The 2017 version of the source signature database is accessed from the NOAA Earth Systems Research Laboratory with this link: https://www.esrl.noaa.gov/gmd/ccgg/arc/?id=123. This version supersedes an earlier version (Sherwood et al., 2016) published as a complement to Schwietzke et al. (2016). Whereas the previous version reported values of δ13CCH4 only, the 2017 version expands the range of geochemical parameters, as described in Sect. 2.4 below. Other minor changes to the database are noted in the database “Readme” file.

The database is separated into fossil fuel and non-fossil fuel sources of CH4. Fossil fuel sources comprise conventional natural gas, coal gas, and shale gas. Shale gas is included as a separate category because of growing interest in CH4 emissions associated with this form of unconventional gas production. Both conventional and shale gas include natural gas coproduced with oil. Coal gas includes both coal mine gases and coal bed methane. All three fossil fuel gas types are representative of reservoir gases measured from producing or previously producing oil or gas wells or coal mines. Data from exploratory wells were excluded, as these are not broadly representative of atmospheric emissions. The database does not currently distinguish between oil and non-oil-associated gas or between different ranks of coal (i.e., lignite, bituminous, and anthracite). However, the database includes the locations of each sample, which may be used to make this distinction based on activity data (e.g., production based on coal rank at a given coal mine). Pipeline (processed) distribution gases are not included in the database, primarily due to lack of data availability. Users of this database should be aware that, due to preferential stripping of alkane components, processed gases may have different molecular compositions than the reservoir gases represented herein. Also, the molecular composition of distribution gases in any region may change over time (Schwietzke et al., 2014b). In comparison with the intentional changes of the molecular composition of natural gas, isotopic signatures are thought to be relatively unaffected by gas processing except for mixing of two or more isotopic end members (Schoell et al., 1993). Geological seepage gases, i.e., the natural source component of the fossil fuel category (Etiope et al., 2008; Etiope, 2009, 2015), are not included in this database. A global database of onshore seeps is discussed in Etiope (2015) and available from CGG (2015). The composition of seepage gases and their influence on the global CH4 budget is the subject of ongoing research.

Non-fossil fuel sources of CH4 in the database consist of modern microbial sources and biomass burning. Modern microbial data are from rice paddies, ruminants (C3- and C4-plant eating cattle, sheep, goats, and their manure), termites, waste and/or landfills, and wetlands (bogs and/or peat, deltas, estuaries, floodplains, lagoons, lakes, marshes, ponds, rivers, swamps and tundra). Biomass burning data are from brush, forests and/or woodlands, grasses, and pastures.

Data were obtained from the peer-reviewed literature, conference proceedings, graduate theses, and government reports and databases. Government databases include the US Geological Survey (USGS) Energy Geochemistry Database (https://energy.usgs.gov/GeochemistryGeophysics/GeochemistryLaboratories/GeochemistryLaboratories-GeochemistryDatabase.aspx), the Geological Survey of the Netherlands (NLOG) database (available by request through http://nlog.nl/en/gas-properties), and the Geoscience Australia (ORGCHEM) database (available by request through http://www.ga.gov.au/search/index.html#/). Google Scholar, Web of Science, the American Association of Petroleum Geologists (AAPG; http://www.datapages.com/), and the Society of Petroleum Engineers (SPE; http://www.spe.org) were used to search for data. The use of English language search tools presented an unavoidable bias in data gathering. Searches focused on publications with gas isotopic data. Since gas compositional analysis is a prerequisite for subsequent isotopic analysis in most laboratories, gas compositional data are included with δ13CCH4 and δ2H data if reported in the original source. Note that the literature contains far more publications with gas compositional data alone. All of the data can be traced back to original sources using the references provided. To maintain data transparency, industry proprietary data were excluded.

