Evaluating ethane and methane emissions associated with the development of oil and natural gas extraction in North America

High-resolution infrared solar absorption spectra encompassing C₂H₆ absorption features near 3.5 μm are regularly recorded under clear-sky conditions at globally distributed ground-based sites within the Network for the Detection of Atmospheric Composition Change (NDACC; see http://ndacc.org and www2.acom.ucar.edu/irwg). The present study focuses on observations obtained as part of this monitoring effort in North America, namely at Eureka, Thule and Toronto, and at Mauna Loa. Spectra recorded at Boulder, in operation since 2009 and not affiliated yet to NDACC, have also been included to increase the latitudinal coverage. Table 1 provides information on these sites. To check the ability of lower-resolution, compact and portable instruments to provide meaningful column measurements of C₂H₆, we further considered observations performed by the portable atmospheric research interferometric spectrometer for the infrared (PARIS-IR; Fu et al 2007), an instrument based on the design of the atmospheric chemistry experiment-Fourier transform spectrometer (ACE-FTS) in orbit on-board the SCISAT satellite since August 2003 (Bernath et al 2005). PARIS-IR has been operated from Eureka each spring since 2006 and from Toronto in 2009 and 2011, side-by-side with the corresponding affiliated NDACC-FTIR spectrometers.

The C₂H₆ retrieval strategies used here are consistent with those described in Franco et al (2015). In brief, two spectral windows encompassing the ^pQ₃ and ^pQ₁ sub-branches near 2976 and 2983 cm⁻¹ were simultaneously fitted, accounting for interferences by water vapor (H₂O), O₃, methyl chloride (CH₃Cl) and CH₄. The third optional window near 2986 cm⁻¹, sometimes affected by water vapor, was used at all sites. The adopted spectroscopic parameters correspond to a combination of HITRAN 2008 for most species (Rothman et al 2009), with HITRAN 2012 for CH₃Cl (Rothman et al 2013), improved parameters for the O₃ lines near 2976.97 cm⁻¹ and a pseudo-linelist for C₂H₆ derived from laboratory cross section spectra recorded by Harrison et al (2010). Model simulations with the CHemical Atmospheric General Circulation Model for study of atmospheric environment and radiative forcing (CHASER; Sudo et al 2002) were used to build the a priori covariance matrix and to determine on a site-by-site basis, a scale factor applied to the v6 climatological a priori vertical profiles employed in the NDACC Infrared Working Group that were derived from the Whole Atmosphere Community Climate Model (WACCM; Garcia et al 2007, Eyring et al 2013). The objective characterization of the information content indicated that mean degrees of freedom for signal (DOFS) are in the range of 1.5–1.7 at all sites, providing sensitivity in an altitude range from the ground up to ∼18 km. The retrieval algorithm and approach adopted for the PARIS-IR low-resolution (0.02 cm⁻¹) observations are consistent with those used for the high-resolution spectra and provide a mean DOFS of approximately 1.0 between the ground and ∼18 km.

Uncertainties associated with the retrieval of C₂H₆ have been evaluated in numerous previous studies (see Franco et al 2015 and references therein), for many different sites with a broad range of atmospheric conditions. All concur in estimating a total systematic error on the total columns close to 6%, dominated by the uncertainty affecting the C₂H₆ line parameters. Reported random errors are in the 2%–4% range for studies making simultaneous use of several windows, with main contribution by the measurement noise.

For this study, a simplified version of CAM-chem (Lamarque et al 2012), a component of the Community Earth System Model, was used to simulate multi-year distributions of C₂H₆ along with tracers of source region contributions. CAM version 4 (CAM4) with bulk aerosols was used at a 1.9° latitude × 2.5° longitude resolution, with the global OH distribution specified using output from a previous full chemistry simulation for year 2000 (Tilmes et al 2015). The meteorology is specified using Modern-Era Retrospective Analysis for Research and Applications meteorological fields corresponding to each year of simulation. The reaction of C₂H₆ with OH was the only chemical loss included. Since this version of the model does not include complete chemistry, it will be called CAM-C₂H₆ throughout this paper.

The initial anthropogenic C₂H₆ emissions inventory used with CAM-C₂H₆ was developed for the Hemispheric Transport of Air Pollutants, Phase II (HTAP2), and is a composite of regional inventories harmonized to represent 2008 and 2010 (Janssens-Maenhout et al 2015). Additional C₂H₆ emissions included in these simulations are biogenic emissions from the online Model of Emissions of Gases and Aerosols from Nature (MEGAN2.1) biogenic emissions model (Guenther et al 2012), and fire emissions from the Fire INventory from NCAR (FINNv1.5; Wiedinmyer et al 2011).

