Characterization of interferences to in situ observations of δ 13 CH 4 and C 2 H 6 when using a cavity ring-down spectrometer at industrial sites

Due to increased demand for an understanding of CH4 emissions from industrial sites, the subject of cross sensitivities caused by absorption from multiple gases on δCH4 and C2H6 measured in the near-infrared spectral domain using CRDS has become increasingly important. Extensive laboratory tests are presented here, which characterize these cross sensitivities and propose corrections for the biases they induce. We found methane isotopic measurements to be subject to interference from elevated C2H6 concentrations resulting in heavier δCH4 by +23.5 ‰ per ppm C2H6 / ppm CH4. Measured C2H6 is subject to absorption interference from a number of other trace gases, predominantly H2O (with an average linear sensitivity of 0.9 ppm C2H6 per % H2O in ambient conditions). Yet, this sensitivity was found to be discontinuous with a strong hysteresis effect and we suggest removing H2O from gas samples prior to analysis. The C2H6 calibration factor was calculated using a GC and measured as 0.5 (confirmed up to 5 ppm C2H6). Field tests at a natural gas compressor station demonstrated that the presence of C2H6 in gas emissions at an average level of 0.3 ppm shifted the isotopic signature by 2.5 ‰, whilst after calibration we find that the average C2H6 : CH4 ratio shifts by+0.06. These results indicate that, when using such a CRDS instrument in conditions of elevated C2H6 for CH4 source determination, it is imperative to account for the biases discussed within this study.


Introduction
With increasing efforts to mitigate anthropogenic greenhouse gas emissions, opportunities to reduce leaks from fossil fuel derived methane (ffCH 4 ) are of particular importance as they currently account for approximately 30 % of all anthropogenic methane emissions (Kirschke et al., 2013).At present, technically feasible mitigation methods hold the potential to half future global anthropogenic CH 4 emissions by 2030.Of this mitigation potential more than 60 % can be realized in the fossil fuel industry (Hoglund-Isaksson, 2012).However, for effective implementation, sources, locations and magnitudes of emissions must be well known.
The global increase in the production and utilization of natural gas, of which methane is the primary component, has brought to light questions in regards to its associated fugitive emissions, i.e. leaks.Recent estimates of CH 4 leaks vary widely (1-10 % of global production; Allen, 2014) and US inventories of natural gas CH 4 emissions have uncertainties of up to 30 % (US EPA, 2016).In addressing this issue, the ability to distinguish between biogenic and different anthropogenic sources is of vital importance.For this reason methane isotopes (δ 13 CH 4 ) are commonly used to better understand global and local emissions, as demonstrated in a number of studies (Lamb et al., 1995;Lowry et al., 2001;Hiller et al., 2014).The discrimination of sources with relatively close isotopic composition such as oil-associated gas and natural gas, which can have isotopic signatures separated by only ∼ 4 ‰ (Stevens and Engelkemeir, 1988), requires precise and reliable δ 13 CH 4 measurements.Ethane (C 2 H 6 ) is a secondary component in natural gas and can be used as a marker to distinguish between different CH 4 sources.Use of the C 2 H 6 : CH 4 ratio provides a robust identifier for the gas of interest.Recent findings in the US found coal bed C 2 H 6 : CH 4 ratios ranging between 0 and 0.045, while dry and wet gas sources displayed differing ratios of < 0.06 and > 0.06 respectively (Yacovitch et al., 2014;Roscioli et al., 2015).
Laser spectrometers, especially those based on cavity ringdown spectroscopy (CRDS), are now a common deployment for site-scale CH 4 measurement campaigns (Yvon-Lewis et al., 2011;Phillips et al., 2013;Subramanian et al., 2015).However, with the advent of such novel technologies, there is a risk of unknown interference from laser absorption which can create biases in measurements.Some examples of this are discussed in Rella et al. (2015) and many others (e.g.Malowany et al., 2015;Vogel et al., 2013;Nara et al., 2012).Using a CRDS instrument we show that the presence of C 2 H 6 causes significant interference to the measured 13 CH 4 spectral lines, thus resulting in shifted reported δ 13 CH 4 values.We propose a method to correct these interferences and test it on measurements of natural gas samples performed at an industrial natural gas site.
The CRDS instruments used throughout this study are Picarro G2201-i analysers (Picarro INC, Santa Clara, USA) which measure gases including CH 4 , CO 2 , H 2 O, and, although not intended for use by standard users, C 2 H 6 .This model measures in three spectral ranges: lasers measuring spectral lines at roughly 6057, 6251 and 6029 cm −1 are used to quantify mole fractions of 12 CH 4 , 12 CO 2 and 13 CO 2 , and 13 CH 4 , H 2 O and C 2 H 6 respectively.The spectrograms are fit with two non-linear models in order to determine concentrations; the primary fit excludes the model function of C 2 H 6 while the second includes this function, thus adding the ability to measure C 2 H 6 (Rella et al., 2015).Such a method for measuring C 2 H 6 concentrations is crude, thus the uncalibrated C 2 H 6 concentration data are stored in private archived files which until now have been used primarily for the detection of sample contamination.The measurements of δ 13 CH 4 and δ 13 CO 2 are calculated using the ratios of the concentrations of 12 CH 4 , 13 CH 4 , 12 CO 2 and 13 CO 2 respectively.
An experimental procedure is presented here which corrects the interference caused by C 2 H 6 on the retrieval of Therefore an open split is included to vent additional gas and retain ambient pressure at the inlets.
δ 13 CH 4 using such a CRDS instrument for application to in situ or continuous measurements of δ 13 CH 4 strongly contaminated by C 2 H 6 , i.e. in the vicinity of ffCH 4 sources.The step-by-step procedure of the experimental methods developed to quantify the cross sensitivities and the proposed calibration for δ 13 CH 4 and C 2 H 6 are depicted in Fig. 1 and presented in detail in Sect. 2. Section 3 encompasses a discussion of the results, including an analysis of the instrumental responses for two spectrometers with an evaluation of the stability and repeatability of the suggested corrections.Finally, field measurements were performed at a natural gas compressor station where the aim was to identify emissions between two natural gas pipelines.In Sect. 5 the importance of the corrections for field measurements is demonstrated by applying our methods to data retrieved during this period while also revealing the instruments' potential to measure C 2 H 6 .

