We describe a new coordination activity and initial results for a global synthesis of eddy covariance CH4 flux measurements.
Atmospheric methane (CH4) is the second-most important anthropogenic greenhouse gas following carbon dioxide (CO2) (Myhre et al. 2013). The concentration of CH4 in the atmosphere today is about 2.5 times higher than in 1750 (Saunois et al. 2016a). The increase in atmospheric CH4 has arisen from human activities in agriculture, energy production, and waste disposal, and from changes in natural CH4 sources and sinks (Saunois et al. 2016a,b, 2017; Turner et al. 2019). Based on top-down atmospheric inversions, global CH4 emissions for the decade of 2003–12 were an estimated ∼420 Tg C yr–1 (range 405–426 Tg C yr–1) (Saunois et al. 2016a). However, some analyses suggest that uncertainties in global CH4 sources and sinks are higher than those for CO2, and uncertainties from natural sources exceed those from anthropogenic emissions (Saunois et al. 2016a). In particular, the largest source of uncertainty in the global CH4 budget is related to emissions from wetlands and inland waters (Saunois et al. 2016a; Melton et al. 2013; Bastviken et al. 2011). Wetland CH4 emissions may contribute as much as 25%–40% of the global total and are a leading source of interannual variability in total atmospheric CH4 concentrations (Bousquet et al. 2006; Chen and Prinn 2006; Saunois et al. 2016a).
Direct, ground-based measurements of in situ CH4 fluxes with high measurement frequency are important for understanding the responses of CH4 fluxes to environmental factors including climate, for providing validation datasets for the land surface models used to infer global CH4 budgets, and for constraining CH4 budgets. Eddy covariance (EC) flux towers measure real-time exchange of gases such as CO2, CH4, water vapor, and energy between the land surface and the atmosphere. The EC technique has emerged as a widespread means of measuring trace gas exchange because it provides direct and near-continuous ecosystem-scale flux measurements without disturbing the soil or vegetation (Baldocchi 2003; Aubinet et al. 2012). There are more than 900 reported active and historical flux tower sites globally and approximately 7,000 site years of data collected (Chu et al. 2017). While most of these sites measure CO2, water vapor, and energy exchange, the development of new and robust CH4 sensors has resulted in a rapidly growing number of CH4 EC measurements (Baldocchi 2014; Morin 2018), primarily in natural and agricultural wetlands (Petrescu et al. 2015).
Since the late 1990s, with a growing number of long-term, near-continuous EC measurements, the EC community has been well coordinated for integrating and synthesizing CO2, water vapor and energy fluxes. This cross-site coordination resulted in the development of regional flux networks for Europe [EuroFlux, CarboEurope, and Integrated Carbon Observing System (ICOS)], Australia (OzFlux), North and South America (AmeriFlux, Large Biosphere Amazon, Fluxnet-Canada/Canadian Carbon Program, and MexFlux), Asia [AsiaFlux, ChinaFlux, KoFlux, and U.S.–China Carbon Consortium (USCCC)], and globally, FLUXNET (Papale et al. 2012; Baldocchi 2014). The resulting FLUXNET database (http://fluxnet.fluxdata.org/) has been used extensively to evaluate satellite measurements, inform Earth system models, generate data-driven CO2 flux products, and provide answers to a broad range of questions about atmospheric fluxes related to ecosystems, land use and climate (Pastorello et al. 2017). FLUXNET has grown steadily over the past 25 years, enhancing our understanding of carbon, water and energy cycles in terrestrial ecosystems (Chu et al. 2017).
Similar community efforts and syntheses for CH4 remain limited in part because EC measurements for CH4 fluxes were rarer until recently. Whereas the earliest EC measurements of CO2 fluxes date back to the late 1970s and early 1980s (Desjardins 1974; Anderson et al. 1984), the first EC CH4 flux measurements only began in the 1990s (Verma et al. 1992; Shurpali and Verma 1998; Fan et al. 1992; Kim et al. 1999), with reliable, easy-to-deploy field sensors only becoming available in the past decade or so. EC CH4 flux measurements became more feasible with advances in sensor development, such as tunable diode laser absorption spectrometers, that allowed researchers to measure previously undetectable trace gas fluxes with higher signal to noise ratios (Rinne et al. 2007; McDermitt et al. 2011). After these new sensors were commercialized, and low-power, low-maintenance open-path sensors were developed that could be operated by solar panels in remote locations, the number of CH4 flux tower measurements increased substantially (Baldocchi 2014; Morin 2018). The rapidly growing number of EC CH4 flux measurements presents new opportunities for FLUXNET-type analyses and syntheses of ecosystem-scale CH4 flux observations.
