An array of instruments for the measurement of trace gases, aerosols, meteorological and aircraft state parameters were installed aboard the National Research Council of Canada Convair 580 research aircraft. Measurements of CH4, CO2, CO and H2O were made using a cavity ring-down spectrometer (Picarro G2401-m) at an interpolated rate of approximately 0.5 Hz with a flow rate of ∼ 435 sccmmin-1. The precision of the CH4 measurement was 2 ppb, and the uncertainty of the measurement at background (∼ 1.9 ppm) was 3.3 ppb (at 2 s). The instrument was calibrated six times before, during and after the project using two standard reference gases traceable to NOAA GMD standards. Methane mixing ratios are reported throughout as dry mole fractions in the paper. Necessary parameters for emissions estimation included temperature (T) measured a using Rosemount probe, dew-point temperature (Td) measured with an Edgetech hygrometer and pressure (P) measured with a Digiquartz sensor. The three-component wind speed (Ux, Uy, Uz) was derived from a Rosemount 858 probe, GPS and Honeywell HG1700 inertial measurement unit. The uncertainties of horizontal and vertical winds on the aircraft are 0.6 and 0.4 ms-1, respectively (Williams and Marcotte, 2000). Geospatial information (latitude, longitude, ellipsoid height altitude) was measured by GPS.

Nitrogen oxides (NO, NO2 and NOy) were measured with a modified trace level chemiluminescent analyzer (Thermo Scientific model 42i-TL). A molybdenum converter (325 ∘C) was used to convert NOy species to NO and an NO2-specific converter (Droplet Measurement Technologies) was used to convert NO2 to NO. Detection limits for NO, NO2 and NOy were determined to be 0.08 ppb (1 s), 0.20 ppb (2 s) and 0.09 ppb (1 s), respectively. Sulfur dioxide (SO2) was measured with a pulsed UV fluorescence analyzer (Thermo Scientific model 43i-TLE) with a detection limit of 0.7 ppb (1 s). Ambient air was drawn in through filtered 6.35 mm (1/4′′) diameter PFA tubing taken from a rear-facing inlet located on the roof toward the rear of the aircraft. Measurements of NO, NO2, NOy and SO2 were made downstream of a constant pressure inlet system maintained at 770 mmHg with a total flow rate of 5 Lmin-1. In-flight zero and background determinations were made several times throughout each flight and the analyzers were calibrated multiple times during the study against National Institute and Standards Technology (NIST) certified reference gases.

Refractory black carbon mass (rBC) was measured with a Droplet Measurement Technologies (DMT) single-particle soot photometer (SP2). Ambient air was subsampled from the main aerosol flow that was brought into the main cabin with a forward-facing shrouded diffuser isokinetic aerosol inlet (Cheng et al., 2018). Benzene, toluene, ethylbenzene and xylenes (BTEX) were measured by a proton transfer reaction time-of-flight mass spectrometer (PTRMS) from the main gas inlet. Further technical details are provided elsewhere (Li et al., 2017). The delay time of each instrument was determined experimentally and through calculations based on sample flow rates and inlet volumes. Total delays are contributed to by the response time of the instruments (1–3 s) and the volume of sampling tubing. Data were adjusted to account for the total delay times of 2–6 s to spatially and temporally synchronize the different measurements (Picarro delay time 6 s). The average speed of the aircraft was 90 ms-1 during the research flights, thus providing a spatial resolution of 90–270 m based upon the internal response time of each measurement.

In total, there were 22 flights with 84 hours of measurements in the AOSR between 13 August and 7 September 2013. The flights were designed for three purposes: measurement of pollutant emissions from facilities (Gordon et al., 2015; Li et al., 2017), measurement of pollutant transformation downwind of the AOSR (Liggio et al., 2016) and comparison with satellite overpasses (Shephard et al., 2015). A total of 13 flights were dedicated to quantifying facility emissions with a minimum of two flights for each of the SML, SUN, CNRL, SAJ and SAU facilities. CH4 above background was not detected during the 2013 flights targeting the Imperial Kearl Lake (IKL) facility, which was not in full production mode at the time (but has since expanded significantly), or from the Suncor Firebag in situ operation. We did detect CH4 above background suspected to originate from the Suncor MacKay River operation (west of SML). We were not able to quantify this source separately; however, emissions from this source are included in one measurement of the total emissions from all mining facilities in the AOSR using a wide downwind screen (see Sect. 3.4 and Fig. 7). Several other flights are not included in the analysis due to unfavorable meteorological conditions, including wind shear problems or insufficient numbers of transects. In total, seven flights were found to be suitable for identifying and quantifying emissions of CH4 from the facilities. Figure 1 displays several of the flight tracks over and downwind of the target facilities north of the Fort McMurray airport.

