Assessing fugitive emissions of CH 4 from high-pressure gas pipelines in the UK

Natural gas pipelines are an important source of fugitive methane emissions in lifecycle greenhouse gas assessments but limited monitoring has taken place of UK pipelines to quantify fugitive emissions. This study investigated methane emissions from the UK high-pressure pipeline system (National Transmission System - NTS) for natural gas pipelines. Mobile surveys of CH 4 emissions were conducted across four areas in the UK, with routes bisecting high-pressure pipelines (with a maximum operating pressure of 85 bar) and separate control routes away from the pipelines. A manual survey of soil gas measurements was also conducted along one of the high-pressure pipelines using a tunable diode laser. For the pipeline routes, there were 26 peaks above 2.1 ppmv CH 4 at 0.23 peaks/km, compared with 12 peaks at 0.11 peaks/km on control routes. Three distinct thermogenic emissions were identi ﬁ ed on the basis of the isotopic signal from these elevated concentrations with a peak rate of 0.03 peaks/km. A further three thermogenic emissions on pipeline routes were associated with pipeline infrastructure.Methane ﬂ uxesfromcontrolrouteswerestatisticallysigni ﬁ cantlylowerthanthe ﬂ uxesmeasured onpipelineroutes,withanoverallpipeline ﬂ uxof627(241 – 1123interquartilerange)tonnesCH 4 /km/yr.Soilgas CH 4 measurementsindicatedatotal ﬂ uxof62.6ktCH 4 /yr,whichequatesto2.9%oftotalannualCH4emissionsin the UK. We recommend further monitoring of the UK natural gas pipeline network, with assessments of transmission and distribution stations, and distribution pipelines necessary. © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