This manuscript describes initial results from a new coordination activity for flux tower CH4 measurements organized by the Global Carbon Project (GCP) in collaboration with regional flux networks and FLUXNET. The goal of the activity is to develop a global database for EC CH4 observations to answer regional and global questions related to CH4 cycling. Here, we describe the objectives of the FLUXNET-CH4 activity, provide an overview of the current geographic and temporal coverage of CH4 flux measurements globally, present initial analyses exploring time scales of variability, uncertainty, trends, and drivers of CH4 fluxes across 60 sites, and discuss future research opportunities for examining controls on CH4 emissions and reducing uncertainties in the role of wetlands in the global CH4 cycle.
FLUXNET-CH4 SYNTHESIS OBJECTIVES AND TASKS.
This activity is part of a larger GCP effort to establish and better constrain the global methane budget (www.globalcarbonproject.org/methanebudget/index.htm), and is designed to develop a CH4 database component in FLUXNET for a global synthesis of CH4 flux tower data. To this end, we are surveying, assembling, and synthesizing data from the EC community, in coordination with regional networks, including AmeriFlux’s 2019 “Year of Methane” (http://ameriflux.lbl.gov/year-of-methane/year-of-methane/), FLUXNET initiatives, and other complementary activities. In particular, this work is being carried out in parallel with the EU’s Readiness of ICOS for Necessities of Integrated Global Observations (RINGO) project, which is working to standardize protocols for flux calculations, quality control and gap-filling for CH4 fluxes (Nemitz et al. 2018). Methane-specific protocols are needed because of the added complexities and high variability of CH4 flux measurements and dynamics (Nemitz et al. 2018).
Our approach is to include all currently available and future CH4 flux tower observations in a global CH4 database, including freshwater, coastal, natural, and managed ecosystems, as well as upland ecosystems that may be measuring CH4 uptake by soils. The initiative is open to all members of the EC community. Database compilation began in 2017 and is ongoing. Data from sites in the Americas can be submitted to AmeriFlux (http://ameriflux.lbl.gov/data/how-to-uploaddownload-data/); otherwise, data can be submitted to the European Fluxes Database Cluster (www.europe-fluxdata.eu/home/sites-list).
In addition to many applications, an ultimate goal of the FLUXNET-CH4 activity is to generate a publicly available, open-access, data-driven global CH4 emissions product using similar machine-learning-based approaches used for CO2 fluxes (Jung et al. 2009; Tramontana et al. 2016). The product will be based on mechanistic factors associated with CH4 emissions and new spatiotemporal information on wetland area and dynamics for constraining CH4-producing areas. This gridded product will provide an independent bottom-up estimate of global wetland CH4 emissions to compare with estimates of global CH4 emissions from land surface models and atmospheric inversions. Recent work has shown the potential to upscale EC CH4 flux observations across northern wetlands, with predictive performance comparable to previous studies upscaling net CO2 exchange (Peltola et al. 2019); however, our focus is on a globally gridded product.
The near-continuous, high-frequency nature of EC measurements also offers significant promise for improving our understanding of ecosystem-scale CH4 flux dynamics. As such, this synthesis also aims to investigate the dominant controls on net ecosystem-scale CH4 fluxes from hourly to interannual time scales across wetlands globally, and to characterize scale-emergent, nonlinear, and lagged processes of CH4 exchange.