The flight patterns designed for the quantification of emissions rates were of two types: (i) screen flights, wherein the aircraft flew transects perpendicular to the plume downwind of one or more facilities, and (ii) box flights, wherein the aircraft flew transects at multiple heights around a single target facility in a box-type pattern (Gordon et al., 2015; Li et al., 2017). The transects were performed at heights from 150 to 1370 m above ground level (a.g.l.), complemented by vertical profiles designed to determine the height of the planetary boundary layer (PBL) and to compare with ground-based measurements.

Following the TERRA methodology (Gordon et al., 2015), the time-resolved measurements were interpolated using covariance kriging to produce a 40 m (horizontal, s) by 20 m (vertical, z) contiguous screen of CH4 mixing ratios. Within TERRA, the CH4 mixing ratios are extrapolated from the lowest transect (∼ 150 m a.g.l.) to the surface using a constant, linear or half-Gaussian extrapolation depending on the type of source and the boundary layer conditions at the time. Uncertainty estimates (see section in the Supplement) are included according to the various types of surface extrapolation applied. Interpolated matrices were constructed for measurements of air pressure (Pair) and temperature (Tair) in order to determine the air mass balance within the box and to convert mixing ratios to mass densities. Spatially equivalent interpolations of wind velocity perpendicular to aircraft motion (U⊤) were created from the vector components of wind speed and direction measurements.

Emissions rates were determined according to the two different mass-balance approaches for screen and/or box flight patterns. Horizontal tracks at multiple altitudes flown perpendicular to the general wind direction produce a virtual screen downwind of the target that is intercepted by emission plumes from the facilities. Fluxes of CH4 moving through each 40×20 m (s×z) pixel can be determined from the interpolated matrices and integrated for a dimensional s by z target area according to Eq. (1): EScreen=∬s1,zbsn,ztCH4-CH4B×U⊤dsdz, where ([CH4]-[CH4]B) is the enhanced mixing ratio of CH4 above background, U⊤ is the horizontal wind velocity perpendicular to the screen (e.g., U×sin⁡θ, θ = angle between wind vector and airplane vector), s1 and sn are the horizontal integration limits along the screen transect, and zb and zt are the bottom and top vertical integration limits. Background mixing ratios of CH4 were determined from the outside edges of the screens away from plume sources. Because [CH4]B varies with height, a vertically variant background profile was subtracted from each vertical measurement column, an approach used in other mass-balance determinations (Cambaliza et al., 2014; Karion et al., 2013). Example vertical profiles of [CH4]B(z) for each day are included in Fig. S1 in the Supplement. The simple mass-balance approach represented by Eq. (1) can be applied to individual downwind screens from other flight paths (i.e., box flights) to determine CH4 emissions from specific sources within a facility.

The second mass-balance method used in this paper is to apply the full box-model TERRA (Gordon et al., 2015) to compute total emissions from all sources within a box, which is made up of multiple (4–6) screens forming a polygon encompassing a facility. This more rigorous mass-balance approach used for calculating total emissions from a facility is represented by Eq. (2): EBox=ECH+ECHT+ECV+ECVT+ECM, where EBox, the total emissions rate from all sources within a box, is the sum of the horizontal advective and turbulent fluxes (ECH and ECHT), vertical advective and turbulent fluxes (ECV and ECVT), and the change in CH4 mass within the box volume (ECM). Because the box includes screens that are downwind, upwind and lateral to sources, incoming (background) and outgoing (background + source) fluxes are determined as a part of the horizontal flux terms (ECH and ECHT). Vertical fluxes through the box top, normally ignored in the conventional mass-balance approaches (Eq. 1), are estimated according to the conservation of air mass within the box volume and the mixing ratio at the top edge of the box. ECM is estimated according to the time derivative of the ideal gas law based on measured changes in pressure and temperature over the flight time (see Gordon et al., 2015, for a full discussion).

The advantage of the box approach (Eq. 2) over the screen approach is a more precise estimate of total emissions by accounting for incoming and outgoing fluxes and meteorological effects within a volume. However, this flight pattern takes more time to completely surround a target facility. The advantage of the screen approach (Eq. 1) is the computation of CH4 fluxes per pixel, which can thus be used to spatially integrate individual emission plumes of arbitrary shapes when the sources can be spatially resolved. Studies applying aircraft mass-balance methods have used single-height transect (Karion et al., 2013; Peischl et al., 2016), single screens (Cambaliza et al., 2014; Walter et al., 2012), spiral (Wratt et al., 2001; Gatti et al., 2014) and full box flight paths (Gordon et al., 2015) for the purpose of determining emissions rates and characterizing meteorological conditions. The aircraft flights presented contained various segments of tracks that allowed applications of all the above methods. In this work we apply a systematic approach deriving information from each of these techniques for a comprehensive top-down characterization of CH4 sources and emissions in the region. Single-height transects are used to determine source chemical signatures by identifying CH4 enhancements and their associations with other trace gas species. Vertical profiles are used to determine the PBL height throughout flights. Single screens are used to determine CH4 emissions rates (Eq. 1) for facilities and their individual sources when plumes are spatially resolved. Box flights are used to determine total CH4 emissions from facilities at a lower uncertainty (Eq. 2) and source-specific emissions are determined where possible (Eq. 1).