The database is separated into fossil fuel and non-fossil fuel (modern microbial and biomass burning sources) data tables for two practical reasons. First, the petroleum geochemistry literature tends to report analyses for discrete samples, for example, production gas analyses from individual wellheads or analyses from discrete stratigraphic horizons in a wellbore. By contrast, the literature on non-fossil fuel sources of CH4 more commonly reports statistical summaries (e.g., multiple measurements at a given location and time) as opposed to discrete sample data; because of this, the non-fossil data comprise n=1973 measurements represented in 107 rows of data. Second, fossil fuel data usually include gas composition of C2+ alkanes and non-alkane gases and isotopic compositions of C2+ alkanes. The non-fossil fuel literature rarely reports data on these additional parameters, even though microbial processes in fact produce C2+, albeit in negligible quantities (< 0.1 %) compared to CH4 (Oremland, 1981; Ladygina et al., 2006; Xie et al., 2013). Rather than trying to fit these two fundamentally different types of data into a common table format, they are presented separately.

Table 2 lists analytical parameters included in the database. For fossil fuel gases, parameters include molar percent composition of non-alkane gases (N2, O2, CO2, Ar, H2, H2S, He) and C1 to C6 alkanes (CH4, C2H6, C3H8, iso-C4H10, n-C4H10, iso-C5H12, n-C5H12, C6H14) as well as δ13C and δ2H isotopic ratios of C1 to C5 alkanes. Though less commonly used in 3-D inverse modeling studies of the global CH4 budget, alkane compositions are important for source attribution in regional air quality and emissions studies (Karion et al., 2013; Pétron et al., 2014; Peischl et al., 2015; Kort et al., 2016). The δ13C and δ2H isotopic signatures of C2+ alkanes may also prove useful as source tracers with future advances in analytical instrumentation. For non-fossil fuel samples, δ13C and δ2H of CH4 are the only parameters provided in the database.

Stable isotopic data are reported in conventional delta notation: δX=(Rsample/Rstandard-1)×1000, in which δX=δ13C or δ2H and R=13C / 12C or 2H / 1H, respectively. δ13C data are reported on the Pee Dee Belemnite/Vienna Pee Dee Belemnite (PDB/VPDB) scale and δ2H data are reported on the Vienna Standard Mean Ocean Water (VSMOW) scale. The Vienna version of the PDB scale, signifying that the original PDB reference material used to define the scale ran out and was replaced with the NBS-19 reference material, is nominally identical to the previous PDB scale (Gröning, 2004). For references in which the scales were not stated explicitly, we assume the use of the PDB/VPDB and VSMOW scales, based on the fact that the use of PDB to define the δ13C scale and VSMOW to define the δ2H scale goes back to the 1950s and early 1960s (Craig, 1953, 1961) and that the oldest reference in the database (Dubrova and Nesmelova, 1968) postdates formal recognition of these scales.

It should be noted that stable isotope laboratories calibrate their data against working and/or secondary standards that have been tied to the PDB/VPDB and VSMOW scales (e.g., Dai et al., 2012). The NG-1, NG-2, and NG-3 suite of natural gas isotopic standards served this purpose beginning around the year 1984 but have since been exhausted (Hut, 1987). The current lack of International Atomic Energy Agency (IAEA) or National Institute of Standards and Technology (NIST) isotopic standards for natural gas or methane remains an ongoing problem. Unfortunately, this level of analytical detail often goes unreported in the database references.

Data screening for the fossil fuel data consisted of the following steps. (1) Location of metadata (country, state or region, basin, formation) were checked for logical compatibility. (2) To aid searching for basin-specific data, wherever possible fossil fuel data were assigned to a corresponding sedimentary basin in the Robertson Tellus Sedimentary Basins of the World (available at http://www.datapages.com/gis-map-publishing-program/gis-open-files/global-framework/robertson-tellus-sedimentary-basins-of-the-world-map). (3) Data duplicates were merged. This step was particularly important for the USGS Energy Geochemistry Database as it includes data from several other sources including the Gas Research Institute report on US natural gas analyses (Jenden and Kaplan, 1989), peer-reviewed papers, and other USGS data reports. For merged duplicates, references to both sources are provided. (4) Obvious outliers, such as individual gas concentrations greater than 100 %, O2 concentrations greater than 21 %, total gas compositions summing to greater than 100 % (plus 10 % to allow for analytical and rounding errors), and positive values of δ13C and δ2H were omitted. For the non-fossil fuel data, no data-screening steps were taken; data are provided as originally reported in the respective sources.