Methane is produced during decomposition under anaerobic or reducing conditions and is transported to the atmosphere via plant-mediated transport, ebullition, and diffusion (Bridgham et al. 2013). During transport, CH4 can pass through unsaturated soil layers and be consumed or oxidized by aerobic bacteria (Wahlen 1993). Process-based biogeochemical models developed and applied at site, regional, and global scales simulate these individual processes with varying degrees of complexity (Bridgham et al. 2013; Melton et al. 2013; Poulter et al. 2017; Castro-Morales et al. 2018; Grant and Roulet 2002). The large range in predicted wetland CH4 emissions rates suggests that there is both substantial parameter and structural uncertainty in large-scale CH4 flux models, even after accounting for uncertainties in wetland areas (Poulter et al. 2017; Saunois et al. 2016a; Melton et al. 2013; Riley et al. 2011). A global EC CH4 database and associated environmental variables can help constrain the parameterization of process-based biogeochemistry models (Saunois et al. 2016a; Bridgham et al. 2013; Oikawa et al. 2017). Furthermore, a key challenge is evaluating globally applicable process-based CH4 models at a spatial scale comparable to model grid cells (Melton et al. 2013; Riley et al. 2011). A globally gridded wetland CH4 emissions product upscaled from EC fluxes can help resolve this issue by providing a scale-appropriate model evaluation dataset. As such, the global CH4 database and gridded product will also be used to parameterize and benchmark the performance of land surface models of global CH4 emissions, providing a unique opportunity for informing and validating biogeochemical models.
METHODS.
Based on a survey of the EC community (announced via the fluxnet-community@george.lbl.gov and AmeriFlux-Community@lbl.gov listservs), information available in regional networks and FLUXNET, and the scientific literature, we estimate that at least 200 sites worldwide are currently applying the EC method for CH4 flux measurements (Fig. 1). Here we focus on findings from across 60 of the ∼110 sites currently committed to participating in our FLUXNET-CH4 activity [Table A1 in the appendix and Table ES1 in the online supplemental material]. Data from this initial set of sites were selected because they were publicly available or were contributed directly by site principal investigators (PIs). We will continue to engage the EC community more broadly and expand the database in the future.
Location of the 200 tower sites that report eddy covariance CH4 flux measurements worldwide. Triangles indicate sites from which data are included in this manuscript, with circles indicating additional flux towers measuring CH4 emissions. The colors of the markers represent the vegetation type based on the International Geosphere-Biosphere Programme (IGBP) definition. See Table ES1 for a list of sites, their characteristics, and years of operation. Sites are overlaid over a map of the differences between the average CH4 emissions over 2000–10 between top-down and bottom-up wetland CH4 estimates. Top-down estimates are represented by the natural fluxes inventoried in NOAA’s CarbonTracker (www.esrl.noaa.gov/gmd/ccgg/carbontracker-ch4/). Bottom-up emissions were produced from an ensemble of 11 Earth system model simulations (Poulter et al. 2017).
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Location of the 200 tower sites that report eddy covariance CH4 flux measurements worldwide. Triangles indicate sites from which data are included in this manuscript, with circles indicating additional flux towers measuring CH4 emissions. The colors of the markers represent the vegetation type based on the International Geosphere-Biosphere Programme (IGBP) definition. See Table ES1 for a list of sites, their characteristics, and years of operation. Sites are overlaid over a map of the differences between the average CH4 emissions over 2000–10 between top-down and bottom-up wetland CH4 estimates. Top-down estimates are represented by the natural fluxes inventoried in NOAA’s CarbonTracker (www.esrl.noaa.gov/gmd/ccgg/carbontracker-ch4/). Bottom-up emissions were produced from an ensemble of 11 Earth system model simulations (Poulter et al. 2017).
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Location of the 200 tower sites that report eddy covariance CH4 flux measurements worldwide. Triangles indicate sites from which data are included in this manuscript, with circles indicating additional flux towers measuring CH4 emissions. The colors of the markers represent the vegetation type based on the International Geosphere-Biosphere Programme (IGBP) definition. See Table ES1 for a list of sites, their characteristics, and years of operation. Sites are overlaid over a map of the differences between the average CH4 emissions over 2000–10 between top-down and bottom-up wetland CH4 estimates. Top-down estimates are represented by the natural fluxes inventoried in NOAA’s CarbonTracker (www.esrl.noaa.gov/gmd/ccgg/carbontracker-ch4/). Bottom-up emissions were produced from an ensemble of 11 Earth system model simulations (Poulter et al. 2017).
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Data standardization, gap-filling, and partitioning.
We used similar data processing procedures as FLUXNET to standardize and gap-fill measurements, and in the case of net CO2 exchange, partition fluxes across sites (http://fluxnet.fluxdata.org/data/aboutdata/data-processing-101-pipeline-and-procedures/). Standard quality assurance and quality control of the data were first performed by site PIs. In nearly all cases, data collected by the local tower teams were first submitted to the data archives hosted by the regional flux networks, where data are prescreened and formatted based on the regional network data protocols. Data from the regional networks then entered our flux processing procedure.
