Quantifying methane emissions from Queensland’s coal seam gas producing Surat Basin using inventory data and an efficient regional Bayesian inversion

Methane emissions across Queensland’s Surat Basin, Australia, result from a mix of activities, including the 10 production and processing of coal seam gas (CSG). We measured methane concentrations over 1.5 years from two monitoring stations established 80 km apart on either side of the main CSG belt located within a study area of 350 × 350 km2. Coupling bottom-up inventory and inverse modelling approaches, we quantify methane emissions from this area. The inventory suggests that the total emission is 173 × 106 kg CH4 yr-1, with grazing cattle contributing about half of that, cattle feedlots ∼ 25%, and CSG Processing ∼ 8%. Using the inventory emissions in a forward regional transport model indicates 15 that the above sources are significant contributors to methane at both monitors. However, the model underestimates approximately the highest 15% of the observed methane concentrations, suggesting underestimated or missing emissions. An efficient regional Bayesian inverse model is developed, incorporating an hourly source-receptor relationship based on a backward-in-time configuration of the forward regional transport model, a posterior sampling scheme, and the hourly methane observations. The inferred emissions obtained from one of the inverse model setups that uses a Gaussian prior 20 whose averages are identical the gridded bottom-up inventory emissions across the domain with an uncertainty of 3% of the averages best describes the observed methane. Having only two stations is not adequate at sampling distant source areas of the study domain, and this necessitates a small prior uncertainty. This inverse setup yields a total emission that is very similar to the total inventory emission. However, in a subdomain covering the CSG development areas, the inferred emissions are 33% larger than those from the inventory. 25 https://doi.org/10.5194/acp-2020-337 Preprint. Discussion started: 15 May 2020 c © Author(s) 2020. CC BY 4.0 License.

loitering to the east of the intake line. While the 'tight' or cattle filter removes most of these sharp peaks in the minutely data, it retains the underlying rise in CH 4 between 0400 and 0600 UTC which mirrors the rise in CO, suggestive of biomass burning signal being transported from further afield. 30 The same cattle filter was applied to the Ironbark data, for consistency, although cattle are fewer and further away at Ironbark and have much less impact on the methane measurements.

S1.2 Diurnal and low wind filtering
Nocturnal low wind conditions can be approximately defined as those with a wind speed at 10-m above ground of less than 2-3 m s -1 , which corresponds to a Richardson number (a stability parameter that is the ratio of buoyant suppression of 35 turbulence to shear generation of turbulence in the lower atmosphere) greater than 0.2 − 0.3 (Luhar et al., 2009). There are considerable occurrences of high methane concentrations at the two sites under such conditions. This is mostly because under these conditions the atmosphere near the ground is typically characterised by strong stable stratification with a very shallow inversion layer so that even small local sources near the ground can lead to very large enhancements in the local methane concentration due to very little vertical atmospheric mixing. Despite being of considerable practical interest, 40 however, these are some of the most difficult conditions to simulate by a flow and dispersion model, particularly at a regional or mesoscale. Thus, one option to circumvent the issue of modelling generally not being able to properly simulate strong inversion conditions at night is to consider daytime hours (1000-1700 h) irrespective of wind speed, and the remaining hours for which the wind speed is greater than 3 m s -1 . The daytime window typically corresponds to periods of strong mixing dominated by convective motions resulting from the solar heating of the ground. For data selection for the two 45 sites, the respective measured wind speeds were used.