In the past decade, unconventional natural gas from shale deposits has been increasingly used as a source of energy, via stimulation through hydraulic fracturing.This technology has raised numerous environmental concerns, including the fugitive emission of methane (CH 4 ) through pre-production, production and transportation processes.Numerous studies have developed life-cycle emissions inventories to assess the impact that hydraulic fracturing has on greenhouse gas emissions (Balcombe et al., 2017;Burnham et al., 2012;Jiang et al., 2011).Incorporated within life-cycle assessments are transmission and distribution losses, including infrastructure such as pipelines and compressor stations that pressurize natural gas for transport along pipelines.Howarth et al. (2011) estimated fugitive emissions from the transmission, storage and distribution phase to total 1.4-3.6%.The figure of 1.4-3.6%has been disputed as too high (Burnham et al., 2012;Cathles et al., 2012) as the data used by Howarth et al. was based on Russian pipelines and was not applicable to the USA (Leliveld et al., 2005); and was based upon unaccounted for gas techniques (the difference between gas produced and sold) which are known to overestimate fugitive emissions (Burnham et al., 2012).Weber and Clavin (2012) downgraded the Howarth et al. (2011) loss rate to 0.8-2.2% for transmission only but cited the same concerns of the above studies; Stephenson et al. (2011) calculated fugitive emissions using facilitylevel factors for transmission pipeline from the 2009 API Compendium (API, 2009) and found a loss rate for transmission pipelines of 0.066% over 1440 km transportation distance.Overall, Weber and Clavin (2012) suggested transmission losses were 1.9 (1.2-2.5)g CO 2 e/MJ.While life-cycle emissions inventories provide insights into fugitive emissions of CH 4 across the oil and gas sector, it is important to quantify losses based upon observations from monitoring data including components of the transmission and distribution system.
Numerous studies have reported emissions from pipeline leaks and related infrastructure.Across the gas transport system in Russia, total transmission, storage and distribution CH 4 losses were found to be on the order of 1.4%; incorporating just the high-pressure transmission system, pipelines were estimated to contribute 999 × 10 6 m 3 CH 4 yr −1 , amounting to 6531 m 3 km −1 yr −1 (Leliveld et al., 2005).A further study of leaks from pipelines in Russia estimated 381 × 10 6 m 3 CH 4 yr −1 (11.3% of total emissions) and emissions from maintenance and repairs to pipelines were estimated at 17.0% of total emissions (Lechtenbohmer et al., 2007).In the USA, leaks from distribution, transmission and gathering pipelines were estimated at 1178.1 × 10 6 m 3 CH 4 yr −1 , 5.7 × 10 6 m 3 CH 4 yr −1 and 186.9 × 10 6 m 3 CH 4 yr −1 respectively (Kirchgessner et al., 1997).A bottom-up survey of CH 4 sources in the Barnett shale region indicated that gathering and transmission pipelines contributed 940 and 230 kg CH 4 h −1 (Townsend- Small et al., 2015).Peischl et al. (2013) conducted a topdown atmospheric survey of CH 4 emissions in Los Angeles and attributed 192 ± 54 Gg CH 4 yr −1 to natural gas, while Townsend-Small et al. (2012) confirmed fossil fuels as the major source of CH 4 in Los Angeles through isotopic analysis.Although Townsend-Small et al. (2012) and Peischl et al. (2013) indicated natural gas pipelines as a likely source of fossil fuel emissions, this was not confirmed through direct measurements from pipelines.
Natural gas pipelines include gathering, transmission and distribution pipelines that have different functions and operate at different pressures.Gathering lines transport natural gas from the wellhead to transmission lines while transmission pipelines transport natural gas from gathering, processing and storage facilities and operate at high pressure.In the UK, gas is delivered to terminals from offshore and is transported around the UK using the National Transmission System (NTS), with 23 compressor stations maintaining operating pressures of up to 85 bar (85 × 10 5 Pa).Distribution networks operate at lower pressure and include service pipelines that connect to customer's meter's or piping and mains lines that supply more than one service line.Several studies have started to quantify the scale of natural gas pipeline leaks, though the majority of research has focused on leaks in cities across the United States.Jackson et al. (2014) measured 5893 natural gas leaks, ranging from 2.5-88.6 ppmv CH 4 , from 2400 road km traveled in Washington, DC.Emissions from four street leaks from natural gas pipelines ranged between 9200 and 38,200 L CH 4 /day.An average loss of 2.7% from natural gas pipelines, 2-3 times higher than the best state estimates (1.1%) was found in Boston, Massachusetts using top-down atmospheric measurements (McKain et al., 2015).Lamb et al. (2015) measured 230 underground pipeline leaks across the USA to create emissions factors for service and mains distribution pipelines and suggested that such systems contributed 197 Gg CH 4 /yr (554 CH 4 Gg/yr, 95% upper confidence limit).
The condition of pipelines is an important factor in contributing to fugitive emissions from natural gas pipelines.Although cast iron and unprotected steel pipes amounted to b10% of all pipeline length in the USA, they contributed 46% of total emissions from pipelines (Lamb et al., 2015).In a further study of fugitive emissions from cast iron mains in Boston, MA, just seven leaks were responsible for 50% of CH 4 emissions measured (Hendrick et al., 2016).The estimated emissions from Lamb et al. (2015) were lower than 2011 US EPA estimates due to the effect of pipeline repairs and replacements from 1992, increasing plastic mains (+150%) while upgrading cast iron (−38%) and unprotected steel (−22%) pipes.Gallagher et al. (2015) found that cities in the USA with pipeline replacement programmes had 90% fewer leaks per mile than cities without, while comparatively few discrete natural gas pipeline leaks were detected in Los Angeles, where cast iron mains are not present (Hopkins et al., 2016).Indianapolis was estimated to have 0.08 leaks/km compared to 0.74 leaks/km in Boston, due to protected steel or plastic mains in Indianapolis and unprotected steel and cast iron mains in Boston (Lamb et al., 2016).Leaks were small in Ithaca, NY, at b0.24 leaks/km, due to only 2.6% of mains being bare steel or cast iron (Chamberlain et al., 2016).
In the UK, the iron mains replacement programme started in 1977 and has an aim of replacing the remaining 91,000 km of iron pipes within 30 m of buildings by 2032 (Dodds and McDowall, 2013).The UK distribution networks total 280,000 km of pipeline, with 7600 km of pipes in the NTS (Dodds and McDowall, 2013).Although there has been an increasing amount of research into leaks from gas pipelines in the USA, there are a limited number of studies elsewhere.Mitchell et al. (1990) suggested that for the UK distribution system, low, medium and high leakage rates were 1.9%, 5.3% and 10.8% respectively and it was argued that leakage rates above 1.9% were more likely.When assessing fugitive emissions of CH 4 from fault zones, Boothroyd et al. (2017) identified natural gas distribution pipelines as a possible source of thermogenic (−41.2‰δ 13 C-CH 4 ) CH 4 , of up to 10.1 ppmv along non-faulted control routes.However, not much else has been done to monitor fugitive emissions from pipelines in the UK.Industry estimates (Nelson, 2003) have provided leak rates for service and mains distribution pipelines in the UK, but these pipelines could be expected to have different leak rates and fugitive emissions of CH 4 than higher pressure transmission pipelines, for which no data are currently available.
Isotopic analysis of δ 13 C-CH 4 has been used to identify natural gas sources of CH 4 from pipelines and other natural gas infrastructure.Jackson et al. (2014) reported pipeline leaks across Washington, DC to have a δ 13 CH 4 isotopic value of −38.2‰, which was statistically indistinguishable from pipeline natural gas (−39‰).Although thermogenic and biogenic CH 4 have ranges of −50 to −20‰ δ 13 C-CH 4 and CH 4 −110 to −50‰ δ 13 C-CH 4 (Whiticar, 1999) respectively, the boundaries are not distinguished as factors such as oxidation and fractionation can affect δ 13 CH 4 composition.Phillips et al. (2013) reported an average δ 13 C-CH 4 of −42.8‰ in Boston, reflecting a natural gas signature that had been altered by fractionation through transport in soil and mixing with background air.Similarly, Townsend-Small et al. (2016) noted the effect of natural background air on natural gas signatures while Arata et al. (2016) observed a mixture of natural gas and biogenic signatures in New Mexico.Thus, isotope analysis is an important tool to identify the source of CH 4 emissions, though source composition can be more complex than defined thermogenic and biogenic boundaries.
In this study, we investigated fugitive emissions of CH 4 from the UK high-pressure NTS.Methane concentration was detected by driving along roads crossing high-pressure gas pipelines and non-pipeline control routes.Isotope analysis of δ 13 C-CH 4 was used to identify the source of fugitive CH 4 emissions.As a follow up one high-pressure gas pipeline was selected for a survey of soil gas measurements.Control surveys were undertaken on the same day in similar meteorological conditions to pipeline surveys.Control routes were surveyed to determine natural background levels of CH 4 as well as emissions in the study area that were not associated with natural gas pipeline leaks, such as from biogenic sources like farming.Control routes were selected to be in areas of similar land use to the pipeline routes, but away from the NTS.On a small number of occasions, the Northumberland control route bisected the high-pressure pipeline network where road layouts meant this was unavoidable, but CH 4 concentrations did not exceed 1.87 ppmv.Pipeline routes were longer than control routes due to taking circuits that traversed pipelines and returning back to cross pipelines as much as possible.Pipeline routes incorporated associated infrastructure to the high-pressure pipelines, such as gas sites where the high-pressure network transports gas to and from.Details of each section of pipeline that was bisected at least once are given in Table 1, with further details provided in the supplementary information.