Within our processing procedure, data were first checked for obvious problems including unit errors, spikes, and out-of-range values based on visualization of the data and statistical metrics. Next, the data were filtered, gap-filled, and partitioned. Friction velocity (
There are as yet no standards for gap-filling CH4 flux measurements and this is an active and ongoing area of research (Nemitz et al. 2018). Gaps in CH4 fluxes were filled using artificial neural networks (ANNs), as they have shown good performance for gap-filling CH4 flux data (Dengel et al. 2013; Knox et al. 2015; Morin et al. 2014a; Nemitz et al. 2018; Goodrich et al. 2015). Details of the ANN routine are provided in Knox et al. (2016) and are summarized here briefly. The ANN routine was optimized for both generalizability and representativeness. To facilitate representativeness, explanatory data were divided into a maximum of 15 data clusters using the k-means algorithm. To avoid biasing toward conditions with better flux data coverage (e.g., summer and daytime), data used to train, test, and validate the ANN were proportionately sampled from these clusters. Several neural network architectures of increasing complexity were tested, ranging from one hidden layer with the number of nodes equal to the number of explanatory data variables (N) to two hidden layers with 1.5N and 0.75N nodes, respectively. The architecture of each neural network was initialized 10 times with random starting weights, and the initialization resulting in the lowest mean sampling error was selected. The simplest architecture, whereby additional increases in complexity resulted in <5% reduction in mean squared error, was chosen and the prediction saved. This procedure was repeated with 20 resamplings of the data, and missing half hours were filled using the median prediction. A standard set of variables available across all sites were used to gap-fill CH4 fluxes (Dengel et al. 2013), including TA, SWIN, WS, PA, and sine and cosine functions to represent seasonality. These meteorological variables were selected since they are relevant to CH4 exchange and were gap-filled using the ERA-I reanalysis data. Other variables related to CH4 exchange such as water table depth (WTD) or 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. These missing data for variables highlight some of the key challenges in standardizing CH4 gap-filling methods across sites and emphasize the need for standardized protocols of auxiliary measurements across sites (cf. “Future research directions and needs” section) (Nemitz et al. 2018; Dengel et al. 2013). ANN gap-filling was performed using MATLAB (MathWorks 2018, version 9.4.0).
Annual CH4 budgets represent gap-filled, half-hourly fluxes integrated over an entire year or growing season. If fluxes were only measured during the growing season, we assumed that fluxes outside of this period were negligible, although we acknowledge that cold season fluxes can account for as much as ∼13%–50% of the annual CH4 emissions in some locations (Zona et al. 2016; Treat et al. 2018b; Helbig et al. 2017a; Kittler et al. 2017).
Uncertainty estimation.
ANNs were also used to estimate annual gap-filled and random uncertainty in CH4 flux measurements (Richardson et al. 2008; Moffat et al. 2007; Anderson et al. 2016; Knox et al. 2018). Here, we focus on assessing the random error, but a full assessment of total flux measurement error also requires quantifying systematic error or bias (Baldocchi 2003). Systematic errors, due to incomplete spectral response, lack of nocturnal mixing, submesoscale circulations, and other factors are discussed elsewhere (Baldocchi 2003; Peltola et al. 2015) and are the focus of other ongoing initiatives.
Random errors in EC fluxes follow a double exponential (Laplace) distribution with a standard deviation varying with flux magnitude (Richardson et al. 2012, 2006). Model residuals of gap-filling algorithms such as ANNs provide a reliable, and conservative “upper limit,” estimate of the random flux uncertainty (Moffat et al. 2007; Richardson et al. 2008). For half-hourly CH4 flux measurements, random error was estimated using the residuals of the median ANN predictions. At each site, the probability density function (PDF) of the random flux measurement error more closely followed a double-exponential (Laplace) rather than normal (Gaussian) distribution, with the root-mean-square error (RMSE) for the Laplace distribution fitted to the PDF of random errors consistently lower than the normal distributed error. From half-hourly flux measurements, random error can also be estimated using the daily differencing approach (Richardson et al. 2012). Random error estimates [σ(δ)], as expressed as the standard deviation of the double-exponential distribution with scaling parameter β, where
Wavelet-based time-scale decomposition.