Methods
The purpose of laboratory tests was to characterize the instruments' response to concentration changes in gases found at fossil fuel sites (e.g.gas extraction or compressor stations), specifically, the cross sensitivities of CO 2 , CH 4 and H 2 O on C 2 H 6 and of C 2 H 6 on δ 13 CH 4 .Presumably there are additional gases with the potential for interference; this study focuses on those reported to have a significant effect on C 2 H 6 and δ 13 CH 4 measurements by Rella et al. (2015).We also define and describe a new procedure to calibrate both C 2 H 6 and δ 13 CH 4 .
In the following chapter the general set-up used for the majority of experiments is described, after which we enter a more detailed description of the processes involved in each step.
of 52 ppm in nitrogen 2.1 Experimental set-up

Method
Each cross sensitivity is measured by creating a gas dilution series designed to control the concentrations of the gas responsible for the interference in steps while keeping concentrations of the other gas components constant (in particular the component subject to interference).The instrument response was evaluated for a large range of concentrations and different combinations of gas components.An example of such a measurement time series can be seen in Fig. S1 in the Supplement.The experimental set-up used includes two CRDS instruments (Picarro G2201-i) running in parallel in a laboratory at ambient conditions (25 • C, 100 m above sea level; a.s.l).The instruments were used in iCO 2 -iCH 4 auto switching mode, of which we consider only the "high precision" mode of δ 13 CH 4 throughout the study.For the dilution series, a working gas is diluted in steps using a set-up of two mass flow controllers (MFC; El-flow, Bronkhorst, Ruurlu, the Netherlands), as shown in Fig. 2. A T-junction splits the gas flow to both instruments; the total flow is greater than the flow drawn into the instruments.Hence to maintain an inlet pressure close to ambient, the set-up includes an open split to vent additional gas.In order to assess variability and error, each experiment is repeated a minimum of three times consecutively.To detect instrumental drift between experiments, a target gas is measured before commencing each dilution sequence.An overview of each targeted cross interference, with information on the gases used and ranges spanned in laboratory tests, can be found in Table 1.

Gases
Throughout the experiments, four categories of gas were used: a zero air gas with measured residual concentrations of To alter the water vapour content of a sample, the experimental set-up described in Fig. 2 was modified by incorporating a humidifier.
The humidifier consists of a liquid flow controller (Liquiflow, Bronkhorst, Ruurlu, the Netherlands) and a mass flow controller (El-flow, Bronkhorst, Ruurlu, the Netherlands) fed into a controlled evaporator mixer (CME) (Bronkhorst, Ruurlu, the Netherlands).The tube departing the CME contains a gas flow of 2 L min −1 and is heated to 40 • C to prevent any condensation.A short description and diagram of the humidifying bench can found in Laurent et al. (2015).
The H 2 O interference on C 2 H 6 was measured by using the humidifier to vary the H 2 O content of zero air gas in the range of 0.25-2.5 % H 2 O, representing the range of real world conditions.The humidifier set-up cannot reliably reach humidity below 0.2 % H 2 O, a range frequently reached when measuring gas cylinders or dried air.This low range was attained using a H 2 O scrubber (Magnesium Perchlorate, Fisher Scientific, Loughborough, UK) connected to the CRDS instrument inlet while measuring ambient air.As the efficiency of the scrubber decreases over time, a slow increase of H 2 O spanning low concentrations in the range of 0-0.5 % can be observed.
The CH 4 interference on C 2 H 6 was measured by creating a dilution series of variable CH 4 content using zero air and a working gas of 6 ppm CH 4 , 360 ppm CO 2 , 310 ppb N 2 O and 50 ppb CO in natural air.Methane concentrations ranged from 0 to 6 ppm.To keep other causes of interference at a minimum, the gas mixture passed through two scrubbers: the first a CO 2 scrubber (Ascarite(ii), Acros Organics, USA) and the second a H 2 O scrubber (Magnesium Perchlorate, Fisher Scientific, Loughborough, UK).As an independent check on the linearity of the response functions, each dilution sequence was repeated at two humidities (0 % H 2 O and 1 % H 2 O) and four C 2 H 6 concentrations (between 0 and 1.5 ppm).
The CO 2 interference on C 2 H 6 was measured with a dilution series ranging 0-1500 ppm CO 2 created by mixing zero air and a working gas of 2000 ppm CO 2 , 1.7 ppm CH 4 and 50 ppb CO in natural air.Any interference due to CH 4 was accounted for during data processing.This test was repeated at four water vapour levels (0, 0.5, 1 and 1.5 %) and five C 2 H 6 concentrations (between 0 and 2.5 ppm).

C 2 H 6 calibration set-up
In order to correctly use the C 2 H 6 data from CRDS instruments, the data must be calibrated to an internationally recognized scale.To achieve this, the set-up described in Sect.2.1 was modified to include the filling of removable samples (1 L glass flasks), the concentrations of which could be independently verified, as shown in Fig. 2. A gas mixture using the C 2 H 6 standard and an ambient air cylinder was created via two MFCs before passing through the flask on its way to the instruments' inlets.Each step in the dilution series requires an individual flask, which was flushed for 20 min and then analysed for 10 min with an average precision of 0.02 ppm C 2 H 6 on the CRDS instrument.The flask is subsequently sealed and removed for analysis on a gas chromatograph (GC) (Chrompack Varian 3400, Varian Inc, USA) which uses National Physics Laboratory (NPL) standards and has an uncertainty better than 5 %.The system is described in more detail in Bonsang and Kanakidou (2001).
In total 17 flasks were filled with gas mixtures spanning from 0 to 5 ppm C 2 H 6 , covering the range expected near a leak of ffCH 4 (Gilman et al., 2013;Jackson et al., 2014).In order to calibrate the linearity of the response at very high concentrations which may be expected from pure natural gas samples, we conducted a measurement at 100 % of the C 2 H 6 standard (52 ppm ± 1 ppm).