The summertime emission rates measured in this study are reported in units of tonnes of CH4 per hour, an appropriate unit given the duration of the flights (i.e., a few hours). We do not attempt to derive annual emissions as the assumptions needed to do so are highly uncertain without measurements in other seasons for a volatile species such as CH4. However, we do make a first-order comparison to emission inventories and other studies that report emissions on an annual basis by downscaling the annual emissions to hourly emission rates using an assumption of a constant temporal factor throughout the entire year. This is appropriate for emission inventories that are based upon the measurement of emissions or emission factors in summer that then upscale their emission rates of CH4 to annual emissions using a constant temporal factor assumption (e.g., GOA, 2014). However, the assumption of a constant temporal factor is far from being validated and further measurements in different months are needed to understand the potential for seasonal variability of fugitive emissions of CH4.

Previous work shows that the box approach has a demonstrated uncertainty of 25–27 % for total emissions of CH4 from a facility in the AOSR (Gordon et al., 2015). Uncertainty due to the extrapolation of CH4 mixing ratios from the lowest height measurements to the surface was estimated to be 15 and 26 % in that study. In contrast, screen approaches used in other studies have estimated uncertainty in the range of 30–50 % (Cambaliza et al., 2014) with the main sources of error attributed to the reliability of plume characterization and the stability of meteorological conditions. In this study, uncertainties in both the box and screen estimates are reduced through (i) a high number of transects over a wide vertical range to accurately characterize vertical structure in the PBL, (ii) reliable measurements of background CH4 (or incoming fluxes for boxes), (iii) measurements of the PBL height to account for meteorological variance and (iv) measurements within time periods of minimal PBL change. In addition, the enhancement of CH4 in the plumes downwind of the facilities and the high precision of the Picarro instrument minimize uncertainties in plume characterization and background CH4. The overall uncertainty for computed CH4 emission rates for an individual determination was estimated to be less than 30 % (see the Supplement for a complete evaluation and discussion of uncertainties).

Two example flights from three different facilities (SML, SUN and CNRL) are presented to demonstrate that CH4 emissions in the AOSR are mainly from three source types: open pit mining, tailings ponds and facility activities. Emissions from the remaining two facilities (SAJ and SAU) were shown to be primarily open pit mining. Source categories were identified by measurements of CH4, NOy, BTEX and rBC. Figure 2 shows measurements from one low-level transect of a screen flight on 16 August 2013 (nine transects in total). This transect was flown at a height of approximately 150 m a.g.l. downwind of the SML and SUN facilities, showing a clear separation of emission sources from the two facilities. Four distinct plumes are visible, labeled A–D, with linear air parcel back trajectories indicated by red arrows. Back trajectories were determined using the wind speeds and wind directions measured on the aircraft at flight level from the positions of maximum CH4, back extrapolated as a general indicator of plume origin. This methodology creates a western bias in our plume origins. A more careful analysis of surface winds at several meteorological stations in the local vicinity at the time of the aircraft transect shows that surface wind directions were from ∼ 140–180∘ (SE) compared to the flight level winds, ∼ 220∘ (SW). The low-level surface winds are likely channeled by the river valley, which runs in a SE to NW direction. Thus, the trajectories of air masses originating at the surface and mixing upwards have a clockwise rotation, a very local effect, placing the actual plume sources further east than the linear tracks show in Fig. 2. Plume A shows a maximum mixing ratio of 2.68 ppm CH4, which is an enhancement of ∼ 0.58 ppm above a background of ∼ 2.1 ppm on this day in this region. This enhanced CH4 is associated with values of 2.3 µgm3 rBC and 47 ppb NOy. The simple linear wind back trajectory places the origin of the air mass near the western edge of open pit surface mining activity ∼ 100 min earlier, although as mentioned the actual source is likely slightly east of that location due to the clockwise rotation of the plumes. The combination of rBC and NOy is indicative of exhaust from heavy hauler diesel trucks that operate in open pit mines. However, significant CH4 emissions are not expected from the truck exhaust, as emissions factors of CH4 from off-road gasoline and diesel combustion indicate that the CH4 / CO2 emission ratio would be 1 to 2 orders of magnitude lower (Environment Canada, 2015) than the ΔCH4 / ΔCO2 observed in this plume (0.58 ppm CH4 / 16.1 ppm CO2). The disturbance of the oil sands at the mine faces by the trucks is a well-known source of CH4 with minor emissions of CO2 and other VOCs (Strausz and Elizabeth, 2003) as well as intermediate volatility organic compounds (Tokarek et al., 2018). Thus, Plume A is interpreted to be a combination of heavy truck exhaust, indicated by the presence of rBC and NOy, that spatially overlaps the mine face source of CH4. Plume D shows a similar chemical profile with a maximum CH4 of 2.40 ppm, ∼ 0.30 ppm above background, associated with elevated levels of NOy (40 ppb) and rBC (1.5 µgm-3). The back trajectory for Plume D is in agreement with an origin at one of two locations of open pit mining activity at SUN. The two plumes show a similar ΔNOy / ΔrBC ratio within the range of 15–30 ppb per µgm-3. We consistently measure this profile of NOy and rBC enhancements from active mines across all five facilities.