Gas measurement and analysis
Methane concentration and δ 13 C-CH 4 were measured using a Picarro Surveyor P0021-S cavity ring-down spectrometer (Picarro Inc., Santa Clara, CA) while driving along pipeline and control routes.The spectrometer has a stated precision of 5 ppb + 0.05% of reading 12 C and all results are expressed as per mille relative to VPDB (Vienna PeeDee Belemnite) based upon a factory supplied calibration.Sample gas was measured at a frequency of 1 Hz through a sample line attached to the roof at the back of the survey vehicle (vertical height of sampling was 1.5 m).The Picarro software mapped wind plumes and identified source areas using wind speed (between 0 and 60 m/s ± 2% @ 12 m/s) and wind direction (0-359°± 3°) data from a 2D anemometer (WindSonic, Gill Instruments, Lymington, UK) attached to the roof the survey vehicle.Measurement location was determined using a GPS A21 (Hemisphere, Scottsdale, Arizona).
The raw concentration data were downloaded from the surveyor and converted into ArcMap (version 10) point shapefiles (Boothroyd et al., 2017).Using the point shapefiles imported into ArcMap, individual pipeline and control route lengths were calculated by converting points to polylines (i.e.connecting data points into lines to create the route) and using the measure tool to calculate the length of the polyline.A total of 467.9 km were traveled along the four pipeline and control routes.The distance between a given data point and the pipeline was calculated to the nearest meter using the Near feature in the ArcGIS toolbox.For control routes, a median line (see Section 2.3.1) between sections of the route traveled was mapped and the distance between it and the nearest point of measurement determined so that pipeline and control routes underwent the same treatment.Elevated CH 4 concentrations were identified as discrete peaks greater than the 99th percentile (2.1 ppmv CH 4 ) of all measured data.Although previous research (Boothroyd et al., 2017) used the 95th percentile to determine peak concentrations, in this study the 95th percentile was 1.94 ppmv CH 4 and so the 99th percentile was chosen to better distinguish higher concentrations of CH 4 .Pipeline and control routes were revisited the next day after the initial survey for δ 13 C-CH 4 isotopic measurements.Areas identified as having elevated CH 4 concentrations were revisited based on time constraints and allowing similar numbers of measurements between pipeline and control routes.To determine isotopic composition, realtime atmospheric measurements were conducted for ten minutes while the survey vehicle remained stationary at a given location.The isotope composition of sources was determined using Keeling plots of δ 13 C-CH 4 against the inverse of CH 4 concentration, with the intercept representing the source composition (Pataki et al., 2003).Thermogenic CH 4 was interpreted to be in the range of −50 to −20‰ δ 13 C-CH 4 and biogenic CH 4 −110 to −50‰ δ 13 C-CH 4 (Whiticar, 1999), though it is noted that mixing of CH 4 sources can occur within these ranges, as discussed in Section 1.

Data analysis
Data were censored relative to the wind direction, any data collected whilst the wind was in opposite half-disk (outside of 90°either side of the data point) from the nearest point on the pipe or control were removed and was not considered in the analysis of variance (ANOVA) and CH 4 flux determination.
For pipeline routes, isotopic data is presented for complete 10 min analytical periods described above and data wind-resolved to the direction of the pipeline, wherein data in the wrong half-disk were removed.Isotopic data was also transformed into 30-second averages for the 10minute analytical period to reduce natural variation from the atmospheric sampling that reduced the quality of regression.Isotopic compositions referred to in the text are from 30-second average data unless otherwise stated.All significant raw data, 30-second average and wind corrected Keeling plots are shown in Figs.S1-3, with a summary comparison and locations given in Table S1.If raw data plots were not significant, 30-second average and wind corrected plots were not created.Significance was judged at the 95% probability of the gradient of the Keeling plot being different from zero.Prior to any analysis, prolonged stationary periods (primarily when changing batteries to the Picarro Surveyor, which required the machine to be turned off for a short period followed by a warm-up period) were removed from the analysis, but periods in stationary traffic were not excluded.