Determining the correction for δ 13 CH 4
Measured δ 13 CH 4 is altered in the presence of C 2 H 6 .To understand the magnitude of this effect, experiments were conducted using the method described in Sect.2.1.The dilution series uses the C 2 H 6 standard and a cylinder filled with ambient air, i.e. with a negligible C 2 H 6 mixing ratio (< 1 ppb), to create concentration values spanning from 0 to 4 ppm C 2 H 6 .As there is only one source of CH 4 in the experiment, the addition of C 2 H 6 should not affect the value of δ 13 CH 4 ; hence any change seen is an apparent shift of δ 13 CH 4 due to C 2 H 6 interference.This concentration range was chosen as it encompasses a C 2 H 6 : CH 4 ratio of 0 to 1, well within the likely range to be measured from fossil fuel sources (Yacovitch et al., 2014).

Calibration of δ 13 CH 4
The reported δ 13 CH 4 was calibrated to Royal Holloway University of London (RHUL) scale using four calibration gases spanning −25 to −65 ‰ that were created by different dilutions of pure CH 4 and CO 2 with ambient air.The aliquots were measured multiple times by isotope ratio mass spectrometry (IRMS) at RHUL.The precision for δ 13 CH 4 , obtainable with this IRMS, is reported as 0.05 ‰ -detailed information on the measurement system can be found in Fisher et al. (2006).The calibration factor is determined from a linear regression and calibrations were performed once a day for 3 consecutive days before and after the laboratory experiments.A target gas was measured regularly to track any drift in δ 13 CH 4 as an independent check on the calibration quality.

Results and discussion
This study focuses on determining a reliable correction and calibration scheme for a Picarro G2201-i when measuring methane sources with C 2 H 6 interference.Findings from the experiments described in Sect. 2 are discussed in detail here.
In order to calibrate δ 13 CH 4 and C 2 H 6 values, there are a series of corrections that must take place beforehand (see Fig. 1).The initial correction to be applied is on C 2 H 6 due to interference from CH 4 , CO 2 and H 2 O. Particular emphasis is placed on this correction due to the discovery of significant non-linear behaviour in the presence of H 2 O, CH 4 and CO 2 in the sample gas.Once the C 2 H 6 has been corrected, the calibration of C 2 H 6 using independent GC measurements, the C 2 H 6 interference correction on δ 13 CH 4 and finally the calibration of δ 13 CH 4 can be effected.
For our results to be applicable to future studies we examine the inter-instrument variability and stability over time, compare our results to current literature and discuss the uncertainties attributed to our results.Throughout this study we refer to raw, uncorrected C 2 H 6 and δ 13 CH 4 concentrations as "reported" to highlight that they may be influenced by interferences and are uncorrected.Within this section negative C 2 H 6 concentrations are often mentioned.We note that this is the "reported" C 2 H 6 concentration by the instrument.Unless otherwise stated, the standard deviation reported is calculated from 1 min averages and depicted as error bars within figures.

Correcting reported C
H 2 O content was found to be the dominating source of interference to reported C 2 H 6 ; its presence decreases the reported concentration of C 2 H 6 with increasing H 2 O concentration.Furthermore, the response function exhibits a hysteresis effect, which, although small, can be considerable when changing from dry to undried air samples (e.g. between dry calibration gas and undried ambient air).There are two distinct instrumental responses, depending on whether dried or undried ambient air are being measured during the night preceding the experiment, which are depicted in Figure 3 by dark and light blue markers respectively.When the CRDS instrument measures dry air prior to the experiment, a discontinuity is observed at 0.16 % H 2 O. Figure 4 shows this effect in more detail; prior to 0.16 % H 2 O the response function exhibits a stable linear response.The correction within this low range was found to be the same for both instruments, 0.44 ± 0.03 ppm C 2 H 6 / % H 2 O.After passing the 0.16 % H 2 O threshold, the response exhibits a discontinuity with a magnitude and subsequent slope that are also dependent on the air moisture beforehand.This is seen in Fig. 4 whereby the discontinuity of two repetitions (A and B depicted by dark and light blue markers respectively) differs in magnitude by 0.1 ppm reported C 2 H 6 .The discontinuity occurs when the instrument passes the 0.16 % H 2 O threshold, both when moving from dry to wet air and vice versa (see Fig. S2).If measuring undried air before the experiment, the interference due to H 2 O can be described well by a linear response (light blue markers in Fig. 3) and potentially causes large biases from the true C 2 H 6 .For example, if measuring at 1 % H 2 O, both instruments display a change in reported C 2 H 6 of approximately −0.9 ppm.The response function calculated for instruments CFIDS 2072 and 2067 differed, showing −0.72 ± 0.03 ppm C 2 H 6 / % H 2 O and −1.00 ± 0.01 ppm C 2 H 6 / % H 2 O with R 2 values of 0.98 and 0.99 respectively.The hysteresis effect is evident when measuring with undried air; the slope was seen to shift after each repetition, in total by 0.1 ppm C 2 H 6 / % H 2 O.

CO 2 interference on C 2 H 6
For both instruments an increase in the CO 2 concentration results in lower reported values of C 2 H 6 , and it is furthermore apparent that the magnitude of this interference is dependent on air humidity.For a dry sample gas (H 2 O < 0.16 % -demonstrated in the left-hand column of Fig. 5), the interference for both instruments is found to be highly stable and well characterized by a linear slope of 1 × 10 −4 ± 1 × 10 −5 ppm C 2 H 6 / ppmCO 2 with a R 2 value of 0.9.There was no measurable difference in slope at any of the C 2 H 6 concentrations tested (see Fig. S3).In contrast, for water vapour levels ≥ 0.5 % H 2 O (see right- hand column of Fig. 5), measurements exhibit a higher scatter between repetitions.This is mainly attributed to a drifting intercept; however the experiments also show a smaller R 2 of 0.8.We calculate a characteristic linear slope of 3.8 × 10 −4 ± 1 × 10 −5 ppm C 2 H 6 / ppm CO 2 and 3.9 × 10 −4 ± 1 × 10 −5 for ≥ 0.5 % water vapour for instruments CFIDS 2072 and 2069 respectively.Therefore, when measuring undried ambient air, the presence of CO 2 at a level near 400 ppm will induce a shift in the reported C 2 H 6 of approximately −0.15 ppm C 2 H 6 , whereas if the air is dried the reported shift is much smaller, at approximately −0.04 ppm C 2 H 6 .