Daily mean C₂H₆ mixing ratio profiles at the closest pixel to each measurement site were used from the CAM-C₂H₆ output. To ensure a reliable comparison with the ground-based measurements, we imposed the vertical resolution and sensitivity of the FTIR retrievals to the model outputs before computing the total columns: the individual mixing ratio profiles were re-gridded onto the vertical FTIR layer schemes and convolved by the averaging kernels (AVKs) according to the formalism of Rodgers and Connor (2003). The re-gridding method is a mass-conservative interpolation, preserving the total C₂H₆ mass above the station altitude and ignoring the mass underneath. Seasonally-averaged AVKs were employed for the smoothing.

GEOS-Chem is a chemical transport model driven by assimilated meteorological fields from the Goddard Earth Observing System of the National Aeronautics and Space Administration (NASA) Global Modelling and Assimilation Office (Bey et al 2001). Our analysis was based on a 2° × 2.5° resolution simulation for 2010, with an 18 month spin-up. We used v10-01 with GEOS-5 meteorological fields and the full chemistry (NO_x–O_x–HC–Aer–Br) mechanism. Ethane emission fluxes over the US and Canada were based on CH₄ fluxes derived from the Greenhouse Gases Observing SATellite (GOSAT) at a 50 km × 50 km spatial resolution (Turner et al 2015). The CH₄ inversions were calculated using a state vector optimally defined with radial basis functions to enable analytical inversion with full error characterization while minimizing aggregation error. A full description of the uncertainty and error characterization of this method for calculating CH₄ emissions is presented in Turner and Jacob (2015).

In order to attribute nationwide percentages of anthropogenic CH₄ emissions, Turner et al (2015) used prior information on source locations. As a result, they estimated CH₄ fluxes for 12 anthropogenic categories. In this study, we only considered the categories and thus locations relevant to C₂H₆, and these include: (1) oil and NG, (2) biofuel, and (3) biomass burning. For these categories, we estimated C₂H₆ emissions using C₂H₆/CH₄ emission ratios from the literature and implemented them into the model using the Harvard-NASA Emissions Component v1.1.005 (Keller et al 2014). To estimate oil and gas C₂H₆ emission fluxes, we used a single C₂H₆/CH₄ emission ratio based on Katzenstein et al (2003). To derive biomass burning and biofuel C₂H₆ emissions, we considered biofuel consumption as residential biomass burning; and we used the C₂H₆/CH₄ emission ratios for temperate forests from Akagi et al (2011). The use of relatively few emission ratios is a large source of uncertainty in this analysis on top of errors associated with the GOSAT based CH₄ fluxes. However, there are relatively few emission ratios available in the published literature, and our approach offers a simple and reproducible implementation in the model. Consistent with the CAM-C₂H₆ outputs, we used the GEOS-Chem mixing ratio profiles accounting for the FTIR vertical resolution and sensitivity.

We investigated C₂H₆ time series derived from the ground-based observations to determine the current abundances of C₂H₆ as well as its recent evolution over the North American FTIR sites. The FTIR daily mean total columns from January 2003 through December 2014 are shown as gray circles in figure 2, while the light blue triangles correspond to the daily-averaged PARIS-IR measurements at Eureka and Toronto. On each frame, the right y-axis scale converts the total columns into approximate surface mole fractions. The scale conversion was determined for each site according to the linear regression between the total column abundance and the C₂H₆ mole fraction in the lowermost layer of the FTIR retrieval scheme. These regressions are characterized by a coefficient of determination (R²) of at least 0.8.

Zoom In Zoom Out Reset image size

Differences in C₂H₆ abundance present in figure 2 are largely a function of site latitude (Simpson et al 2012) and altitude (Gardiner et al 2008, Angelbratt et al 2011); the bulk of the C₂H₆ molecules are found in the lowermost tropospheric layers. The mean column for each site (±1σ) is as follows: 1.7 ± 0.5 (Eureka; 610 m a.s.l.), 1.8 ± 0.6 (Thule; 225 m a.s.l.), 2.1 ± 0.4 (Toronto; 174 m a.s.l.), 1.6 ± 0.5 (Boulder; 1612 m a.s.l.) and 0.6 ± 0.2 molec cm⁻² (Mauna Loa; 3396 m a.s.l.). There are also seasonal and interannual differences in the C₂H₆ burden between the sites. Mainly driven by the rate of reaction with OH, the typical seasonal peak-to-peak amplitude for C₂H₆ is on the order of 50% at Boulder and Toronto, and it increases to ∼80% at the remote sites. It is worth noting the good agreement between the PARIS-IR and FTIR total columns above both Eureka and Toronto (figures 2(a) and (c)).