This database was not subject to a data quality assessment. The data were generated from countless laboratories in different countries over a span of 5 decades. Source publications also span a wide range of academic rigor, from conference proceedings to peer-reviewed journals. Milkov (2010) analyzed natural gas data from the West Siberia Basin and found that Soviet-era papers from the 1970s reported δ13CCH4 values that were too negative by ∼ 7 ‰ compared to data generated in the late 1990s by US, German, and Russian labs, while Soviet-era papers from the 1980s reported values that were too positive by ∼ 4.5 ‰. We make no attempts to correct for these systematic errors; rather we caution users of this database to evaluate and use the data appropriately. By sheer number of samples (n=10706) and data sources, systematic errors inherent in any single dataset average out over the whole database, while random errors have a negligible impact on measures of central tendency.

Fossil fuel sources comprise 8734 data records from 149 published sources. Table 3 provides a summary of the number of countries, basins, fields, formations, and published source by gas type (conventional gas, coal gas, shale gas) and specified analytical parameter (δ13CCH4, δ2HCH4, C2H6 : CH4). Non-fossil fuel sources comprise 1972 data records from 41 published sources. Table 4 provides a summary of the number of countries, regions, and published sources by CH4 source (rice paddies, ruminants, termites, waste, wetlands, biomass burning) and parameter (δ13CCH4, δ2HCH4). Finally, Table 5 provides unweighted statistical summaries by CH4 source and parameter.

Figure 1 shows global maps of the number of samples in each country by gas type. These maps are further broken down by parameter (δ13CCH4, δ2HCH4, C2H6 : CH4) in Fig. 2. Representativeness of the fossil fuel data is assessed by comparison of sample counts from each country to that country's coal and natural gas production volumes from the BP Statistical Review of World Energy, 2016 (http://www.bp.com/en/global/corporate/energy-economics/statistical-review-of-world-energy.html) (Fig. 3). This was done at the level of individual countries owing to difficulty in obtaining production statistics at the sub-country level for all the countries in the database. We note that reservoir gases vary compositionally and isotopically within individual countries, basins, fields, and formations (Fig. 4). Within an individual formation, for example, natural gas can range from microbial gas in shallow and/or thermally immature areas, to oil-associated gas in deeper or thermally mature areas, to unassociated dry gas in thermally over-mature areas. Similarly, the type (i.e., rank) of coal gas data presented for any specific country may not be representative of the dominant coal type produced in that country.

Despite isotopic and compositional variability within countries, country-level analysis is the finest practical spatial resolution that can be assessed for the global dataset. Shale gas was excluded from this analysis of representativeness since shale gas production is limited mostly to Canada and the US. For the parameter δ13CCH4, the database is representative of 84 % of global natural gas production and 80 % of global coal production for the time period 2000–2015. For conventional gas, the countries with the highest numbers of samples with δ13CCH4 are the US (n=2042), China (834), Russia (556), Canada (402), and Australia (400) (Fig. 3). Countries with no conventional gas data include Algeria, Malaysia, Turkmenistan, the United Arab Emirates, and Venezuela, which together account for 12.2 % of global natural gas production. For coal gas, the countries with the largest sample sizes include the US (722), China (196), Australia (110), and Poland (105) (Fig. 3). Countries with no coal gas data representation include India, Indonesia, Kazakhstan, Ukraine, and Colombia, which together account for 14.5 % of global coal production. For the parameter δ2HCH4, the database is representative of 73 % of global natural gas production and 74 % of global coal production. For C2H6 : CH4 ratio data, the database is representative of 76 % of global natural gas production and 31 % of global coal production. Sample biases can be mitigated by weighting values by each country's fraction of global gas or coal production (Schwietzke et al., 2016) or by other methods suited to the specific data use.

Representativeness is generally poorer for the non-fossil data, owing in part to the smaller total sample sizes and the lack of data for several key areas. For example, there are few microbial or biomass burning data from Southeast Asia and Africa, two areas of significant wetland, termite, and biomass burning CH4 emissions. Arctic wetlands are also underrepresented in the database. These areas constitute important data gaps that should be targeted for more intensive data mining and/or future field studies.