Plume B (Fig. 2) shows the highest CH4 mixing ratio at 4.19 ppm, which is an enhancement of ∼ 2.09 ppm above background. The back trajectory from the position of the maximum CH4 places the air mass over the western edge of Mildred Lake Settling Basin (MLSB) tailings pond ∼ 20 min earlier. The CH4 enhancement occurs simultaneously with a decrease in NOy and rBC and an enhancement of total BTEX from ∼ 0 ppb to a maximum of 7.6 ppb. Tailings ponds are known to contain significant quantities of BTEX compounds due to waste streams of mature fine tailings containing naphtha diluent flowing into the pond (Small et al., 2015). This is similar to the chemical profile observed in Plume C, with a back trajectory placing the air mass over one of several possible SUN facility tailings ponds shown in Fig. 2 (Ponds 6, 5 and 2–3 in Small et al., 2015). This indicates the presence of CH4 emissions from multiple tailings ponds. The anaerobic digestion of organic matter in the tailings pond is the primary mechanism for the production of this biogenic CH4 (Siddique et al., 2012). For Plume C the measured mixing ratio enhancements are 0.25 ppm CH4 and 2.3 ppb of BTEX. The lower CH4 enhancement compared to Plume B suggests that less CH4 is emitted from this pond, in agreement with Small et al. (2015). The peak-to-peak CH4 / BTEX ratios from Plume B and Plume C are ∼ 300 and ∼ 100 ppbppb-1, respectively. The difference in measured inter-facility CH4 / BTEX ratios could arise from a number of factors including different pond ages, history, depth, methanogenic behavior or the use of different diluents in each facility. The in-plume correlations of CH4 with the associated tracers (NOy, rBC and BTEX) for each of the Plumes identified in Fig. 2 are shown in Fig. S2 in the Supplement.

Our observations are qualitatively consistent with pond-specific industry-reported CH4 emission factors, which present SML and SUN Ponds 2–3 (Small et al., 2015) as the highest CH4-emitting tailings ponds in the region. We consistently measured relative enhancements from plumes downwind of SML and SUN according to the pattern of Plumes B and C in Fig. 2, demonstrating the feasibility of using BTEX compounds as tracers for the CH4 being emitted from tailings ponds. We expect that BTEX would be greatly reduced from the tailings ponds of those facilities using paraffinic froth treatment (e.g., SAJ) instead of naphtha. In such cases, light hydrocarbons could in principle be used as tracers for the tailings ponds emissions of CH4. However, we did not detect methane plumes above the detection limit that were distinct from the open pit mining plumes of CH4 associated with rBC and NOx for any facilities other than SML and SUN.

Elevated plumes from facility stacks are the primary sources of SO2 in the AOSR due to the bitumen upgrading process. Hence, a significant enhancement of SO2 can be used as a tracer for plant or stack CH4 sources. However, this is not measured at the height shown in Fig. 2, which shows a maximum SO2 of only 5 ppb for this transect between Plumes C and D at 150 m a.g.l. For the same flight (Fig. 2), maximum SO2 was 131 ppb for a transect ∼ 350 m above the ground, with an associated narrow peak of CH4 with a maximum mixing ratio of 2.11 ppm. While higher-altitude SO2 plumes were frequently measured downwind of various facilities over the course of the aircraft campaign, in most cases no significant CH4 enhancements were observed in these plumes. A similar case is discussed in Sect. 3.2 in which we show the full range of vertical measurements and a lack of enhanced CH4 in the SO2 plume. Ground-level CH4 from tailings ponds and open pit mine faces therefore dominates the CH4 emissions in the region, with minor contributions from industrial plants.