Methane fluxes are highly dynamic and vary across a range of time scales (Sturtevant et al. 2016; Koebsch et al. 2015). For example, in wetlands with permanent inundation, the seasonal variation of CH4 exchange is predominantly controlled by temperature and plant phenology (Chu et al. 2014; Sturtevant et al. 2016). Ecosystem CH4 exchange also varies considerably at both longer (e.g., interannual; Knox et al. 2016; Rinne et al. 2018) and shorter (e.g., weeks, days, or hours; Koebsch et al. 2015; Hatala et al. 2012; Schaller et al. 2018) time scales. Wavelet decomposition is a particularly useful tool for investigating scale in geophysical and ecological analysis (Cazelles et al. 2008; Torrence and Compo 1998), because it can characterize both the time scale and location of patterns and perturbations in the data. Partitioning variability across temporal scales can help to isolate and characterize important processes (Schaller et al. 2018).
The maximal overlap discrete wavelet transform (MODWT) was used to decompose the time scales of variability in gap-filled CH4 flux measurements, as described in Sturtevant et al. (2016). The MODWT allows the time series to be decomposed into the detail added from progressively coarser to finer scales and either summed or treated individually to investigate patterns across scales. We reconstructed the detail in the fluxes for dyadic scales 1 (21 measurements = 1 h) to 14 (214 measurements = 341 days). Since patterns generated by ecological processes tend to occur over a scale range rather than at one individual scale, the detail over adjacent scales were summed to analyze four general time scales of variation (Sturtevant et al. 2016). These time scales included the “hourly” scale (1–2 h) representing perturbations such the passage of clouds overhead and turbulent scales up to the spectral gap, the “diel” scale (4 h to 1.3 days) encompassing the diel cycles in sunlight and temperature, the “multiday” scale (2.7 to 21.3 days) reflecting synoptic weather variability or fluctuations in water levels, and the “seasonal” scale (42.7 to 341 days) representing the annual solar cycle and phenology. Data were wavelet decomposed into the hourly, diel, and multiday scales using the Wavelet Methods for Time Series Analysis (WMTSA) Wavelet Toolkit in MATLAB.
Statistical analysis.
We tested for significant relationships between log-transformed annual CH4 emissions and a number of covariates using linear mixed-effects models as described in Treat et al. (2018b). The predictor variables of CH4 flux we evaluated included: biome or ecosystem type (categorical variables), and continuous biophysical variables including mean seasonal WTD, mean annual soil and air temperature (TMST and TMAT, respectively), net ecosystem exchange (NEE), GPP, and ER. When considering continuous variables, we focused on freshwater wetlands for comparison with previous CH4 synthesis activities. Soil temperature was measured between 2 and 25 cm below the surface in different studies. The results below are presented for GPP and ER covariates that are partitioned using the nighttime flux partitioning algorithm (Wutzler et al. 2018; Reichstein et al. 2005), although similar findings were obtained using daytime partitioned estimates. Additionally, individual sites or site years were excluded when gaps in measurements exceeded two consecutive months, which explains the differences in the number of sites and site years in the “Environmental controls on annual CH4 emissions across freshwater wetland sites” section below.
Mixed-effects modeling was used because of the potential bias of having measurements over several years, with site included as a random effect in the analysis (Treat et al. 2018b). The significance of individual predictor variables was evaluated using a χ2 test against a null model using only site as a random variable (Bates et al. 2015), with both models fit without reduced maximum likelihood. For multiple linear regression models, we used the model selection process outlined in Zuur et al. (2009). To incorporate annual cumulative uncertainty when assessing the significance of trends and differences in annual CH4 fluxes across biomes and ecosystem types, we used a Monte Carlo simulation that randomly draws 1,000 annual cumulative uncertainties for each estimate of annual CH4 flux. For each random draw the significance of the categorical variable was tested using a χ2 test against the null model with only site as a random variable. We report the marginal r2 (r2m), which describes the proportion of variance explained by the fixed factors alone (Nakagawa and Schielzeth 2013). The mixed-effects modeling was implemented using the lmer command from the lme4 package (Bates et al. 2015) for R statistical software.
RESULTS AND DISCUSSION.
Geographic and temporal coverage of eddy covariance CH4 flux measurements.