Correcting concentration for distance
As there was no fixed distance from the Target (pipeline or control line), the CH 4 concentration would be expected to decline to ambient with distance.Consequently, any difference between Areas (Vale of Pickering, Durham, Northumberland, and Vale of Eden) or Target could be ascribed to distance away from the survey line at each point of measurement.The dynamic plume approach of Hensen and Scharff (2001) was used to control for the distance away from the survey line.A 3D Gaussian plume model was applied to the data of each pipeline or control survey, where the concentration of methane (in mg CH 4 / m 3 ) above the ambient methane concentration (typically 1.5 ppmv -1.29 mg/m 3 ) at a point away from a source is given by: where: x = shortest distance from point of measurement to the pipeline (m); y = the perpendicular distance along the fault of the measurement (zero m in this study); z = the height of the detector above the ground surface (1.5 m); Q = the source strength (mg/s); u = the wind speed resolved along x (m/s); H = the height of the source (m); and σ y and σ z = dispersion terms in the directions y and z.The dispersion terms are approximated as σ y = I y x, and σ z = I z x and in near surface conditions we assumed that there is no stable stratification and that therefore I z = I y = 0.5.Wind speed was resolved to the shortest distance to the target (u x ) by calculating the shortest distance (x) to the pipeline (or control line) from the point of measurement along with wind speed and direction at height z.Prior to analysis for pipeline distance, data when the wind direction was from the wrong half-disk was removed.Data, first recorded as ppmv, were converted to mg/m 3 with knowledge of the air pressure and temperature conditions on the day.No allowance for buoyant lift-off was given as methane release at the source was assumed to be passive and diffusive, wherein H = 0, meaning the measured concentration above ambient (C) could be determined having allowed for distance x and angle of the source to the measurement location.As the source location was assumed to be from a pipeline, data from the control survey was analyzed using the same method, but was corrected using Eq. ( 1) to a median line rather than the pipeline.Consequently, methane concentrations corrected to the pipeline should be statistically significantly greater than those corrected to a median control line if the pipeline is a source of methanei.e. the pipeline was hypothesized to have statistically higher concentrations of methane compared to background levels and non-pipeline sources of methane in the same study area.Pipeline and control surveys were also corrected for distance traveled (Table 2).Periods when the survey vehicle was stationary or slow moving led to multiple measurements at one location or in close proximity and thus weighting for distance traveled removed multiple measurements from a given location.

Analysis of variance
A two factor survey design was adopted, with data assessed using analysis of variance (ANOVA).The factorial design and use of ANOVA allowed determination of whether pipelines had a significant impact upon CH 4 fluxes and whether there was variation in flux between study areas.The first factor was the area with four levels (Vale of Pickering, Durham, Northumberland and Vale of Eden), and the second factor was the nature of the source (target) which had two levelspipeline or control.Pipeline and control were replicated across the four study areas.An interaction term between the two factors allowed assessment of significant differences between each pipeline survey and its respective control survey.
The data were Box-Cox transformed to assess for outliers and these were removed if present (Box and Cox, 1964).The data were then tested for normality using the Anderson-Darling test (Anderson and Darling, 1952) and if necessary the data were log-transformed.The Levene test was used to test for the homogeneity of variance.The Tukey test was used post hoc to assess where significant differences lay between factor levels.The proportion of variance explained by factors was assessed by the generalized ω 2 .(Olejnik and Algina, 2003).To avoid type I errors all probability values were assessed as significant if the probability of difference from zero was N95%, but if the probability was close to this value then it is reported.Results are expressed as least squares means as these are better estimates of the mean for that factor level (i.e. the mean for the pipeline or control or individual mean of the four areas) having taken account of the other factors and interactions that were included in the analysis.
The flux from the pipeline and control lines were summed for each area and divided by the distance over which they were collected.Once ambient-corrected (i.e.calculate the flux of CH 4 above ambient), projected to the proposed source (calculate flux coming from direction of pipeline) and then distance corrected (distance from survey line) the fluxes were also calculated once significant sources had been identified.