We next compare the profiles from SML and SUN to a third facility, CNRL Horizon. Figure 3 shows a similar transect at ∼ 150 m a.g.l. from the 2 September flight in the vicinity of CNRL. The bottom panel of Fig. 3 shows that there was considerable wind divergence at this time (see back trajectory arrows for A, B and C). This wind divergence was also present in the next pass of the aircraft on the south side of CNRL at a height of 300 m (not shown). While this divergence aids in the visualization of source separation, it invokes uncertainty in the mass-balance determinations. The emission rates on 2 September were determined using 10 transects from a flight much earlier in time than that shown in Fig. 3, when the winds were more consistent in direction (NNW).

While Plume A shows a small enhancement of ∼ 1 ppb BTEX downwind of the tailings pond, in contrast to SML and SUN no significant CH4 was associated with it. This is consistent with the pond-specific emission factors presented in Small et al. (2015) that do not list the CNRL tailings pond as a significant source of CH4. The primary Plume B included a CH4 mixing ratio up to 2.24 ppm (an enhancement of ∼ 0.34 ppm above background) associated with 12 ppb NOy and 0.7 µgm3 rBC downwind of the CNRL mine. Consistent with the previously described open pit profile and the back trajectory, we identify Plume B as an open pit mining source of CH4. A secondary Plume C was measured with a maximum CH4 of 2.02 ppm (an enhancement of ∼ 0.12 ppm) east of the open pit mine. The lack of associated species does not relate the origin of Plume C to a tailings pond or an open pit mine source of CH4. The plume is downwind of the main CNRL plant and closer in horizontal proximity to SO2 plumes measured during higher-altitude transects. This suggests a CH4 source near the main plant that could originate from venting or flaring activity, electricity cogeneration using natural gas or natural gas leakage. Thus, the primary source of CH4 from the CNRL facility is open pit surface mining activity with a secondary undetermined source from the main plant.

Source profiles of CH4 are further compared to measurements of ethane (C2H6). Source-attribution studies for CH4 commonly use higher ethane-to-methane ratios (EMRs) as a signature for oil and gas emissions on both a regional (Peischl et al., 2016) and global (Hausmann et al., 2016) scale, while low EMR ratios can be indicative of microbial sources of methane that do not emit ethane (agriculture, landfills, wetlands, etc.; Smith et al., 2015). C2H6, along with other VOCs, was measured from 20 s grab samples collected in 3 L Summa canisters. The VOCs were analyzed offline using GC-MS and GC-FID methods described elsewhere (Li et al., 2017). Table 1 shows C2H6 measurements from three different flights (14 August, 16 August and 2 September) when canister sampling overlapped the plume descriptions listed previously. In all cases shown, enhancements of C2H6 above background (0.8–1.5 ppb) were in the range of only 1–2 ppb, normally the highest enhancements for each flight (within 1 ppb of 95th percentile). The small emissions rates of ethane (EMRs < 1.4 %) across flights contrast with the high EMRs (i.e., 40–50 %) seen for conventional oil and gas fields in other regions of North America (Peischl et al., 2016; Smith et al., 2015) and are lower than all the possible EMR source scenarios tested in Hausmann et al. (2016). The problems associated with determining EMR ratios from a combination of continuous CH4 measurements and the discrete canister sampling of ethane from aircraft have been highlighted recently, and it was shown that actual EMR ratios determined in this way can be off by up to a factor of 2 (Smith et al., 2015). Thus, the limited EMR data shown in Table 1 are not intended to be a comprehensive measure of EMR in the AOSR but simply to support the conclusion that the major sources of methane from the facilities in the AOSR are microbial in nature without a significant co-emission of ethane. The low EMRs are consistent with previous measurements in the region (Simpson et al., 2010) and are an indication of the unique character of unconventional bitumen sources. As such, global estimates of the relative contributions of oil and gas emissions to decadal increases in atmospheric CH4 that are based on C2H6 and CH4 measurements in the free troposphere (Hausmann et al., 2016) would not capture AOSR emissions due to the low C2H6 emissions in this region.

The source chemical profiles in Sect. 3.1 can be used in combination with the screen mass-balance method (Eq. 1) to isolate and quantify categories of AOSR emissions. As an example, we show the 14 August flight surrounding the SML facility, which consisted of a box and screen path flown in rapid succession. Figure 4 shows an image of the interpolated aircraft measurements from the box path creating a contiguous mesh superimposed on a map of the region. Winds were from the south at 186 ± 48∘ over the course of the day. Three distinct ground-based plumes of CH4 are visible: a primary plume (Plume N) on the northern screen (∼ 6500 m wide) exiting the box, a secondary plume (Plume NW) at the northwest corner (∼ 7000 m wide) exiting the box and a smaller plume (Plume E) on the eastern screen (∼ 3000 m in width) entering the box from outside the SML facility boundary. The lowest aircraft transect was at a height of ∼ 150 m a.g.l., with maximum CH4 mixing ratios of 3.00, 2.60 and 2.63 ppm, respectively, for the three plumes. Mixing ratios of CH4 below 150 m a.g.l. are based on a linear extrapolation of interpolated pixels to the surface corresponding to maximum surface mixing ratios of 3.48, 3.17 and 3.06 ppm for the primary (N), secondary (NW) and tertiary (E) plumes, respectively. Extrapolation to the surface is the primary source of uncertainty for surface sources in this method; however, the uncertainty can vary between flights depending on the meteorological conditions (Gordon et al., 2015). As a part of our uncertainty analysis in the Supplement, we have included an uncertainty associated with the differences in emission rates that arise from the use of linear, constant and half-Gaussian extrapolations in the calculations.