CH 4 interference on C 2 H 6
The CH 4 effect on C 2 H 6 , as shown in Fig. 6, is less prominent by at least an order of magnitude than both the H 2 O and CO 2 interferences.At dried ambient CH 4 concentrations a typical change in reported C 2 H 6 of approximately −0.008 ppm is observed within both instruments.Dried air experiments show a high scatter of points between repetitions, and R 2 values of 0.4 and 0.6 for instruments CFIDS 2072 and 2067 respectively are calculated.Despite its large uncertainty, the data suggest that both instruments display a similar response with a statistically significant slope within the range of C 2 H 6 concentrations tested (see Fig. S3).In light of this we use a weighted mean to calculate a linear response of 9 × 10 −3 ± 2 × 10 −3 ppm C 2 H 6 / ppm CH 4 for dry air measurements for CFIDS 2067, and ± 5 × 10 −3 ppm C 2 H 6 / ppm CH 4 for CFIDS 2072.The results obtained at 1 % H 2 O show little correlation (as shown in the right-hand column of Fig. 6), with both instruments displaying a R 2 value of 0.2.An ANOVA test suggests the slopes are not significantly different from zero; thus we omit a CH 4 correction for this case.

Combining the CO 2 , CH 4 and H 2 O correction on C 2 H 6
To fully take into account all (known) C 2 H 6 crosssensitivities, the corrections to reported C 2 H 6 need to be combined.Due to the non-linearity of the discontinuity in reported C 2 H 6 at 0.16 % H 2 O and its subsequent slope we choose to report correction coefficients for the two found linear regimes, i.e. for continuous measurements with sample humidities below 0.16 % and sample humidities above 0.16 %.Within each range the proposed correction formula is given as follows: If the humidity is limited to less than 0.16 % before and during measurements, Both instruments demonstrated good agreement for all the correction factors calculated at < 0.16 % H 2 O.
Corrections for measurements undertaken at concentrations higher than or equal to 0.16 % H 2 O are A = ± 2 × 10 −5 ppm C 2 H 6 / ppm CO 2 for CFIDS 2067.

C 2 H 6 calibration
To make use of the corrected C 2 H 6 it should be calibrated to match an internationally recognized scale.This is achieved by measuring whole-air samples by CRDS and independently on a calibrated gas chromatograph, as discussed within Sect. 2. The calibration factor is determined by comparing the corrected C 2 H 6 resulting from CRDS and C 2 H 6 as confirmed by the GC and plotted in Fig. 7a.The relationship was found to be linear throughout the range of 0-5 ppm C 2 H 6 with a slope of 0.505 ± 0.007 and 0.52 ± 0.01 for instruments CFIDS 2072 and 2067 respectively.The results are reported in Table 2 from which we can see the intercept of the calibration for instrument CFIDS 2072 shifts between the experiment in February and that in October, while the slope remains constant throughout the measured time period.The change in the intercept is attributed to a C 2 H 6 baseline drift which we have monitored over time using regular target gas measurements; an example is given in Fig. 7b.To account for this drift and any elevated baselines (such as that of CFIDS 2067 -see Table 2), a regular measurement of a working gas is necessary, from which the instrument offset can be calculated.For the full calibration, we thus suggest using Eq. ( 2), Figure 8.During a dilution sequence of ambient gas with C 2 H 6 , the CH 4 concentration decreases from its nominal concentration 1948.7 ppb ± 0.32 ,ppb as the contribution from C 2 H 6 is increased.Thus both 12 CH 4 and 13 CH 4 undergo a similar decrease as the gas is diluted.However, what is observed is an increase in the reported value of 13 CH 4 , suggesting C 2 H 6 interference.The 12 CH 4 axis is plotted to the left in light green, whereas the 13 CH 4 axis is plotted to the right in dark green at a different scale.Error bars represent the standard deviation, the 12 CH 4 markers are larger than their associated error bars.
where D is the calibration factor (slope) for the instrument, i.e. for CFIDS 2072 D = 0.505 ± 0.007 and (WGS) the baseline drift determined using the working gas. (2)

δ 13 CH 4 correction
By measuring the shift of the reported δ 13 CH 4 in C 2 H 6contaminated samples, we have observed that the instrument reports heavier values of δ 13 CH 4 in the presence of C 2 H 6 .
The shift is a result of increased reported 13 CH 4 in samples containing C 2 H 6 (see Fig. 8).This is most likely caused by the overlapping of spectral lines within the 6029 wave number region (Rella et al., 2015).We calculate the δ 13 CH 4 correction by taking the slope of δ 13 CH 4 (the difference between the reported δ 13 CH 4 and the initially reported one of the C 2 H 6 -free gas) and the corrected C 2 H 6 to CH 4 ratio.
The ratio is used to permit the calculation of the δ 13 CH 4 response function per ppm CH 4 as the magnitude of interference is dependent on CH 4 concentration (Rella et al., 2015).
The significance of the interference on δ 13 CH 4 concentrations is illustrated in Fig. 9; as the C 2 H 6 : CH 4 ratio increases, the change in the reported δ 13 CH 4 increases linearly.Results obtained from tests carried out throughout the year, for both instruments are noted in Table 3 and plotted in Fig. 9.The correction equation can be expressed as follows: These corrections contain the inherent δ 13 CH 4 offset of the instrument.When calibrating the δ 13 CH 4 to a known scale (as described in Sect.2.5) any instrumental offset will be incorporated within the calibration.Therefore, the correction equations can be simplified to Also highlighted in Fig. 9 is the typical measurement range for the majority of ffCH 4 sources related to dry and wet natural gas relative to calibrated C 2 H 6 / CH 4 ratios given on the upper abscissa, whereby dry gas refers to natural gas that occurs in the absence of condensate/liquid hydrocarbons (C 2 H 6 : CH 4 = 1-6 %) while wet gas typically contains higher concentrations of complex hydrocarbons (C 2 H 6 : CH 4 > 6 %; Yacovitch et al., 2014).It is clear that within this range the bias on methane isotopic signatures is significant; dry gas will alter the reported δ 13 CH 4 by 0.8-4 ‰, while wet gas can cause a shift of up to 13 ‰ depending on its C 2 H 6 : CH 4 ratio.