S1.3 Filtering for biomass burning events
Both methane and CO are emitted from biomass burning. CO is not present in other large methane sources of interest, including those compiled in the bottom-up emissions inventory. Methane emissions from power stations, domestic wood heating and vehicles on the other hand would contain CO, but the modelled CH 4 signals for these sources are predicted to be 50 virtually undetectable at Burncluith. The majority (89%) of CSG methane source emissions in the bottom-up inventory are not from combustion. Importantly, emissions from the less well known migratory or seepage sources, which also do not originate from combustion, would not be screened out by a CO filter. Thus, comparisons of observed CH 4 with model simulations are more accurately made by excluding hourly periods with large CO enhancements above the background CO concentration. 55 A plot of the measured CO vs CH 4 concentrations at Burncluith after applying the above cattle and low wind filtering is shown in in Figure S2 (orange circles). Two distinct groupings are apparent; the data group with high magnitudes of CO concentration likely represents dominant contributions from combustion sources. Enhancements of CO above background at Burncluith are mostly observed during north-westerly and easterly winds, consistent with the locations of the occasional forest burn offs and the wood fire in the dwelling adjacent to the monitoring station, respectively. (The background CO 60 concentration was calculated using the same methodology as the background CH 4 (Section 4.2). To filter out such events, we chose an hourly mean CO enhancement of 10 ppb (above the background) as a cut off, which is about twice the one standard-deviation uncertainty in the observed CO around the estimated background CO variation without considering the CO enhancement periods. The hourly mean data points after removing the data points with CO values greater than 10 ppb above background concentrations are shown as blue dots in Figure S2. This CO filter further removed about 22% of the 65 filtered Burncluith data.

S2. Bottom-up methane emission inventory
The following is a brief account of how the bottom-up methane emissions from the various sectors for the year 2015 were compiled. Full details are given in a report by Katestone (2018).

S2.1 Grazing cattle 70
The information used to estimate methane emissions for grazing cattle included: • Methane emission factor for grazing cattle based on direct measurements (Harper et al., 1999). 75 The number of grazing cattle were calculated in each NRM region and was distributed uniformly across the region. This was then multiplied by the emission factor to give the corresponding methane emissions. There were 1,086,059 grazing cattle in the study area.

S2.2 Feedlots
The following information was used to compute methane emissions for cattle in feedlots: 80 • National Pollutant Inventory (NPI) data for the 2014/15 reporting year with the ANZSIC (Australian and New Zealand Standard Industrial Classification) description "Beef Cattle Feedlots (Specialised)".
• Queensland Government datasets including Lot and plan boundaries (Property boundaries Queensland cadastral dataset) and Locations and standard cattle unit numbers contained in NRM regions (Department of Agriculture and Fisheries). 85 • Methane emission factor for "enteric fermentation" and "manure management" for non-dairy cattle from the Food and Agriculture Organization (FAO) of the United Nations (FAOSTAT, 2016). A total emission factor combines the above two methane sources at a feedlot.
There were 235 cattle feedlots in the study area. The number of cattle per feedlot was estimated. The methane emissions were calculated by multiplying the number of cattle per feedlot by the emission factor. 90

S2.3 Coal Seam Gas (CSG) activities
The locations of CSG wells and processing facilities were based on data available through DNRM and methane emissions data and calculations were provided by the operators.
The information used to calculate the CSG methane emissions included: locations of CSG wells and processing facilities, methane emissions data and reporting prepared for the National Greenhouse and Energy Reporting (NGER) program and 95 directly from the operators, quantity of gas combusted and the volume of produced water.
The calculation methods are consistent with the NGER program where methane is classified as a greenhouse gas and is quantified and reported in terms of carbon dioxide equivalents.
Combustion: Emissions of methane due to combustion of CSG (including flaring) and diesel were calculated as the product of the quantity of fuel type, the energy content of fuel type, an appropriated emission factor and the GWP of methane. 100

Fugitive emissions (other than venting or flaring): These fugitive methane emissions include emissions from (NGER 105
Determination 2008, Section 3.70 (Clean Energy Regulator, 2016)): a gas wellhead through to the inlet of a gas processing plant, a gas wellhead through to the tie-in points on gas transmission systems (if processing of natural gas is not required), gas processing plants, well servicing, gas gathering, gas processing and associated waste water disposal. The emissions are calculated as the product of the total quantity of natural gas, the appropriate emission factor and the GWP of methane.