Soil CH 4 measurements
A detailed, follow-up study of the Vale of Pickering pipeline mobile survey was conducted, with the mobile survey used to determine sections of the Vale of Pickering pipeline that were, a priori, sections where leaks had and had not been identified.Three sections were chosen, two identified as having leaks and one with no identified leak.
The approach used for surveying the sites was based on that of Boothroyd et al. (2016), which measured leaks from abandoned oil and gas wells by comparing soil gas CH 4 concentrations above well pads to those from control fields.For each survey line assessed in this study, an agricultural field containing the NTS pipeline was surveyed, with a neighbouring field of identical land use and soil type used as the control field.In each of the surveyed fields, soil CH 4 measurements were made at equidistant intervals along a transect line.For the pipeline field the survey transect line was followed as close to the pipeline as possible (located by the position of gas company's own field markers).Readings were taken at an approximately 8-m spacing with locations of measurements confirmed by GPS.Due to restrictions on identical land use or crops/animals in the fields, neighbouring fields could not always be used as the control.In these cases, control lines along the far edge of the pipeline field were used, ensuring the greatest distance between the control and survey lines.In total 18 pipeline and 18 associated control fields were surveyed.
Soil CH 4 concentrations were measured in parts per million (ppmv) using an EcoTec TDL-500 portable tunable diode Laser Methane/Gas Analyser with a detection range of 0-10,000 ppmv (Geotechnical Instruments Ltd., Leamington Spa, UK).The measurements were made with a suction cup, connected to the TDL, and placed onto the soil surface for up to 10 sa time based upon the tube delay of the instrument.Measurements were made between 09:30 and 19:30.Prior to each period of data collection the detector was calibrated to a 500 ppmv standard.During soil gas measurement the air temperature, air pressure, relative humidity and dew point were recorded (Commeter C4141 digital Thermo-Hygro-Barometer, Comet System, the Czech Republic).
All data from the survey were considered relative to their control, which were considered as ambient CH 4 conditions under the soil and land use for the weather conditions on the sample day.Pipeline data was ratioed to the average of the CH 4 soil gas concentration for its respective control field and was therefore a relative percentage of the ambient control concentration, i.e. values above 1.0 were interpreted as a leak.
The relative concentration data from the survey was considered as a two-factor ANOVA.The first factor was whether the survey line was a priori considered as containing leaks or not, this factor had two levels (leak or no leak).The second factor was the difference between the survey lines which had 18 levels, one for each survey line measured (i.e. the 18 pipeline and control fields).Data underwent the same treatment as outlined above, with Box-Cox transformation and the normality and Levene tests.The ANOVA was first applied without any covariates and then the ANOVA was repeated using air temperature, air pressure, relative humidity and dew point as covariates -all the covariates were tested for normality and transformed as required.All results from ANOVA are presented as least squares means and post-hoc analysis was performed using Tukey's pairwise comparisons at 95% probability, this was taken as the detection limit within the experimental design which in turn was used to estimate a determination distance, i.e. the maximum horizontal distance along the soil surface for which a significant leak could have been detected.
To assess the magnitude of fluxes for those leaks detected from the soil gas survey the diffusion modeling approach developed by Boothroyd et al. (2016) was used.To model the fluxes from measured leaks, Fick's first law of diffusion was applied.This first law assumes a steady state diffusive flux from a region of high to low concentration proportional to the concentration gradient, which in 2-dimensions can be expressed as: where: J = the diffusive flux (mg CH 4 /m 2 /s); D = diffusion coefficient (m 2 /s); and φ = the concentration of CH 4 in soil (mg CH 4 /m 3 ).Eq. ( 1) was solved assuming that the flux was at steady state over time in 2dimensions using an explicit finite difference method with Δx and Δy = 0.1 m and a distance 3 m either side of the pipe was found to be sufficient to capture the variation back to an ambient concentration: the boundary conditions were chosen such that φ was at the ambient CH 4 concentration as measured for the control field.The pipeline was located at the centre of the base of the grid and the central grid cell was given a concentration equivalent to that in the pipeline at a depth of 1.2 m below soil surface.Firstly, the model was developed fitting the observed values of φ assuming observed values for equivalent to φ at 10 cm depth; the concentration in the pipeline was taken as the maximum value observed in the field measurements; and using D as a fitting parameter.Secondly, the value of D was set based upon the approach proposed by Ridgwell et al. (1999).Initially the flux model fitted the ambient CH 4 soil concentration seen in the control field (φ); the pipeline concentration value was taken as a relative to the ambient, with the fitting parameter the diffusion coefficient (D).The approach used to set the value of D was that proposed by Ridgwell et al. (1999) using the equations: where T soil [K] is the ambient temperature (°C), f the fractional total porosity, f air the fractional air-filled porosity and f clay the fraction of claysized particles present in the soil.The ambient temperature (T soil ) was taken as the average temperature measured on the sampling day by the Thermo-Hygro-Barometer.The value of f clay used (0.3) was taken from Avery (1980) with the soil being a mineral loam soil, standard for the UK having a total porosity of 0.52.The concentration in the pipeline was taken as 100% methane.Using Eqs. ( 2) and (3) the calculated value of D soil = 0.086 cm 2 /s; this single value was used throughout the diffusive modeling process.