We characterize the recent atmospheric evolution of C₂H₆, considering the daily means, with a statistical bootstrap resampling tool that combines a linear function and a Fourier series accounting for the intra-annual variability of the data sets (Gardiner et al 2008). A third-order Fourier series is used to adjust the seasonal variations, except at the Arctic sites, Eureka and Thule, where a second-order fit is appropriate to capture the seasonality over the spring to fall time period (no observations during the polar night). As in Franco et al (2015), we consider separately the daily mean subsets before 1 January 2009 (when enough observations are available), and after 1 January 2009. The fitted seasonality is drawn as a blue curve in each frame of figure 2, while the green and red solid lines designate the corresponding linear regressions. At Eureka (figure 2(a)), a linear regression (red dashed line) is calculated solely on the basis of the 2009–2014 PARIS-IR campaign measurements. Figure 3 shows the relative annual C₂H₆ rates of change, using either the January 2003 or January 2009 computed total columns as reference. The available annual rates of C₂H₆ over both the 2003–2008 and 2009–2014 time periods, along with the 2σ confidence level, for the five FTIR sites are plotted in blue. The equivalent annual rates derived from the Jungfraujoch (table 1) time series investigated in Franco et al (2015) are added for comparison.

Zoom In Zoom Out Reset image size

From 2003 to 2008, we observe a slow decline of the C₂H₆ total columns between −1.0 and −1.5% yr⁻¹, with consistent rates within the different latitudes (figure 3(a)). The Mauna Loa rate of change is not significant at the 2σ confidence level (0.9 ± 1.1% yr⁻¹). However, when using the dataset from inception, we compute a decrease of −1.0 ± 0.3% yr⁻¹ over 1995–2008. These results agree with the global decline of the atmospheric C₂H₆ burden from the mid-1980s up to the end of the 2000s, which was widely reported in the scientific literature (e.g., Gardiner et al 2008, Angelbratt et al 2011, Franco et al 2015, and references therein). It was attributed to reduced fugitive fossil fuel emissions that were thought to also contribute significantly to the slowing growth rate of atmospheric CH₄ since the mid-1980s (Aydin et al 2011, Simpson et al 2012).

However, as at Jungfraujoch (Franco et al 2015), the FTIR measurements over the North American sites show a reversal of the long term decline in C₂H₆ around 2009, followed by a sharp and significant (at the 2σ confidence level) increase at all FTIR sites (figure 3(b)). The observed rates of change are very consistent and vary between ∼3% yr⁻¹ at the remote stations and ∼5% yr⁻¹ at mid-latitudes. The increase in C₂H₆ deduced from the PARIS-IR and FTIR observations at Eureka are very consistent as well. We hypothesize that this recent increase is the result of the recent massive growth in the exploitation of shale gas and tight oil reservoirs in North America, where the drilling productivity began to grow rapidly after 2009 (Schneising et al 2014, Warneke et al 2014, Vinciguerra et al 2015). Of the observation sites presented here, Boulder is located closest to a highly productive oil and gas basin, the Denver-Julesberg Basin. We suggest that is why the Boulder data show a large variability in the C₂H₆ columns (5.0 ± 4.5% yr⁻¹), influenced by local air mass circulation. Similar elevated mixing ratios of C₂H₆ have been observed at the Boulder Atmospheric Observatory by Gilman et al (2013).

We compared C₂H₆ simulations produced with CAM-C₂H₆ for the period 2003–2014, using the bottom-up anthropogenic inventory HTAP2, with the remote-sensing measurements. The C₂H₆ total columns derived from this simulation, represented by the dashed curve in figure 2, greatly underestimate the FTIR observations at all sites in terms of C₂H₆ abundance and seasonal amplitude, with a low bias that is far larger than the typical systematic uncertainty of 6% affecting the total columns. Furthermore, the model does not reproduce the recent C₂H₆ increase.

Based on comparisons between the model results and the FTIR measurements, we estimate that the global HTAP2 anthropogenic C₂H₆ emissions must be doubled to simulate the observed C₂H₆ abundance prior to 2009. We refer to this quantity of C₂H₆ emissions as HTAP2x2. An additional increase in C₂H₆ emissions beyond HTAP2x2 is required to simulate the recent rise from 2009 to 2014. Given that C₂H₆ emissions related to the oil and gas sector represent up to 80% of the total anthropogenic C₂H₆ emissions over North America, and if we assume that the missing emissions during this period resulted from the recent increase in oil and gas extraction in North America, we can use the model to estimate the increase in C₂H₆ emissions needed to match the observations. To match the 2014 observations at the continental sites, the North American annual contribution has to almost double over the period 2009–2014. This corresponds to a 1.2 Tg increase of the North American annual C₂H₆ emissions in 2014 over the 2008 emission rates. An overview of the total C₂H₆ emission rates used for the CAM-C₂H₆ simulations is provided in table 2.