S2.4 Coal mining 110
Methane emissions for four coal mines in the study were calculated from the following information (only the dominant coal extraction process was included): • DNRM Mining lease surface areas. 115 • Fugitive methane emission factor for extraction of coal in Queensland of 0.02 tonnes CO 2 -e per tonne of raw coal (DoE, 2016).
The methane emissions for each mine were calculated by multiplying the amount of ROM coal by the appropriate emission factor. They were allocated uniformly across the mining lease areas associated with each mine as identified from the DNRM 120 dataset.

S3. Estimating the background methane concentration
Data from the measured concentration time series were retained if they occurred between 1200 -1500 h local time (typically the time of highest boundary layer height and maximum trace gas homogeneity during the diurnal cycle) and the hourly standard deviation of concentration was less than or equal to 1 ppb, indicating very well mixed conditions. This filtered 125 dataset was then used to derive a smooth curve. Based on the method described by Thoning et al. (1989), the filtered dataset was fitted with a function consisting of a cubic polynomial and three harmonics. This fit is then subtracted from the filtered data and the residuals further filtered with a band-pass filter of 80 days. The original function fit is then added back to the filtered residuals to give a smooth fit through the data. These operations are performed iteratively (with hours lying outside twice the standard deviation around the fit excluded) until the fit converges. An interpolation routine then produced the fitted 130 background CH 4 concentrations at each of the hourly timestamps of the original measured data.

S4. Model performance for meteorology
In Figure S3, the modelled winds are qualitatively similar to those observed, with the most frequent modelled wind direction also from the north-east quadrant, and winds from the south-west quadrant modelled at a relatively smaller frequency in agreement with the observations. The modelled winds at Burncluith and Ironbark are more similar than those observed. At 135 Burncluith the model underestimates the frequency of low wind speed events (< 2 m s -1 ), which mostly occurs at night, and overestimates the frequency of higher wind speed events (> 4 m s -1 ) from the north-east sector. The wind speed distribution at Ironbark is better modelled than that at Burncluith-one reason for this could be that Burncluith has several tall trees in the vicinity which may weaken the flow field and whose influence is not properly accounted for in the model. There is also a difference in the height at which winds are given: the model height is 10 m whereas, it is 7.6 m at Burncluith and 5.8 m at 140 Ironbark. Generally, in the surface layer, winds get stronger with height, and, therefore, one factor in the modelled winds being stronger than the observation could be the height difference.
meteorological evaluation of the model here, except to report that the overall model-data correlation coefficient (r) for wind speed at Ironbark was 0.68 and it was 0.66 at Burncluith. Another parameter, the Index of Agreement (IOA, = 0 no 145 agreement, = 1 perfect agreement), which, unlike the correlation coefficient, is sensitive to differences between the observed and model means as well as to certain changes in proportionality (Willmott, 1981) was 0.82 for Ironbark and 0.76 for Burncluith. As judged from the IOA values, the overall TAPM performance for winds for the Surat Basin is satisfactory and comparable to those in other studies (e.g., Luhar and Hurley, 2003;Hurley et al., 2005) (also see papers in TAPM citation database https://scholar.google.com.au/scholar?oi=bibs&hl=en&cites=13876071272134760358). 150 Figure S1. Burncluith minutely-mean data for 10 January 2017. Blue curve represents the default filtering, red curve the cow filtering ("tight"). Where the two curves cannot be seen separately, they overlap. Figure S2. Hourly mean concentrations of CO versus CH 4 measured at Burncluith selected for 1000-1700 for all wind 160 speeds and for 1800-0900 for wind speed greater than 3 m s -1 (orange circles). The data group with high magnitudes of CO concentration likely represents dominant contributions from combustion sources. The data marked with blue dots are the measurements when hourly mean CO concentrations are within 10 ppb of the background CO concentration at the time of measurement and are selected to represent contributions from non-combustion sources.