Figure 5 shows a natural gas genetic characterization plot of δ13CCH4 versus δ2HCH4, first presented in Whiticar et al. (1986) and modified in Whiticar (1989, 1999). The characterization framework in Fig. 5 and in other plots of δ13CCH4 versus alkane molecular compositions (Bernard, 1978; Schoell, 1983; Faber and Stahl, 1984) were originally developed by researchers at the German Federal Institute for Geosciences and Natural Resources in the 1970s and early 1980s. These plots were derived largely from proprietary industry data. Because the data could not be publicized, the characterization plots were published without showing the underlying data used in their development. These characterization schemes are still widely used to this day, despite that fact that the literature data on gas isotope ratios and compositions has expanded by orders of magnitude since the 1980s. Figure 5 shows the distribution of conventional gas, coal gas, and shale gas in relation the major genetic fields: thermogenic, microbial CO2 reduction, and microbial fermentation. It also shows the field for gases from geothermal, hydrothermal, and crystalline rocks. Overall, the low percentage of samples falling outside any of the principal genetic fields in Fig. 5 indicates that this original classification scheme captures essentially the full range of isotopic variability in natural gases; however, the breakdown of sample counts by genetic origin changes with revision to the classic characterization scheme. For example, while the canonical thermogenic field assumes a δ13C value of -50 or -55 ‰ as the limit between thermogenic and microbial CH4 (Stahl, 1974; Schoell, 1983; Whiticar et al., 1986), recent work extends the thermogenic field to isotopically lighter values; see below.

Figure 6 shows a more recent version of the δ13CCH4 versus δ2HCH4 plot, updated in Etiope (2015) based on a previous, unpublished version of a fossil fuel reservoir dataset. This diagram distinguishes more types of thermogenic gas, following Etiope and Sherwood Lollar (2013) and Hunt (1996) and reports an updated genetic field for abiotic gas, i.e., gas formed by chemical reactions of inorganically derived gases such as carbon dioxide (CO2) and hydrogen (H2) and not from degradation of organic matter (Etiope and Schoell, 2014). The thermogenic field in Fig. 6 extends to δ13C = -67 ‰ due to the existence of low-maturity thermogenic gas (Rowe and Muehlenbachs, 1999; Milkov and Dzou, 2007) and secondary alterations (biodegradation; Milkov, 2010, 2011) that would otherwise be mistaken for primary microbial gas.

Of the 8734 fossil fuel samples in the database, a subset of n=2861 have both δ13C and δ2H data and are thus represented on the plot. For conventional gas (n=1951 δ13C - δ2H data pairs), a majority (78 %) of the samples plot within the thermogenic field. A smaller percentage of samples plot within the microbial field (17 %) or the abiotic field (5 %). For coal gas (n=511), data are more evenly distributed between thermogenic (56 %) and microbial (39 %) fields, with a smaller percentage falling within the abiotic (2 %) field. Because of overlapping genetic fields, percentages sum to > 100 %. Additionally, it is important to outline that conventional or coal gases falling within the abiotic field actually have a dominant thermogenic origin: these δ13C-enriched gases are, in fact, mainly from over-mature (late-stage catagenesis) source rocks from northwestern Germany (Rotliegend) and China (Songliao and Tarim basins). Further refinement of the genetic characterization plot should therefore account for these late-stage thermogenic gases. Shale gas data (n=396) fall almost entirely within the thermogenic field (91 %), with the majority of the data clustered toward the dry gas (TD in Fig. 6) end of the thermogenic maturity spectrum. Non-fossil source data (rice paddies, ruminants, waste, wetlands, termites) plot entirely within the microbial fermentation field. Biomass burning has a characteristically enriched isotopic signature, falling within the abiotic field despite a fundamentally different generation pathway compared to abiotic natural gas. A revision of the genetic diagram is in fact in progress (Milkov et al., 2017), and statistics of our database will be readjusted, taking into account this new reassessment of microbial versus thermogenic isotopic genetic characterization.