δ 13 CH 4 calibration
Full instrument calibrations as described in Sect.2.4 were performed once in 2014 and once in 2015.The δ 13 CH 4 values obtained for the calibration gases by RHUL are measured by IRMS and are therefore not subject to interferences.The calibration gas aliquots were measured with an average standard deviation of 0.03 ‰.To calibrate δ 13 CH 4 CORRECTED , the δ 13 CH 4 CORRECTED was calculated for each calibration gas and used within the linear regression.The calibrations were linear with R 2 > 0.99 on both occasions and no change (within our uncertainties) was observed between the two tests.By measuring an ambient air target regularly, we later detected a shift in the δ 13 CH 4 baseline.Two further calibrations were performed in 2016 to assess this incident which confirmed that the offsets of the linear regressions were significantly shifted, while the slopes agreed well with previous calibrations.Therefore, to account for a baseline drift, it is important to measure a target gas regularly and amend the offset of the calibration equation accordingly.

Typical instrumental performance and uncertainties
In order to characterize the repeatability of the C 2 H 6 measured by the CRDS instrument, we have measured several targets and monitored the changes of the reported C 2 H 6 signal over time.The raw signal is a measurement every 3 s, which displays on average a standard deviation of 90 ppb.By aggregating the data to 1 or 30 min intervals, the precision can be improved and a standard deviation of 20 or 8 ppb is reached.Furthermore, the 1 min standard deviation at 52 ppm C 2 H 6 is 180 ppb.Thus by assuming a linear relationship the typical performance for 1 min averages is 20 ppb ±0.3 % of reading.
Of course, there are some substantial uncertainties attributed with the C 2 H 6 correction and calibration which need to be accounted for when discussing the uncertainty of the calibrated C 2 H 6 concentrations.With regards to the C 2 H 6 correction for 1 min averages, if measuring dried ambient air the propagation of uncertainties are negligible with respect to the raw instrumental precision (20 ppb).However, if using 30 min averages the uncertainty augments from 8 to 10 ppb.Elevated CH 4 , CO 2 and H 2 O signals (> 5 ppm, > 1000 ppm, > 0.2 % respectively) will induce increased C 2 H 6 uncertainty regardless of aggregation time.After calibration, the correction factor increases to 2 1/2 times that of the corrected C 2 H 6 , so at ambient air concentrations calibrated C 2 H 6 has an uncertainty of 30 ppb.
The repeatability of δ 13 CH 4 for 1 min averages on our instrument is a standard deviation of 0.66 ‰.The standard deviation is reduced to 0.29 and 0.09 ‰ by aggregating the raw data for 5 and 30 min respectively.For the correction of δ 13 CH 4 due to C 2 H 6 , error propagation of the factors applied in Eq. ( 4) must be taken into account.Therefore, at ambient concentrations, the uncertainty of a 1 min average will increase to 0.9 ‰.

Generalizability of corrections and calibrations
The experiments in this study were repeated multiple times and performed on two instruments to better understand how the instrument responses change over time and how they vary between instruments.The C 2 H 6 correction and calibration, and δ 13 CH 4 correction experiments were repeated on CFIDS 2072 over the course of a year to determine any temporal drifts.
The coefficients of the C 2 H 6 correction were examined over a 4-month period.Methane, carbon dioxide and water vapour coefficients for dried gas displayed no noticeable variation over this time frame.Both CH 4 and CO 2 coefficients for undried gas also showed good stability throughout this period; however the undried H 2 O coefficient is seen to vary significantly (±0.1 ppm C 2 H 6 / % H 2 O).As discussed previously, the H 2 O correction is subject to a hysteresis effect, which makes analysis of its long-term variation difficult.As we did not find a clear temporal pattern of the variations, we therefore suggest that this coefficient is not likely to be time dependent.
The calibration of C 2 H 6 was calculated twice within a 9month period (see Table 2).No variation of the slope of the response function is observed within this time frame.The intercept is prone to drift in time as discussed previously.
The δ 13 CH 4 correction has been examined three times throughout a 6-month period (see Table 3).The variability of the slope observed over 6 months is 1 ‰ ppm C 2 H 6 / ppm CH 4 .Given that the error attribution of each experiment is approximately ±1 ‰ ppm C 2 H 6 / ppm CH 4 , this variability is not statistically significant.The intercepts show good agreement with no variation outside the expected uncertainties.
The comparison of both CRDS instruments showed good agreement for all calculated C 2 H 6 correction coefficients, with the exception of the undried H 2 O coefficient at > 0.16 % H 2 O.For this coefficient we calculate a difference of 0.3 ppmC 2 H 6 / % H 2 O between that of CFIDS 2072 and CFIDS 2067.The variance may be the consequence of spectrometer differences, a long-term hysteresis effect or differences in their past use (mostly dried samples on CFIDS 2072 and mostly undried samples for CFIDS 2067).
The slopes derived for the C 2 H 6 calibration of both instruments correspond well, with no significant difference seen between the two.The intercepts differ by approximately 0.6 ppm, thus suggesting a distinct difference between intrainstrumental C 2 H 6 baselines.
The slopes of the δ 13 CH 4 correction were found to be in good agreement between the two instruments.Where the instruments differ is with regards to their δ 13 CH 4 baselines, thus causing the observed disparity in intercept (seen in Table 3) of approximately 3 ‰.
To the best of our knowledge, at this time there is only one published study reporting on a correction due to C 2 H 6 interference on an isotopic Picarro analyser.Rella et al. (2015) have studied the interference using a Picarro G2132-i, a high-precision CH 4 isotope-only CRDS analyser which uses similar analysis algorithms and spectral regions to that of the Picarro G2201-i.Rella et al. (2015) obtained C 2 H 6 correction parameters of A = 0.658 ppm C 2 H 6 / ppm H 2 O, B = 5.5 ± 0.1 × 10 −3 ppm C 2 H 6 / ppm CH 4 , C = 1.44 ± 0.02 × 10 −4 ppm C 2 H 6 / ppm CO 2 in 2015.Factors B and C for CH 4 and CO 2 respectively agree well with the dried air coefficients attained within this study.The H 2 O coefficient, as suggested by Rella et al. (2015) differs from both that of CFIDS 2072 and CFIDS 2067 but confirms the variability of this factor between instruments when measuring undried air samples.Lastly, Rella et al. (2015) report a correction factor for δ 13 CH 4 of 35 ‰ ppm CH 4 / ppm C 2 H 6 which indicates a different response to C 2 H 6 contamination of the different instrument series.
4 Source identification at a natural gas compressor station In order to quantify the effect of C 2 H 6 contamination in a real world situation, we have applied the corrections and calibrations discussed in this paper to measurements taken at a natural gas site, with the aim of distinguishing emissions between two natural gas pipelines.In the following section we demonstrate the effect of C 2 H 6 interference on δ 13 CH 4 at a fossil fuel site and discuss the alternative approach of using calibrated C 2 H 6 : CH 4 ratios to distinguish source signatures, a method which has not been previously tested on a Picarro G2201-i.