Methane peaks and isotopes
Time series plots of each survey are displayed in Fig. 2, showing wind corrected pipeline measurements, and control measurements.The change in the control dataset following wind correction to the median line is also displayed to show the effect of dataset treatment.The Vale of Pickering had 15 pipeline peaks (2.13-2.95ppmv CH 4 ) with a mean of 2.49 ± 0.07 ppmv.There were four peaks on the control (2.11-2.91 ppmv CH 4 ), reduced to three when wind corrected.A peak of 2.48 ppmv was recorded 28 m from the pipeline adjacent to a natural gas processing facility (Fig. 3A) but the isotopic analysis (Vale of Pickering -Pipeline 1, full data −63 ± 6‰ δ 13 C-CH 4 , Fig. S1) was not from the direction of the pipeline and indicated biogenic CH 4 (Table S1).Of the 15 pipeline peaks, the Keeling plot regression was either insignificant in the 30-second average dataset or unsampled for 10 peaks, with biogenic CH 4 at Vale of Pickering -Pipeline 2 (Fig. S2), applicable to five of the measured peaks.One thermogenic signature was identified on the control, Vale of Pickering -Control 1 (−38 ± 3‰ δ 13 C-CH 4 , Fig. S2).
Only one CH 4 peak (2.20 ppmv) was above 2.1 ppmv on the Vale of Eden pipeline route, with an unsampled isotopic signature.Of the eight locations analyzed on the pipeline and control for isotopic composition, one was thermogenic (Vale of Eden -Pipeline 2, −38 ± 1‰ δ 13 C-CH 4, Fig. S2), 34 m from the pipeline.Although excluded as not from the correct wind direction on the pipeline sampling day, the location's isotope data was from the direction of the pipeline and had a maximum concentration of 10.18 ppmv CH 4 in the isotope raw data.The control route had nine peaks (2.13-2.76ppmv CH4, mean 2.25 ± 0.07 ppmv), reduced to five in the wind-corrected dataset.
To summarise, six thermogenic methane sources were identified on the pipeline routes, three of which were associated with peaks above 2.1 ppmv CH 4 from the pipeline sampling data (a peak density of 0.03 thermogenic peaks/km, Table 3).One thermogenic peak was identified on control routes.Thirty-five pipeline peaks were observed at a density of 0.31 peaks/km (Table 3).Excluding peaks identified as biogenic (from farm yards and arable land) on wind corrected pipeline routes, 26 peaks were observed, at 0.23 peaks/km traveled, ranging from 0.05 peaks/km on the Vale of Eden route to 0.76 peaks/km on the Durham route (Table S2).Control routes had 0.09 peaks/km from the full dataset and 0.11 peaks/km from the wind-corrected data.Wind-corrected control routes ranged from 0.04 peaks/km for the Durham route to 0.25 peaks/km for the Vale of Eden route, with two having a greater peak density than their respective pipeline routes.When accounting for the number of peaks observed on the control, the overall peak density from pipelines was 8 peaks at 0.07 peaks/km.Accounting for peaks from wind-corrected control data, the total number of pipeline peaks was 14 at a density of 0.12 peaks/km.

Flux from pipeline survey
Methane flux for each pipeline survey was scaled having accounted for the flux from its respective control survey.The greatest flux was from Northumberland pipeline and the smallest from the Vale of Eden pipeline (Table 4).For two pipelines the IQR included zero, and it may be concluded that there was no flux from these pipelines.
The flux of methane from the seven sites identified with having a thermogenic methane composition was also calculated from the 10 min isotope analytical periods.The pipeline sites had a mean of 33.2 ± 20.7 tonnes CH 4 /yr with a range of 0.1-131.9tonnes CH 4 /yr (Table 5).Included within this was the Durham -Pipeline 4 thermogenic measurement, though this was likely from a mains service/distribution pipeline rather than from the high pressure network and thus the flux calculated to the nearest high pressure pipeline may be inaccurate as a consequence.The flux for the control (103.1 tonnes CH 4 /yr) is greater than the mean for the pipeline fluxes and all but one of the individual measurements.The control flux was calculated to the median control line, a distance of 508 m, but was recorded on a housing estate and most likely represents a natural gas leak from a service/distribution pipeline from a much closer distance than N500 m.Thus, in reality, the control flux is likely to be much smaller than when projected to the median line.If the individual pipeline fluxes are scaled to the 97.5 km (Table 2) from distance corrected datasets, this gives a flux per km range of 1-1352 kg CH 4 /km.When the flux rates are upscaled to the 7600 km of total NTS pipeline, this gives a range of 6.0--10,278.5 tonnes CH 4 /yr across the NTS, with a mean of 2588 ± 1614 tonnes CH 4 /yr.Given however that the six pipeline fluxes represent all potential thermogenic CH 4 detected from the NTS during the survey, the total methane flux would be 199.2tonnes CH 4 /yr at 2043 kg CH 4 /km across the 97.5 km surveyed.When applied to the entire NTS network, the annual flux from all the detected thermogenic fugitive emissions equates to 15,530.4 tonnes CH 4 /yr (12,665.7 tonnes CH 4 /yr excluding Durham -Pipeline 4).The control line applied to the same distance of 7600 km would give 10,901.4tonnes CH 4 /yr but as stated, the flux is inflated by distance to the median line compared to the actual location of the CH 4 source.The flux from the control does reflect uncertainty in emissions from across the natural gas transportation sector, given it is a representation of mains and service distribution fugitive emissions.

ANOVA
Anderson-Darling test showed that log-transformation was sufficient to normalise the data and the Box-Cox transformation showed that only 5 out 12,445 data were removed.The ANOVA of the projected fluxes showed that both factors and the interaction term were significant.The most important factor was the target with the control lines significantly lower than the pipelines, where the least squares mean for pipelines was 2770 ± 84 mg CH 4 /m 3 /s whereas for the control it was 903 ± 46 mg CH 4 /m 3 /s.There were significant differences between all areas with the largest least squares mean being for the Vale of Pickering and the lowest being for Durham.Differences between areas can be  ascribed to differences between days of sampling as well as the differences in the background for each area.The difference between areas does not necessarily represent the differences between the pipelines but this can be estimated from the significant interaction term (Table 6).

