 Seasonality parameters estimated using TIMESAT software for soil temperature (TS, from every probe).

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
Year	Data year
Probe_name	Temperature probe name as given in data files
Soil_temp_depth_m	Depth of soil temperature probe (m), with negative values being under the surface
Start_TS_(DOY)	Season start for elevated TS (DOY), point “f” in Fig. 1
End_TS_(DOY)	Season end for elevated TS (DOY), point “h” in Fig. 1
Base_value_TS_(C)	Baseline TS during non-elevated season (C), average of points “a” and “b” in Fig. 1
Ampl_TS_(C)	Amplitude of TS during elevated temperature season (C), difference between point “e” in Fig. 1 and Base_value_TS
Peak_TS_(DOY)	Day of maximum elevated TS (DOY), point “g” in Fig. 1
Peak_value_TS_(C)	Maximum value of TS (C), point “e” in Fig. 1

Table B7 Installation depths for soil temperature probes.

Column	Description
SITE_ID	Site identification code as assigned by regional flux data network
Year	When relevant, information about time span of probe location; if blank, assume constant probe depth
Probe_name	Temperature probe name as given in data files
Soil_temp_depth_m	Depth of soil temperature probe (m), with negative values being under the surface
Additional_notes	When relevant, additional information about site