Site description
Located in an industrial park in northern Europe, the campaign took place at a natural gas compressor station in summer 2014.Such stations serve the distribution of natural gas; their key purpose is to keep an ideal pressure throughout the transmission pipelines to allow continuous transport from the production and processing of natural gas to its use.The visited compressor site comprises two major pipelines with their corresponding compressors.The two pipelines carry gas of different origins to the site, where after pressurization, they are combined for further transmission.The site topography is flat and open with the surrounding area being predominantly farmland and in close proximity to a major road.FFCH 4 emissions were expected to emanate from various sources on site such as the compressors, methane slip from turbines and fugitive emissions due to the high pressure of gas (Roscioli et al., 2015).Other possible methane sources in the nearby region were identified as traffic and agriculture, including a livestock holding situated less than 500m south-west of the site.

Continuous measurements of CH
Two instruments were utilized for continuous measurements throughout the 2-week field campaign: a CRDS instrument (CFIDS 2072, characterized in detail in previous sections) and an automatic gas chromatograph with a flame ionization detector (GC-FID; Chromatotec, Saint-Antoine, France) measuring VOCs (light fraction C 2 -C 6 hydrocarbons), described in detail in Gros et al. (2011).They were located at a distance of approximately 200-400 m from the pipelines and compressors.
The air measured by the CRDS instrument was dried consistently to < 0.16 % H 2 O using a Nafion (Perma Pure LLC, Lakewood, USA).The δ 13 CH 4 was calibrated using the method described previously in Sect. 2. Every two days, 20 min measurements of two calibration gases were made to calibrate the CH 4 and CO 2 data and to track any drift in the isotopes.A C 2 H 6 free working gas was measured every 12 h and used simultaneously as a target gas for the calibration of CH 4 and CO 2 , and to track any drift in the C 2 H 6 baseline for the calibration of C 2 H 6 .
The GC-FID was calibrated at the beginning and end of the campaign using a certified standard gas mixture (NPL, National Physics Laboratory, Teddington, UK).The sampling time is a 10 min average every half an hour; 10 min of ambient air is measured after which the following 20 min are used to analyse the input.Grab samples of pure natural gas were taken of both pipelines, with the aim of characterizing the two differing gas supplies.The 0.8 L stainless steel flasks were evacuated prior to sampling to a pressure of the order of 10 −6 mbar, after which they were filled to ambient pressure when sampling.The flasks were measured independently in the laboratory with a manual GC (described in Sect.2.4) and, after dilution with zero air, by the CRDS instrument.
4.2 Impact of C 2 H 6 on δ 13 CH 4 observations at the field site To quantify the effect of C 2 H 6 interference on δ 13 CH 4 a total of 16 events were selected from the 2-week field campaign, with criteria defined as a peak exhibiting both increasing CH 4 concentrations and a change in δ 13 CH 4 signature for a minimum of 1 h.Two such events are plotted in Fig. 10.Event 1 represents the majority of events measured during the field campaign, in which CH 4 and C 2 H 6 are well correlated.This particular event has a maximum concentration of 11 ppm CH 4 and 0.6 ppm C 2 H 6 .On average the selected events have peak concentrations of 5 ppm CH 4 and 0.3 ppm C 2 H 6 .The methane isotopic signature was char- acterized using the Miller-Tans method (Miller and Tans, 2003), in which δ 13 CH * 4 CH 4 values are plotted against CH 4 to calculate the isotopic signature of the methane source in situations where the background is not constant.In order to avoid bias stemming from using ordinary least squared (OLS) regression, the York least squares fitting method was implemented, thus taking into account both the X and Y errors (York, 1968).All events excluding one were found to have δ 13 CH 4 signatures characteristic of natural gas, corresponding on average to −40 ‰.A single event (Event 2 plotted in Fig. 10) was detected with a δ 13 CH 4 signature of −59 ‰ ± 1.5 ‰.Such a signature suggests a biogenic source and, due to the south-westerly wind direction throughout the event (where the livestock holding is located), suggests the source is likely to originate from livestock, either as ruminant or manure emissions.
If the data are left uncorrected, sources containing C 2 H 6 substantially bias the calculated isotopic signature of CH 4 events.This is demonstrated in Fig. 10c where, for Event 1, the slope of points after C 2 H 6 correction (in blue) is shifted in comparison to the slope derived from points left uncorrected (in red), signifying a modification of the δ 13 CH 4 signature.Corrected δ 13 CH 4 suggests a signature of −40.0 ‰ ± 0.1 ‰, while uncorrected values imply −37.8 ‰ ± 0.08 ‰.When no C 2 H 6 is present, i.e.Event 2, there is no disparity between the raw and corrected δ 13 CH 4 slope, resulting in a δ 13 CH 4 signature of −59 ‰ ± 1 ‰ for both methods.For the 15 natural-gas-related events, the average shift induced due to uncorrected data is 2 ‰.Consequently the bias in isotopic signatures due to C 2 H 6 means that uncorrected data will always overestimate the source when a simple two endmember mixing model is applied.

Continuous field measurements of ethane
As an independent verification of the CRDS performance we compared two time series of C 2 H 6 which were measured simultaneously by the CRDS and GC-FID during the natural gas field campaign by using a co-located air inlet.The CRDS data were averaged to identical time stamps as the GC-FID, i.e. a 10 min average every 30 min.From which we calculated a root mean squared error (RMSE) of 13 ppb.Given the precision of C 2 H 6 measured by the CRDS instrument is 10 ppb for 10 min averages, and the uncertainty on the GC-FID is 15 %, we conclude that this is an extremely good agreement.
Furthermore, the flask samples, taken on the 4 July 2014, were measured by the CRDS to have a C 2 H 6 : CH 4 ratio of 0.074 ± 0.001 ppm C 2 H 6 / ppm CH 4 and 0.046 ± 0.003 ppm C 2 H 6 / ppm CH 4 for the gas within Pipeline 1 and Pipeline 2 respectively.On the same day gas quality data from the on-site GC recorded a C 2 H 6 : CH 4 ratio of 0.075 ppm C 2 H 6 / ppm CH 4 and 0.048 ppm C 2 H 6 / ppm CH 4 respectively.Although the error associated with the later figures is unknown, the strong agreement between the two verifies our correction and calibration strategy of C 2 H 6 .