Unwrapped curtain plots of CH4, BTEX, NOy, rBC and SO2 from the 14 August box flight (Fig. 4) are shown individually in Fig. 5, projecting the 3-D virtual box onto a 2-D grid. The same three plumes from Fig. 4 are highlighted by dotted boxes in red (N screen), yellow (NW corner) and black (E screen). The red and yellow boxes show sources originating from within the SML facility and the black boxes show a source originating outside of the SML boundaries and entering the box from the east. The largest SML CH4 plume is associated with > 10 ppb BTEX and the absence of rBC and SO2, with some NOy (∼ 20 ppb). This is consistent with the chemical signature associated with tailings pond emissions discussed previously. The NW plume is associated with > 60 ppb of NOy and up to 5 µgm-3 of rBC, with minimal BTEX and SO2, consistent with the expected chemical signature from open pit surface mining. The smaller plume on the E screen is associated with elevated BTEX and CH4 and is likely a plume from one of the SUN tailings ponds as the winds indicate the plume is entering the box. The elevated plume in Fig. 5 (orange circles) with ∼ 100 ppb of SO2 and ∼ 30 ppb of NOy is traced to the SML upgrader activities, but with no enhancement of CH4 above background on this day. A second NOy plume is visible at the northeastern corner of the box, not associated with any of the identified CH4 source types. This NOy plume likely originates from traffic on the main highway that passes between the SML and SUN facilities and/or trucks and other vehicles operating in and around the main SML facility.

Boundaries of the plumes from separate sources are estimated using the tracer species listed in Fig. 5 by evaluating where the chemical signatures reached background levels. However, the SML tailings pond and open pit mine plumes were not completely resolved from one another, overlapping within a range of ∼ 800 m. The uncertainties in the emission rates due to plume overlap were estimated by contracting and expanding the horizontal integration boundaries (s) by 800 m on each side (a total of ± 1600 m) as part of the sensitivity analysis. A vertically varying background profile ([CH4]B(z)) is determined using data from the upwind southern screen, as mentioned previously. Using a spatially identical screen of perpendicular wind U⊤(z), the fluxes are determined through each pixel and the total source emission is calculated by integrating the pixels within the plume boundaries (Eq. 1). CH4 emissions rates are computed to be 6.4 ± 1.2 tonnes per hour (th-1) for the SML main tailings pond and 2.7 ± 0.6 th-1 for the SML open pit mine source. It is possible that the CH4 plume from the SML tailings pond includes CH4 emissions from the main SML plant facility (flaring, venting, natural gas leakage, etc.) that cannot be spatially separated from one another due to their proximity; however, we anticipate that the magnitude of these emissions are minor and captured within the error intervals listed.

This screen-based mass-balance approach for determining specific source emission rates (Eq. 1) is applied to flights with appropriate conditions for plume separation. Mean emissions rates of CH4 from specific sources within the facilities SML, SUN, CNRL, SAU and SAJ are shown in Fig. 6. SML emissions rates are the average of three mass-balance flights over 2 days (two on 14 August and one on 16 August). Two flights on separate days were used for each of the SUN (16 and 29 August), SAJ (21 August and 6 September) and SAU (29 August, 6 September) facilities. One CNRL flight (2 September) had northerly wind conditions showing plume separation on a southern screen. No significant daily variability is observed as the emissions rates for the same source agree within error. Duplicate and triplicate estimates for the same source are combined using an error-weighted uncertainty (the Supplement). SML and SUN had significant open pit mining emissions of CH4 of 2.8 ± 0.4 and 1.8 ± 0.2 th-1, respectively, and were the only facilities with tailings ponds emissions above the detection limit, 6.4 ± 0.8 and 2.4 ± 0.3 th-1. CNRL had open pit mining emissions (2.6 ± 0.7 th-1) and significant emissions originating from the main plant facility (1.0 ± 0.3 th-1). Plumes of CH4 from SAJ and SAU were only attributed to open pit emissions of 1.2 ± 0.2 and 1.4 ± 0.2 th-1, respectively.