The solid orange line in figure 2 signifies the C₂H₆ total columns simulated by CAM-C₂H₆ with the revised HTAP2x2 inventory and North American emission increases from 2009 onwards. Similar to the FTIR time series, the annual rates of change relative to January 2009 derived from this new CAM-C₂H₆ simulation are calculated and reported in figure 3. The results show a better agreement with the remote-sensing time series for both the abundance of C₂H₆ and its seasonal amplitude, especially at the northern sites. The mean daily biases of the model to the FTIR fitted seasonality over the period 2009–2014 are as follows: −4 ± 16% (Eureka), −4 ± 14% (Thule), −18 ± 13% (Toronto), −15 ± 15% (Boulder) and +37 ± 18% (Mauna Loa). Globally, the revised HTAP2x2 inventory allows the model to simulate significant C₂H₆ increases between ∼2.5 and ∼4% yr⁻¹ from 2009 onwards, although the model still slightly underestimates the rates of change observed at mid-latitudes. Similar findings are deduced from the comparison between CAM-C₂H₆ and FTIR at Jungfraujoch. These results support the hypothesis that the enhanced North American oil and gas extraction largely contributes to the recent C₂H₆ increase detected, as assumed previously by Franco et al (2015). For the remote non-continental Mauna Loa site, the CAM-C₂H₆ results overestimate the FTIR data (figure 2(e)), suggesting that the emissions in Asia (more directly upwind of Mauna Loa) should not be increased as much as 2 times the original HTAP2 inventory.

In this section, we compare the results of GEOS-Chem simulations performed with a new top-down inventory of anthropogenic C₂H₆ emissions, based on CH₄ measurements from GOSAT, to the FTIR observations. Since this new inventory is only for 2010, we compare the simulated C₂H₆ columns to the FTIR measurements for this year only (figure 4). In order to provide more robust statistics for the representation of the C₂H₆ seasonal cycle at each site, the available daily mean columns from the 2003–2014 FTIR time series are de-trended and scaled to the year 2010 according to the annual rates of change reported in figure 3. In figure 4, the FTIR monthly means are represented as blue circles along with 1σ standard deviation as error bars, while the blue curve and the shaded area correspond to the running average of the daily mean total columns (with a 2 month wide integration and a 15 day time step) and the associated 1σ standard deviation, respectively. Similarly, the red curve and shaded area designate the GEOS-Chem results. The total C₂H₆ emissions provided from the top-down inventory for the year 2010 are reported in table 2.

Zoom In Zoom Out Reset image size

GEOS-Chem properly simulates the late winter maximum abundance of C₂H₆ (figure 4) and reproduces the FTIR total columns found at the two mid-latitude sites, Toronto and Boulder (figures 4(c) and (d)). Both stations are located in the vicinity of intense oil and NG basins: several studies have shown a volatile organic compound source signature associated with the oil and gas industry near Boulder (Pétron et al 2012, Gilman et al 2013, Swarthout et al 2013, Thompson et al 2014), while Toronto is located north of one of the most active basins in the US, the Marcellus shale gas play. Due to its relatively coarse horizontal resolution, the model dilutes local enhancements in C₂H₆ and hence does not simulate the secondary maximum observed at Boulder during summertime (which is likely due to local air mass circulations).

GEOS-Chem shows a high-bias for the summertime C₂H₆ total columns at most sites, especially the remote sites (Eureka, Thule and Mauna Loa). The levels of OH, the main tropospheric sink of C₂H₆, are globally lower by ∼10%–15% in GEOS-Chem v10-01 with respect to previous model versions. This leads to less C₂H₆ oxidation at the time of OH maximum and to an underestimation of the seasonal cycle. This is particularly relevant for the polar sites, where the lower insolation dampens further the photochemical oxidation, causing C₂H₆ to accumulate in summer. We cannot rule out a contribution from transport; however, direct transport of pollutants into the arctic is less frequent during summer (Klonecki et al 2003).

The assumptions with regard to the C₂H₆/CH₄ ratios could also induce uncertainties in the inferred C₂H₆ emissions and could contribute to the biases observed between GEOS-Chem and some FTIR results. Indeed, such ratios largely vary in space and time, e.g., according to the types of facilities and the composition of the NG (Mitchel et al 2015, Subramanian et al 2015).