A long-standing view in the petroleum geochemical literature held that “more than 20 % of the world's discovered gas reserves are of biogenic origin” (Rice and Claypool, 1981). This biogenic gas was loosely defined by cutoffs of δ13CCH4 < -55 ‰ and < 2 % C2+ alkanes (C2H6 through pentane (C5H12)). For conventional natural gas in the current database, 14 % of the samples have δ13CCH4 < -55 ‰ and 23 % have a C2H6 : CH4 ratio < 0.02. These percentages envelope the original Rice and Claypool (1981) estimate. However, it is now known that natural gas within the δ13CCH4 and % C2+ cutoffs encompass primary microbial gas (i.e., biogenic gas in Rice and Claypool, 1981; formed from microbial CO2 reduction and methyl fermentation in shallow sediments), secondary microbial gas (formed from biodegradation of thermogenic hydrocarbons; Zengler et al., 1999; Head et al., 2003; Jones et al., 2008), and low-maturity thermogenic gas (Rowe and Muehlenbachs, 1999; Milkov and Dzou, 2007). Analysis of the δ13C and molecular ratios of C2+ alkanes and CO2 is often the only means of distinguishing between these three types of gas (Milkov, 2011).

At the global level, primary and secondary microbial gases are thought to account for ∼ 3–4 % and ∼ 5–11 %, respectively, of conventional recoverable natural gas reserves (Milkov, 2011). Secondary microbial gas accounts for a larger share of global conventional gas production: giant Cenomanian gas pools of secondary microbial CH4 (mean δ13CCH4=-51.8 ‰) found at depths < 1500 m in the West Siberia Basin alone account for ∼ 17 % of the global conventional gas production (Milkov, 2010).

Microbial methanogenesis is even more significant for coals (Rice, 1993), with an approximately even distribution between thermogenic and microbial genetic origins (Figs. 5, 6). The two largest coal mines in the world (North Antelope Rochelle and Black Thunder mines) are located in the Powder River Basin, Wyoming, US. Coal gas from these formations is microbial (fermentation) in origin (mean δ13CCH4=-59.1 ‰, n=267; mean δ2HCH4=-309.6 ‰, n=118). However, as discussed above, we note that some gas, traditionally considered microbial because of its low δ13C values, may actually have a thermogenic origin. Coals can also generate secondary microbial gas (Scott et al., 1994).

Figures 7 and 8 show normalized probability distributions of δ13CCH4 and δ2HCH4 for fossil fuel and modern microbial processes (with their respective subcategories) and biomass burning sources of CH4. The distributions show wide overlap between different CH4 source categories, thus highlighting the critical need for robust weighting schemes that result in globally or regionally representative measures of central tendency (discussed below).

Data distributions for modern microbial processes have relatively normal distributions with tight overlap between the different subcategories. The distributions for biomass burning show characteristic bimodality, caused by differences between isotopically lighter C3 and isotopically heavier C4 vegetation. Fossil fuel δ13C and δ2H exhibit left-skewed (conventional and shale gas) or bimodal (coal) distributions arising from the presence of microbial and low-maturity thermogenic gas as described above. This also leads to relatively wider data ranges than the non-fossil categories.

Figure 7 also indicates the δ13C of atmospheric CH4 (∼ -53.6 ‰) before fractionation by photodegradation, calculated as measured atmospheric δ13CCH4 (mean -47.3 ‰ in the year 2016; White et al., 2017) plus an average fractionation factor ε=-6.3±0.8 ‰ (Schwietzke et al., 2016). The δ13C of the atmosphere before fractionation represents the “hinge point” upon which CH4 emissions fluxes are estimated by isotopic mass balance (e.g., Whiticar and Schaefer, 2007). Modern microbial processes have δ13CCH4 signatures falling to the left of the hinge point; thus, lower δ13CCH4 requires lower emissions to isotopically balance fossil fuel and biomass burning sources; higher δ13CCH4 requires higher emissions. Conversely, fossil fuel and biomass burning source categories have δ13CCH4 signatures falling to the right of the hinge point, thus lower δ13CCH4 requires higher emissions; higher δ13CCH4 requires lower emissions. Biomass burning falls furthest from the hinge point (mean δ13CCH4=-26.2±4.8 ‰, unweighted by proportion of C3 and C4 vegetation). Therefore, it has the most leverage on the isotopic mass balance.