The plume-targeting screen mass-balance method described here is unable to resolve emissions of CH4 from multiple sources not characterized by the chemical profiles described in Sect. 3.1 if they cannot also be spatially separated. Because spatial s and z constraints are manually chosen by plume boundaries from chemical profiles, minor emissions may contribute to an overestimation of the emissions from an individual source when highly coincidental in space such that the sources are not separable. For example, the emissions from the main plant were identifiable in the case of CNRL due to the separation and orientation of the plant, the open pit and the tailings ponds with respect to the winds. This was not the case for the other major facilities in the AOSR where many of the sources were highly coincidental in space. It is possible and even likely that other major facilities in this study also have CH4 emissions from their main plants (venting, cogeneration, natural gas leakage, etc.) that are identified as tailings pond emissions or open pit emissions due to close proximity and our inability to deconvolute the sources spatially or chemically. However, we expect that the total emission rates of CH4 from each facility are still accurate.

Emissions rates from each flight and individual sources (where possible) using the screen mass-balance method are tabulated in the Supplement (Tables S1–S5). We did not measure a detectable tailings pond source of CH4 from CNRL, SAJ and SAU. Associated enhancements of rBC and NOy with CH4 suggest that the CH4 source from SAJ and SAU is also predominantly open pit mining. The results using the screen mass-balance approach (Eq. 1) are further verified in Sect. 3.3 using emissions rates for each facility determined from the box approach (Eq. 2).

The total emissions rates of CH4 from each facility determined using the box mass-balance method (TERRA) are tabulated in the Supplement (Tables S1–S5) along with the determinations using the screen approach. Where multiple screen estimates or multiple box estimates were available, uncertainty-weighted (1/σ2) averages were determined for each method for each facility and are summarized in Table 2. While the box method is in some cases based on the same downwind measurements as the screen approach, the two methods have several key differences (described in Sect. 2.3) and are treated as independent estimates. In particular, the box method does not resolve specifically targeted, individual plumes and instead determines the net outgoing flux from the closed volume surrounding the facility. Thus, consistency between the two estimates is evidence that the primary sources of CH4 from facilities in the AOSR are tailings ponds, open pit mines and facility emissions captured by the source characterization in Sect. 3.1 and 3.2. In general, the total emissions from each facility using the screen and box methods are in agreement within uncertainty, which adds confidence to the measured emission rates reported here. In the final row of Table 2, we calculate a weighted average emission rate for each facility using all screen and box measurements. The CH4 emission rates from the facilities are 8.6 ± 0.9, 4.2 ± 0.4, 3.6 ± 0.5, 1.3 ± 0.2 and 1.5 ± 0.2 tCH4h-1 from the SML, SUN, CNRL, SAJ and SAU facilities, respectively.

The total CH4 emissions from the five mining facilities in the AOSR, obtained by summing the best estimates (i.e., uncertainty-weighted average of multiple measurements; bottom row, Table 2) of the individual facility emission rates, is given in the final row and column of Table 2. The five-facility total emission rate of CH4 is 19.2 ± 1.1 tCH4h-1. A final independent estimate of total AOSR emissions was obtained from a flight on 16 August utilizing an independent transect screen ∼ 75 km wide (16 August Screen B) downwind of all major mining facilities in the AOSR (excluding Imperial Kearl Lake and Suncor Firebag operations, but also inclusive of any emissions from the Suncor MacKay River in situ facility). The details of this flight are given in the Supplement Table S6. The interpolated screen from the 16 August flight (total OS) is shown in Fig. 7. The screen was constructed from 10 aircraft horizontal transects from 250–900 m a.g.l. Enhancements of CH4 were measured over a wide horizontal subrange of ∼ 60 km of the entire ∼ 75 km screen. Winds were perpendicular to the plane from the southwest (225∘), showing a large flux of CH4 through the screen from upwind sources. The highest measured mixing ratios of CH4 were 2.67 ppm at the ∼ 250 m a.g.l. transect. Background CH4 in the region was ∼ 2.00 ppm taken as a vertical profile from the wings of the screen. Using the screen method (Eq. 1), the emissions rate was determined to be 23.0 ± 3.7 tCH4h-1, which represents the emissions from all major facilities within the AOSR domain. This AOSR total is only slightly larger than the previous five-facility total emission rate of 19.2 ± 1.1 th-1, but not statistically so, demonstrating the reproducibility of our measured estimates. It is entirely possible that there are other minor sources of CH4 included in this larger number from smaller industrial operators in the region, trucks and vehicles on the main highway, and wetland emissions. In fact, the Canadian GHGRP inventory (see Sect. 3.5) indicates that there is an additional 0.13 tCH4h-1 emitted upwind and 0.17 tCH4h-1 emitted downwind of the aircraft screen (Fig. 7) from minor industrial facilities within the AOSR (both numbers downscaled from the facility-reported annual emissions). The amount of CH4 emitted from vehicles on the highway, however, is expected to be smaller. The fact that the two numbers are not statistically different supports the determination that the majority of the CH4 in the AOSR is emitted from the five major industrial facilities in the region. The two values are combined here using an error-weighted uncertainty, resulting in a final AOSR facility emissions estimate of 19.6 ± 1.1 tCH4h-1 measured during a summertime period.