Vale of Pickering soil CH 4
In total 1209 soil CH 4 measurements were taken, with 631 CH 4 measurements in pipeline fields and 578 in control fields.The mean value of pipeline soil gas measurements in Pickering was 1.40 ± 0.33 ppmv, with a mean of 1.43 ± 0.38 ppmv for the control fields.The relative concentrations (i.e.all 18 pipeline measurements in a field were made relative to their equivalent control measurement) of CH 4 in Pickering had a mean value of 0.985 ± 0.225, and were normally distributed.Of the relative measurements, 324 out of 631 soil gas measurements were lower than ambient with the smallest relative concentration of CH 4 as 0.341.
The ANOVA showed that all factors were significant.Of the 18 survey lines 8 were significantly greater than the sampling day ambient, 5 significantly lower and 5 with no significant difference with the greatest being 131% above ambient and the smallest value 56% lower than ambient.Using post-hoc analysis, the smallest leak detected was 3% above ambient (1.03 relative concentration).Anything smaller than 1.00 was inferred as no leak and below 1.03 (detection limit) and 1.00, analytically inferred as ambient.Assuming that the smallest detectable leak (3% above ambient) was measured directly over the point source of the leak gives an estimate for the smallest flux detectable by this experimental design in each area.For Pickering, this would be 15.6 kg CH 4 / leak/yr.So as to find over what distance it would have been possible within this experimental design to measure a leak given the detection limit of the equipment it was assumed that the measurement was directly above the leak in the pipeline.Diffusion modeling given this assumption of measurement directly over the pipeline shows that there was a detectable concentration of soil CH 4 concentration up to 5 m away, therefore, the experimental design was capable of measuring a leak 2.5 m either side of the point of measurement on the ground.Taking the determination distance into account with readings every 8 m means 2860 m actual pipe length was surveyed.Given the number of leaks detected (i.e.measurements with relative value above 1.03) and the actual distance of pipeline surveyed (2860 m) then for this pipeline a leak was detected every 9.32 m.The average length of pipeline (between joints) is 10 m (Institution of Gas Engineers and Managers), therefore it can be inferred that this study has detected leaks from all pipeline joints.Fig. 5 shows a linear relationship between the relative CH 4 concentration and the CH 4 flux.The average flux from soil gas CH 4 measurements was 8.24 ± 0.4 kg CH 4 /km/yr.The Pickering pipeline is part of the National Transmission System (NTS) of 7600 km of pipeline; the average flux scaled up for the national pipeline estimates a flux of 62.6 kt CH 4 /yr.
The cross-sectional area from which CH 4 was leaking was estimated using Ramskill's non-choked mass flow equation (Ramskill et al., 1986): where: Q = the mass flow rate (kg/s); C = discharge coefficient, A = discharge hole area (m 2 ), k = C p /C v with C p and C v [L 2 T −2 θ −1 ] the specific heat at constant pressure (p) and volume (v), ρ = real gas density (kg/m 3 ); P A = the atmospheric pressure (Pa); and P = the absolute upstream pressure (Pa).The pressure (P) is taken as 85 bar (8.5 MPa) compared to atmospheric pressure (1 bar = 0.1 MPa).
Using the largest leak detected the cross-sectional area (assuming a uniform hole) was calculated to be 0.0582 mm 2 or 5.82 × 10 −8 m 2 , just larger than the area of a pixel on a modern computer display (0.055 mm 2 ), with a diameter of approximately 0.136 mm.

Discussion
The use of mobile survey equipment enabled the detection of methane leaks from high-pressure natural gas pipelines and associated infrastructure.Across four distinct areas, 26 leaks were detected from 114.6 km, having removed data from wind directions away from the pipeline and not counting confirmed biogenic sources, providing an overall leak detection rate of 0.23/km driven (range 0.05-0.76/km).It is not possible to compare this leak density to other studies in the UK, either of the National Transmission System or mains and service distribution networks due to an absence of research into fugitive emissions from pipelines in the UK.However, studies of mains and service distribution pipelines from the USA have reported leak densities across multiple cities: Durham, NC, 0.14-0.20/km;Cincinnati, OH, 0.29/km (Gallagher et al., 2015); and Ithaca, NY, 0.24/km (Chamberlain et al., 2016) were comparable to the UK NTS leak rate found in this study.Washington, DC, Boston and Manhattan had higher leak densities of 2.44-2.66leaks/km (Gallagher et al., 2015;Jackson et al., 2014;Phillips et al., 2013) and consequently the leak density reported here for the UK NTS is on the low end of those reported from US studies, with even the highest density somewhat short of leakage rates where unprotected steel and cast iron mains were monitored.Most of the leaks reported from the above US studies were classified above 2.5 ppmv CH 4 , though the Ithaca study classified leaks above 1.93 ppmv; as such, the leak density for the UK NTS would be expected to rise relative to this, given that leaks were classified in this study as discrete peaks above 2.1 ppmv.Furthermore, this comparison is undertaken with available studies on leaks from pipelines but distribution pipelines in cities compared with high-pressure pipelines in predominantly rural areas must be considered to be limited as they do not operate under the same pressure.Consequently, it is difficult to place the results of this study in context for transmission systems, particularly in terms of UK emissions given that most studies focus upon USA city mains and service distribution systems that operate at lower pressure and often comprise different construction materials.The soil gas survey of the Vale of Pickering pipeline gives a very different impression of the fluxes from pipelines.This walkover survey was, on average, able to detect a "leak" from every pipe joint.It is perhaps now better to talk not of a leak from a pipeline but that the walkover survey was measuring the in situ properties of the high pressure transmission network.It should be noted that this detection rate was only possible because of the experimental design used by the study, i.e. larger numbers of measurements and only ever judged relative to a control with covariates measured throughout.However, this more detailed and close up survey was able to give a higher estimate of the flux from the network than estimated from the drive by survey.
The 7600 km of NTS of which the Pickering pipeline is a part, is estimated to emit a fugitive CH 4 flux from the entire pipeline of 62.6 kt CH 4 /yr or a CO 2 equivalent of 1570 kt CO 2eq /yr across the whole of the UK.The UK Greenhouse Gas inventory calculates emissions from the six direct GHGs covered under the Kyoto Protocol: carbon dioxide (CO 2 ), methane (CH 4 ), nitrous oxide (N 2 O), hydrofluorocarbons (HFCs), perfluorocarbons (PFCs) and sulphur hexafluoride (SF 6 ).The inventory takes into account fugitive emissions from coal mines, oil and gas upstream processing and solid fuel transformation (DECC, 2014) however does not take into account fugitive emissions from the gas transmission pipelines.The UK GHG inventory values are stated in CO 2eq /yr.The UK 2014 GHG emissions inventory of total GHG was 557,300 kt CO 2eq /yr with the CH 4 contribution being 53,500 kt CO 2eq /yr (DECC, 2016).Even though the UK GHG inventory does not account for fugitive CH 4 emissions, the emissions calculated in this study (1570 kt CO 2eq /yr) represents an additional 2.9%.However, this study only considered the high pressure transmission system in the UK and therefore, emissions from the rest of the transmission and distribution system all the way to the customer would have to be better accounted for in the future.