In Fig. 8 the pre-fractionation hinge point is more poorly constrained, owing to greater uncertainty in measured atmospheric δ2HCH4 (-95 ± 5 ‰) and, more importantly, uncertainty in the estimated fractionation factor ε=-235±80 ‰ (Gierczak et al., 1997). Modern microbial δ2HCH4 signatures are within the range of the estimated pre-fractionation atmosphere. Biomass burning and fossil fuel signatures fall to the right of the hinge point. Hence, lower δ2HCH4 requires higher emissions and higher δ2HCH4 requires lower emissions for both these categories.

Unweighted mean δ13CCH4 and δ2HCH4 for modern microbial processes and biomass burning (Table 5) are generally within about 2 ‰ of typical values used in published CH4 budget models (Schwietzke et al., 2016). By contrast, fossil fuel δ13CCH4 and δ2HCH4 summary statistics (Table 5) show wider disparity with source signature values used in published CH4 budget models (Table 1). Remarkably, unweighted mean δ13CCH4 for conventional natural gas (-44.0 ± 10.7 ‰) is identical to the value (-44 ‰) originally indicated by Craig et al. (1988), Stevens and Engelkemier (1988; thermogenic gas), Quay et al. (1991; oil-associated gas), and Whiticar (1989, 1993). However, this value is 4–6 ‰ lighter than the range of -38 to -40 ‰ typically used in more recently published models (Gupta et al., 1996; Lassey et al., 2000, 2007; Tyler et al., 2007; Neef et al., 2010; Monteil et al., 2011; Rigby et al., 2012; Ghosh et al., 2015; Schaefer et al., 2016). For coal gas, unweighted mean δ13CCH4 (-49.5 ‰) is even more significantly depleted compared to typical values of -35 to -37 ‰ assumed in virtually all previous studies. These canonical values were likely derived from bituminous and anthracite coal, which is isotopically heavier than lignite and subbituminous coal (Rice, 1993; Zazzeri et al., 2016), yet lignite and subbituminous coal account for more than half of world coal production (World Energy Council, 2013). Similarly, mean δ2HCH4 for natural gas (-194 ‰) and coal (-232 ‰) is 10–15 ‰ and 60–120 ‰, respectively, lighter than literature values. Shale gas also exhibits lower mean δ13CCH4 (-42.5 ‰) and δ2HCH4 (-167 ‰) than indicated in the CH4 budget literature. For all fossil fuel data (un-categorized), unweighted mean δ13CCH4=-44.8 ± 11 ‰ (1 SD) and δ2HCH4=-197±50 ‰.

These results highlight the possibility that widespread use of too-heavy δ13CCH4 and δ2HCH4 fossil fuel source signatures could have led to systematic underestimation of fossil fuel emissions in the CH4 budget literature. Indeed, Schwietzke et al. (2016) reanalyzed the global CH4 budget using weighted source signature data calculated from an earlier version of this database (Sherwood et al., 2016) and showed that total fossil fuel emissions (excluding geological seepage) are about 50 % higher than previously estimated.

Database users are encouraged to adopt appropriate weighting criteria for estimating spatially averaged source signatures. For instance, at the global level, Schwietzke et al. (2016) developed a method to weight fossil fuel δ13CCH4 data at the country level and non-fossil fuel δ13CCH4 data at the emissions subcategory level. Weighting fossil fuel δ13CCH4 data at the basin level may be practical for some countries with a sufficient sample size. Basin-level gas production statistics may be used in the weighting procedure as a proxy for basin-level CH4 emissions. However, note that basin-level CH4 emissions may be correlated with basin-level δ13CCH4. A basin with mature dry gas and no associated oil production (and thus relatively heavy δ13CCH4) typically employs less gas processing infrastructure (gas separators, combustors, storage tanks) than a basin with associated gas production (and thus relatively light δ13CCH4). The former is therefore likely to emit less CH4 per unit of gas production than the latter. This is substantiated by CH4 emissions estimates from multiple US oil and gas basins. For example, the dry gas basins Marcellus Shale and Fayetteville are estimated to emit on average 0.3 and 1.9 %, respectively, per unit of gas produced (Peischl et al., 2015), whereas the wet gas Denver and Uinta basins emit on average 4.1 and 8.9 %, respectively, per unit of gas produced (Karion et al., 2013; Pétron et al., 2014). Thus, using gas production statistics to weight individual basins without knowledge of the respective CH4 emissions may introduce biases.