We identified 200 sites worldwide that are applying the EC method for CH4 (Fig. 1; Table ES1); wetlands (including natural, managed, and restored wetlands) comprise the majority of sites (59%), with rice agriculture (10%) as the second-most represented vegetation type. The predominance of wetland and rice paddy sites in the database is unsurprising because many studies are designed to target ecosystems expected to have relatively large CH4 emissions. However, there are also sites in ecosystems that are typically smaller sources or even sinks of CH4 such as upland forests (13%) and grasslands (8%). Additionally, six sites (∼3%) are urban, with another five sites measuring CH4 fluxes from open water bodies. Although identified sites span all continents except Antarctica, the majority are concentrated in North America and Europe, with a growing number of sites in Asia (Fig. 1; Table ES1).
Measurements of CH4 fluxes cover a broad range of climates and a large fraction of wetland habitats (Fig. 2), with the tropics and tropical wetlands notably underrepresented. As discussed below (see “Future research directions and needs” section), one important goal of FLUXNET and the regional networks is to increase site representativeness and extend measurements in undersampled regions. Increasing the number of tropical sites is particularly important for CH4 because more than half of global CH4 emissions are thought to come from this region (Saunois et al. 2016a; Dean et al. 2018). Furthermore, compared to northern wetlands, their biogeochemistry remains relatively poorly understood (Mitsch et al. 2009; Pangala et al. 2017). We expect the number of CH4 flux sites and their geographic and temporal coverage to continue to increase, as has occurred through time for CO2, water vapor, and energy flux measurements in FLUXNET (Pastorello et al. 2017; Chu et al. 2017).
Distribution of sites by mean annual air temperature and precipitation. Tower locations are shown as circles or triangles (see Fig. 1), with vegetation type in color based on the IGBP definitions (CRO = croplands; DBF = deciduous broadleaf forests; EBF = evergreen broadleaf forests; ENF = evergreen needleleaf forests; GRA = grasslands; MF = mixed forests; URB = urban and built-up lands; WAT = water bodies; WET = permanent wetlands). Gray dots represent annual mean temperature and total precipitation from the CRU TS 3.10 gridded climate dataset over the entire landmass (Harris et al. 2014), whereas blue dots represent grid cells with >25% wetland fraction as estimated using the Global Lakes and Wetlands Database (Lehner and Döll 2004). Temperature and precipitation grid cells included in this figure were averaged from 1981 to 2011, at 0.5° resolution.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Distribution of sites by mean annual air temperature and precipitation. Tower locations are shown as circles or triangles (see Fig. 1), with vegetation type in color based on the IGBP definitions (CRO = croplands; DBF = deciduous broadleaf forests; EBF = evergreen broadleaf forests; ENF = evergreen needleleaf forests; GRA = grasslands; MF = mixed forests; URB = urban and built-up lands; WAT = water bodies; WET = permanent wetlands). Gray dots represent annual mean temperature and total precipitation from the CRU TS 3.10 gridded climate dataset over the entire landmass (Harris et al. 2014), whereas blue dots represent grid cells with >25% wetland fraction as estimated using the Global Lakes and Wetlands Database (Lehner and Döll 2004). Temperature and precipitation grid cells included in this figure were averaged from 1981 to 2011, at 0.5° resolution.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Distribution of sites by mean annual air temperature and precipitation. Tower locations are shown as circles or triangles (see Fig. 1), with vegetation type in color based on the IGBP definitions (CRO = croplands; DBF = deciduous broadleaf forests; EBF = evergreen broadleaf forests; ENF = evergreen needleleaf forests; GRA = grasslands; MF = mixed forests; URB = urban and built-up lands; WAT = water bodies; WET = permanent wetlands). Gray dots represent annual mean temperature and total precipitation from the CRU TS 3.10 gridded climate dataset over the entire landmass (Harris et al. 2014), whereas blue dots represent grid cells with >25% wetland fraction as estimated using the Global Lakes and Wetlands Database (Lehner and Döll 2004). Temperature and precipitation grid cells included in this figure were averaged from 1981 to 2011, at 0.5° resolution.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
Long-term CH4 flux time series are key to understanding the causes of year-to-year variability and trends in fluxes (Chu et al. 2017; Euskirchen et al. 2017; Pugh et al. 2018). The longest continuous record of CH4 flux measurements, from a fen in Finland (Rinne et al. 2018), is now ∼14 years and ongoing (Table ES1). Three other sites have measurements exceeding 10 years; however, the median length is 5 years, with most sites established from 2013 onward (Table ES1). Longer time series are also important for both exploring the short- and long-term effects of extreme events on fluxes and tracking the response of disturbed or restored ecosystems over time (Pastorello et al. 2017). Furthermore, they can help address new and emerging science questions, such as quantifying CH4 feedbacks to climate with rising temperatures and associated changes in ecosystem composition, structure and function (Helbig et al. 2017a,b; Dean et al. 2018), and the role of wetland emissions in atmospheric CH4 variability (McNorton et al. 2016; Poulter et al. 2017).