Emissions of anthropogenic greenhouse gases are estimated by ECCC in Canada's GHG Inventory, which forms the basis for Canada's annual report to the United Nations Framework Convention on Climate Change, UNFCCC (ECCC, 2016). Currently, industrial facilities that emit more than 50 ktCO2eqyr-1 are required to report their emissions annually to ECCC using the Greenhouse Gas Reporting Program (GHGRP), which is Canada's legislated, publicly accessible inventory of facility-reported greenhouse gas (GHG) data (ECCC, 2017a). Although the GHGRP inventory data are not necessarily used in Canada's GHG Inventory, changes are being proposed to expand monitoring requirements in the GHGRP, including lowering the reporting threshold to 10 ktCO2eqyr-1 in order to enable direct use of the reported data in Canada's GHG Inventory (ECCC, 2017b). Emissions of CH4 from all five major oil sands facilities discussed in this paper are present in the GHGRP Emissions Inventory on an annual basis. The annual emission rates of CH4 extracted from the inventory were downscaled to hourly emissions rates for comparison with our measurements with an assumption of equal seasonal and diurnal profiles 365 days a year and 24 h per day for consistency with upscaling factors used to generate annual emissions (see Fig. 8). While this may be questioned, it should be noted that fugitive emissions of CH4 from mine faces and tailings ponds in the inventories are estimated based upon emission factors measured at oil sands facilities during summer months (June–September), which are then upscaled from hourly emissions to annual emissions using the same assumption that we used to downscale (365×24), as per the recommendation by the government of Alberta (GOA, 2014). Specifically, it is noted from the GOA report that emissions of gaseous species such as CH4 (and CO2) are not temperature dependent (unlike VOCs that have temperature-dependent vapor pressures; Li et al., 2017). The argument used to justify the use of a constant seasonal temporal factor in the GOA report is that temperatures at depth in a tailings pond are said to remain relatively constant throughout the year, and thus biogenic gas formation continues in the winter (GOA, 2014). For mine faces, the GHG component of the oil sand does not change with temperature and is likely released completely in a short period of time after being mined. Thus, the government recommendation to oil sand facilities in preparing annual emission estimates of fugitive GHGs is that reduction factors should not to be used in extrapolating summertime emissions over the rest of the year (GOA, 2014). Figure 8 shows a comparison of the total measured emission rates of CH4 from the five industrial facilities (2013), the total measured CH4 emission rate in the AOSR from the single downwind screen on 16 August 2013 and the sum of the facility emission rates from the Canadian GHGRP Emissions Inventory for 2013 expressed in hourly units. The combined facility emissions rate of 19.6 ± 1.1 th-1 is approximately 48 ± 8 % higher than the five-facility total of 13.2 th-1 extracted from the inventory for 2013. Facility-to-facility comparisons show higher measured than reported emission rates for three out of the four facilities (the SML and SAU facilities are combined as one in the inventory). In contrast, for CNRL our measured emission rate is 1.2 th-1 lower than the inventory. Since we have determined the composition of SML, SUN and SAJ emissions to be primarily from tailings ponds and open pit mining, there appears to be an underestimation in the inventory of those particular area sources within these sites.

These discrepancies indicate a need for inventory reconciliation between the bottom-up and top-down estimates. It has been shown that it is possible to reconcile divergent bottom-up and top-down CH4 estimates for the Barnett Shale by using more comprehensive activity factors and better characterization of emissions from high emitter sites (Lyon et al., 2015) and continuous monitoring to identify these super emitters (Zavala-Araiza et al., 2015). Currently, bottom-up estimates in the AOSR are accomplished by systematic surface flux chamber measurements of area sources (surface mines, tailings ponds) to derive area-based emissions factors (GOA, 2014). While surface flux chamber measurements (Klenbusch, 1986; Conen and Smith, 1998) are estimated to be 50–124 % of the true emissions rate for a homogenous source (Klenbusch, 1986), it is unclear how the uncertainty propagates when the emissions factors are scaled to the full surface area of the heterogeneous AOSR emissions sources. The official survey protocol for open pit sources attempts to minimize the possibility of underestimating emissions by explicitly requiring fugitive surveys to include sampling at a range of locations within the open pit mine, where safe to do so, including high-priority zones (disturbed in the last week), normal-priority zones (disturbed from 1 week to 6 months ago) and low-priority zones (disturbed > 6 months ago; GOA, 2014). However, it seems that the recent core sampling methodology outlined by Johnson et al. (2016) has great promise and reduced uncertainty for estimating fugitive emissions from open pit mining.