Use of continuous observations of C 2 H 6 : CH 4 by CRDS
The instruments' capability to now measure interferencecorrected and calibrated C 2 H 6 opens the door for using another proxy for source apportionment, namely the C 2 H 6 : CH 4 ratio (Yacovitch et al., 2014, Roscioli et al., 2015, Smith et al., 2015).The C 2 H 6 : CH 4 ratio that characterizes each source is determined by the slope of the C 2 H 6 to CH 4 relationship.This method was applied to the 16 events identified within the natural gas field campaign, again using the York linear regression method, taking into account both X and Y error.Two examples of this method are displayed in the bottom panel of Fig.

Combined method for CH 4 source apportionment
To distinguish which pipeline the emissions originate from, we compare both the δ 13 CH 4 signature and the C 2 H 6 : CH 4 ratio source apportionment methods.The two pipelines were characterized from the whole-air samples taken on 4 July 2014; although the gas within the pipelines is subject to change as incoming gas varies, we assume here that this did not occur throughout the short duration of the campaign (24 June to 4 July 2014).The data collected from the aforementioned 16 events are compiled within Fig. 11, which illustrates the distribution of δ 13 CH 4 signature vs. C 2 H 6 : CH 4 ratios.The results from the flask measurements, i.e. characteristics of Pipeline 1 and 2, are plotted as dashed purple and red lines.Both methods clearly identify the biogenic source, seen as an outlier in the bottom left corner of the plot.Furthermore, both methods are able to distinguish between the two pipelines.The isotopic signatures of the natural gas events (on average 40.2 ‰ ± 0.5 ‰) are clustered near the isotopic signature of Pipeline 1, which has a δ 13 CH 4 signature of 40.7 ‰ ± 0.2 ‰, thus suggesting the majority of the measured methane is an emission from this pipeline.When considering the C 2 H 6 : CH 4 ratio a similar conclusion may be drawn as the mean C 2 H 6 : CH 4 ratio is 0.069 ± 0.002 ppm C 2 H 6 / ppm CH 4 , much like that of Pipeline 1 at 0.074 ± 0.003.A future study will address the shift in measured events to the left of Pipeline 1 in Fig. 11 by using additional VOC data from the GC-FID to aid source identification.The uncorrected 16 events are also plotted in Fig. 11 as circular markers.These are found in the top righthand corner of Fig. 11 and do not correspond well with either of the pipelines, thus reconfirming the importance of the corrections.

Concluding remarks
This study focuses on measurements of C 2 H 6 contaminated methane sources by a CRDS (Picarro G2201-i), with emphasis on correcting δ 13 CH 4 and (although not intended for use by standard users) C 2 H 6 for cross-interferences before calibration.Our extensive laboratory tests suggest that CRDS instruments of this model are all subject to similar interferences (as expected as they scan the same spectral lines) and that they can have a significant impact on reported concentrations and isotopic signatures if not accounted for properly when measuring industrial natural gas sources.For now, we suggest using constant, instrument-specific correction factors if possible or the ones found in this study (summarized in Fig. 12).As our study period only encompasses 1 year it is clear that the stability of the correction over the full life-time needs to be monitored further.To fully exploit the reported C 2 H 6 data, we suggest drying gas samples to < 0.16 % H 2 O, calibrating the instrument and taking frequent measurements of a working gas (or set of working gases) to monitor and correct for the instrumental baseline drift.
The results of our field campaign demonstrate the extent of the interferences of C 2 H 6 on δ 13 CH 4 for a real world application and also support the validity of our C 2 H 6 correction and calibration through the comparison with an independently calibrated GC-FID.In our case, when measuring wet gas emissions we detected an average shift in isotopic signature of 2.5 ‰ due to C 2 H 6 interference; however the extent of this bias will vary according to the contribution of C 2 H 6 , therefore affecting each ffCH 4 source to a different degree which can cause problems for source determination.The results reported here are important for all future work of CRDS in fos-sil fuel regions (where sources consist of a C 2 H 6 : CH 4 ratio between 0 and 1 ppm C 2 H 6 / ppm CH 4 ) to create awareness of such interferences and correct for them accordingly.Our CRDS instrument is sufficient for measurements of strongly variable C 2 H 6 sources, where if using calibrated 1 min C 2 H 6 data, concentration variations above 150 ppb are required to achieve a signal-to-noise ratio of 5. Thus for industrial natural gas sites it offers a new opportunity to use continuous C 2 H 6 : CH 4 observations as a means of source determination that is independent from δ 13 CH 4 methods.The recently released G2210-i analyser is dedicated to C 2 H 6 : CH 4 ratio measurements and as such achieves a higher precision, making it suitable for a wider variety of ethane sources.
Finally, we successfully combined both the δ 13 CH 4 and C 2 H 6 : CH 4 ratio source apportionment methods.At the natural gas compressor site both methods clearly distinguish biogenic sources from that of natural-gas-based sources.Combining those two independent methods yields a better fingerprint of the source and spurious C 2 H 6 or δ 13 CH 4 can be more easily identified.Lastly, by characterizing both the δ 13 CH 4 and C 2 H 6 : CH 4 ratio of our source, we gain insight into the formation and source region of the gas (Schoell, 1983).

Figure 1 .
Figure 1.Flow chart illustrating the steps involved to calibrate C 2 H 6 and δ 13 CH 4 .The number in the top right-hand corner corresponds to the subsection in which the methods of each step are explained in detail.

Figure 2 .
Figure 2. General set-up.The dilution and working gas are connected via two MFCs to two CRDS instruments in parallel.In red is the placement of an optional glass flask used for the C 2 H 6 calibration only.The flow is greater than that of the instruments' inlets.Therefore an open split is included to vent additional gas and retain ambient pressure at the inlets.