CH4 fluxes and trends across biomes and ecosystem types.
Half-hourly and annual net CH4 fluxes for the 60 sites currently included in the database exhibited strong variability across sites (Figs. 3 and 4). Across the dataset, the mean half-hourly CH4 flux was greater than the median flux, indicating a positively skewed distribution with infrequent, large emissions (Fig. 3a), similar to findings from chamber-based syntheses (Olefeldt et al. 2013; Turetsky et al. 2014). Mean and median CH4 fluxes were smaller at higher latitudes and larger at lower latitudes (Fig. 3b), comparable again to trends in CH4 fluxes observed in predominantly chamber-based syntheses (Bartlett and Harriss 1993; Turetsky et al. 2014; Treat et al. 2018b).
(a) Probability density function, and (b) cumulative frequency distribution of half-hourly CH4 flux (FCH4) data for sites currently included in the database (60 sites) aggregated by biome. Thin lines represent individual sites, whereas thicker lines present sites aggregated by biome. All cases are approximated by kernel density estimation. Note that whereas the x axis is scaled between −50 and 900 nmol m–2 s–1 for visualization purposes, some CH4 fluxes exceed this range.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
(a) Probability density function, and (b) cumulative frequency distribution of half-hourly CH4 flux (FCH4) data for sites currently included in the database (60 sites) aggregated by biome. Thin lines represent individual sites, whereas thicker lines present sites aggregated by biome. All cases are approximated by kernel density estimation. Note that whereas the x axis is scaled between −50 and 900 nmol m–2 s–1 for visualization purposes, some CH4 fluxes exceed this range.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
(a) Probability density function, and (b) cumulative frequency distribution of half-hourly CH4 flux (FCH4) data for sites currently included in the database (60 sites) aggregated by biome. Thin lines represent individual sites, whereas thicker lines present sites aggregated by biome. All cases are approximated by kernel density estimation. Note that whereas the x axis is scaled between −50 and 900 nmol m–2 s–1 for visualization purposes, some CH4 fluxes exceed this range.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
(a) Histogram of annual CH4 fluxes (FCH4; g C m–2 yr–1) measured with eddy covariance and published in the synthesis by Baldocchi (2014), and (b) histogram of our annual CH4 fluxes including additional site years of data estimated from the 60 sites listed in Table A1.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
(a) Histogram of annual CH4 fluxes (FCH4; g C m–2 yr–1) measured with eddy covariance and published in the synthesis by Baldocchi (2014), and (b) histogram of our annual CH4 fluxes including additional site years of data estimated from the 60 sites listed in Table A1.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
(a) Histogram of annual CH4 fluxes (FCH4; g C m–2 yr–1) measured with eddy covariance and published in the synthesis by Baldocchi (2014), and (b) histogram of our annual CH4 fluxes including additional site years of data estimated from the 60 sites listed in Table A1.
Citation: Bulletin of the American Meteorological Society 100, 12; 10.1175/BAMS-D-18-0268.1
The continuous nature of EC flux measurements is well suited for quantifying annual ecosystem-scale CH4 budgets, along with accumulated uncertainty (cf. “Gap-filling performance and uncertainty quantification” section). Annual estimates of net CH4 flux for each of the 60 sites in the flux tower database ranged from −0.2 ± 0.02 g C m–2 yr–1 for an upland forest site to 114.9 ± 13.4 g C m–2 yr–1 for an estuarine freshwater marsh (Rey-Sanchez et al. 2018), with fluxes exceeding 40 g C m–2 yr–1 at multiple sites (Fig. 4b). These emissions are of a considerably broader range and have much higher annual values than in an earlier synthesis by Baldocchi (2014), which included published values from 13 sites (Fig. 4a); median annual CH4 fluxes (±SE) in that study were 6.4 ± 1.9 g C m–2 yr–1, compared with 10.0 ± 1.6 g C m–2 yr–1 for our expanded database. Annual CH4 sums in our database were positively skewed, with skewness increasing with additional observations due largely to the inclusion of high CH4-emitting freshwater marsh sites (Fig. 4).