Conclusions
This study used two approaches to the measurement of CH 4 emissions from high pressure gas pipelines (70 to 85 bar).Both approaches used demonstrated significant emissions from pipelines relative to background control.i) Leak rate from a mobile pipeline survey was 627 (241-1123 interquartile range) tonnes CH 4 /km/yr.The flux from thermogenic CH 4 sources was 199.2 tonnes CH 4 /yr across 97.5 km surveyed.Scaled up to the NTS, confirmed thermogenic fluxes amount to 15.5 kt CH 4 /yr.ii) A walkover survey of soil gas CH 4 found that it was possible to detect elevated CH 4 for every pipeline joint.Scaling results for the entire UK national transmission system showed a pipeline emission of 62.6 kt CH 4 /yr or a CO 2 equivalent of 1570 kt CO 2eq /yr across the whole of the UK which is 2.9% of total annual CH4 emissions.iii) Further research is required into the scale of fugitive emissions from pipeline infrastructure in the UK.Transmission and distribution stations are known to be sources of CH 4 , while little research has been conducted in recent years on distribution pipeline emissions, beyond industry surveys.This study has reported potential emissions from transmission stations and distribution pipelines and would recommend further work to better quantify their impact on GHG emissions.

•
Fugitive CH 4 emission from UK National Transmission System high pressure pipeline • Pipelines had 26 peaks of CH 4 above 2.1 ppmv, control routes 18 peaks.• Six thermogenic CH 4 sources from pipelineannual flux of 15.85 kt CH 4 /yr • Soil CH 4 survey detected leaks from joints in-situ.CH 4 /yr, 2.9% of UK CH4 inventory.G R A P H I C A L A B S T R A C T a b s t r a c t a r t i c l e i n f o 1. Introduction

Fig. 1 .
Fig. 1.Map of study pipeline and control survey routes.Letters A-D refer to panels in Fig. 3.

Fig. 2 .
Fig. 2. Time series methane concentrations for Vale of Pickering, Durham, Northumberland and Vale of Eden pipeline surveys.Wind corrected pipeline, complete control and wind corrected control surveys displayed.Inset wind corrected control plots are at the same scale as the complete control time series as way of comparison to show where data has been retained or removed.Times are GMT.Start times for each survey differ due to different travel distances on day of sampling; and Durham, Northumberland and Vale of Eden sampling was conducted during British Summer Time (GMT + 1).

Fig. 5 .
Fig. 5. Pickering CH 4 flux calculations.The measured CH 4 soil gas concentration measurement relative to the its control (relative CH 4 concentration) for each leak observed in Vale of Pickering study area in comparison to the CH 4 flux for each the leak as calculated by diffusion modeling.

Table 1
High-pressure transmission pipeline network traversed during study.Year Comm = year commissioned.MOP = maximum operating pressure.

Table 2
Sample size (n) and distance traveled (km) for distance corrected datasets.

Table 3
The number of peaks N2.10 ppmv CH 4 observed from pipeline and control routes.Pipeline peaks disseminated into pipeline minus biogenic and pipeline minus biogenic & full control peaks.Note distances are different to flux calculations that remove ambient measurements.

Table 4
Pipeline flux having accounted for control routes.IQR = inter-quartile range.

Table 5
Methane flux from thermogenic CH 4 sources.

Table 6
The least squares means of the target and area * target terms.