Figure 3 .
Figure 3.An example of the results from a H 2 O interference experiment spanning the range 0-1 % H 2 O.The reported C 2 H 6 is altered due to the addition of water vapour when measuring zero air (< 1 ppb C 2 H 6 ).Dark and light blue markers signify the response when dried and undried ambient air have been measured overnight by the instrument prior to the experiment respectively.Error bars signify the standard deviation of each measurement.

Figure 4 .
Figure 4.The discontinuity seen for instrument CFIDS 2072 for two repetitions denoted by different colours.After the discontinuity at 0.16 % the subsequent slope clearly differs between the two repetitions.Both instruments display a discontinuity at 0.16 % H 2 O.Each point represents a 1 min average, the error bars represent the standard deviation of the raw data.

Figure 5 .
Figure 5. Relationship between reported C 2 H 6 and concentration changes of CO 2 for instruments CFIDS 2072 and 2067 at varying values of H 2 O, at 0 ppm C 2 H 6 (within our instrumental precision).For each plot the bottom axis indicates the concentration of the targeted gas (CO 2 ).Plots (a) and (b) are at 0 % H 2 O, (c) and (d) are experiments at varying humidities, distinguishable by colour.The legend denotes repetitions of the experiment.The error bars in each plot denote the standard deviation of each measurement.The R 2 values for the experiments at 0 % H 2 O are 0.9 and 0.8 for all other H 2 O experiments for both instruments.

Figure 6 .
Figure 6.Relationship between reported C 2 H 6 and concentration changes of CH 4 for both instruments at 0 ppm C 2 H 6 (within our instrumental precision).For each plot, the bottom axis indicates the increase in concentration of the targeted gas.The vertical bars in each plot denote the standard deviation of each point.The legend denotes repetitions of the experiment.Plots (a) and (b) are at 0 % H 2 O.The R 2 values are 0.4 and 0.6 for instruments CFIDS 2072 and 2067.Plots (c) and (d) show the response at 1 % H 2 O.These two plots have a R 2 value of 0.2.

Figure 7 .
Figure 7. (a) Ethane calibration calculated from measurements of flask samples by both the GC and CRDS.The x-axis is the corrected C 2 H 6 (C 2 H 6 COR ) using the corrections described previously.The y-axis is the C 2 H 6 as measured by a manual GC.The error bars indicate the standard deviation of each flask measurement, for certain flasks error bars are smaller than their respective markers.(b) 30 min target measurements over a period of 4 days, from 13 to 16 November 2015.The standard error of each target is smaller than the plotted marker.The baseline C 2 H 6 is seen to drift with time.

Figure 9 .
Figure 9.The effect of C 2 H 6 on reported δ 13 CH 4 .The slopes of reported δ 13 CH 4 vs. the C 2 H 6 CORRECTED : CH 4 ratio are shown for three tests taken throughout the course of 1 year.Triangular markers imply whole-air sample measurements, while square markers are derived from direct measurements.Error bars indicate the standard deviation.In the presence of C 2 H 6 the instrument reports heavier values of δ 13 CH 4 .The typical range of (calibrated) C 2 H 6 : CH 4 of dry and wet gas are highlighted in pink and green respectively, corresponding to the top axis.
4.1.3Grab sample measurements of CH 4 , δ 13 CH 4 & C 2 H 6 in pure natural gas samples

Figure 10 .
Figure 10.Ethane and methane content of two selected peaks.Methane and ethane 1 min averaged time series is shown in (a) and (b) for Event 1 and (e) and (f) for Event 2. Miller-Tans plots of the corresponding peaks are shown in (c) and (g), blue for the corrected δ 13 CH 4 due to C 2 H 6 , and red representing uncorrected δ 13 CH 4 .Event 1 includes elevated C 2 H 6 emissions and thus displays a difference between the slope before and after C 2 H 6 correction, corresponding to a shift in isotopic signature.Event 2, with no C 2 H 6 shows no alteration in slope.The slopes of C 2 H 6 vs. CH 4 are shown in (d) and (h), signifying the C 2 H 6 : CH 4 ratio of the emission.Errors of both the isotopic and C 2 H 6 : CH 4 signatures are calculated from the standard error of the slope.

Figure 11 .
Figure 11.Distribution of 16 events according to their C 2 H 6 : CH 4 ratios and isotopic signature.The red and purple dashed lines signify the characterizations of Pipeline 1 and 2 respectively as measured by the CRDS instrument from flask samples taken on the 4 July 14.For corrected and calibrated data (square markers), both the isotopic signature and C 2 H 6 : CH 4 ratios identify the biogenic source (bottom-left point) and suggest the natural gas emissions emanate from Pipeline 1. Circular markers represent the uncorrected data which does not agree with the flask sample measurements of pipelines 1 or 2. The error bars indicate the standard error of the slope calculated from Miller-Tans and C 2 H 6 vs. CH 4 plots for δ 13 CH 4 signature and C 2 H 6 : CH 4 ratio respectively.

Figure 12 .
Figure12.Flow chart illustrating the steps and the corresponding equations to calibrate C 2 H 6 and δ 13 CH 4 as determined from this study.The coefficients are the mean of both CRDS instruments tested.We suggest removing H 2 O from gas samples prior to analysis.

Table 1 .
Description of the gas mixtures used to determine the cross-sensitivities of the interference of CH 4 , H 2 O and CO 2 on C 2 H 6 and the interference of C 2 H 6 on δ 13 CH 4 .The respective ranges spanned during laboratory tests, and the typical range at a natural gas site are noted on the right-hand side.

Table 2 .
Summary of C 2 H 6 calibration factors calculated for both instruments CFIDS 2072 and 2067.

Table 3 .
The various response functions calculated for the δ 13 CH 4 correction due to C 2 H 6 .
* Flask measurement.(δ 13CH 4 ) CORRECTED = (δ 13 CH 4 ) RAW − E * C 2 H 6 CORRECTED /CH 4 + F, (3)where E is the slope of the response function and F is the intercept.E and F are +23.6 ± 0.4 ‰ ppm CH 4 / ppm C : CH 4 ratio calculated is significantly shifted by approximately +0.06.The average raw C 2 H 6 : CH 4 ratio for the 15 natural gas events is 0.132 ± 0.007 ppm C 2 H 6 / ppm CH 4 , while the biogenic events C 2 H 6 : CH 4 ratio calculated is negative and thus impossible. 6