As suggested from Fig. 3b, annual wetland CH4 emissions differed significantly among biomes, even when considering accumulated uncertainty [average Monte Carlo χ2 = 13.4 (12.1–14.7, 95% confidence interval), degrees of freedom (df) = 3, p < 0.05] (Table 1). Median CH4 emissions were significantly lower for tundra wetlands (2.9 ± 1.3 g C m–2 yr–1) than temperate wetlands (27.4 ± 3.4 g C m–2 yr–1). Higher CH4 emissions were observed from subtropical/tropical wetlands (43.2 ± 11.2 g C m–2 yr–1), based on only three site years of data; however, emphasizing the need for additional flux tower measurements in the tropics.
Number of site years and characteristics of CH4 fluxes (g C m–2 yr–1) currently included in the database. Fluxes are compared with measurements reported in a recent synthesis of predominantly chamber-based CH4 flux measurements. Biome type was extracted from Olson et al. (2001) using site coordinates and includes tundra, boreal/taiga, temperate, and tropical/subtropical. Wetland CH4 emissions differed significantly across biomes, with letters indicating significant differences (α = 0.05) among biomes. Note that similar to our tower only dataset, values from Treat et al. (2018b) represent measured annual fluxes derived from a smaller dataset where measurements were made in the growing season and nongrowing season.
Whereas annual boreal/taiga wetland CH4 emissions were comparable to values reported in a recent synthesis of predominantly chamber-based CH4 flux measurements (Treat et al. 2018b), our tower-based measurements are ∼50% lower and over 6 times higher for tundra and temperate wetlands, respectively (Table 1). The inconsistencies highlighted in Table 1 not only reflect the differences in the number and location of sites between datasets, but also the discrepancies resulting from different measurement techniques. Several studies have noted considerable differences in CH4 emissions measured using EC and chamber techniques, with estimates from chambers often higher than those from the EC measurements (Schrier-Uijl et al. 2010; Hendriks et al. 2010; Meijide et al. 2011; Krauss et al. 2016). This distinction highlights the need for additional studies investigating the systematic differences caused by the different spatial and temporal sampling footprints of these methods (Krauss et al. 2016; Morin et al. 2017; Windham-Myers et al. 2018; Xu et al. 2017). Characterizing discrepancies between measurement techniques may also help constrain bottom-up estimates of CH4 emissions and reduce the disagreement of ∼15 Tg C yr–1 between bottom-up (139 Tg CH4 yr–1) and top-down (125 Tg CH4 yr–1) estimates of CH4 emissions from natural wetlands (Saunois et al. 2016a).
Annual CH4 emissions also differed significantly across ecosystems [average Monte Carlo χ2 = 45.5 (39.3–50.1), df = 9, p < 0.001; Fig. 5], with median fluxes highest for freshwater marshes (43.2 ± 4.2 g C m–2 yr–1) and lowest for upland ecosystems (1.3 ± 0.7 g C m–2 yr–1). Treat et al. (2018b) also observed the highest annual emissions in marshes and reported a similar median value for temperate marshes (49.6 g C m–2 yr–1). Wet tundra and bogs had significantly lower annual emissions than marshes (Fig. 5), which in part reflects their presence in colder boreal and tundra systems, as well as differences in vegetation type, nutrient status, and hydrological regime (Treat et al. 2018b). Low median CH4 emission was observed from salt marshes in our dataset (0.8 ± 2.9 g C m–2 yr–1), because high sulfate concentrations inhibit methanogenesis (Poffenbarger et al. 2011; Holm et al. 2016). Even drained wetlands converted to agricultural land can be large sources of CH4 associated with seasonal flooding (Fig. 5). Median annual CH4 flux from rice was 12.6 ± 1.6 g C m–2 yr–1, which is slightly lower than the IPCC default value of 15 g C m–2 yr–1 (Sass 2003).
Annual CH4 fluxes (FCH4; g C m–2 yr–1) among ecosystem types for the 60 sites currently included in the database (Table A1). Letters indicate significant differences (α = 0.05) among ecosystem types. Median value, first quartile, and third quartile are pre