[1] Pétron et al. (2012) have recently observed and analyzed alkane concentrations in air in Colorado's Weld County and used them to estimate the volume of methane vented from oil and gas operations in the Denver-Julesburg Basin. They conclude that “the emissions of the species we measured are most likely underestimated in current inventories”, often by large factors. However, their estimates of methane venting, and hence of other alkane emissions, rely on unfounded assumptions about the composition of vented natural gas. We show that relaxing those assumptions results in much greater uncertainty. We then exploit previously unused observations reported in Pétron et al. (2012) to constrain methane emissions without making assumptions about the composition of vented gas. This results in a new set of estimates that are consistent with current inventories but inconsistent with the estimates in Pétron et al. (2012). The analysis also demonstrates the value of the mobile air sampling method employed in Pétron et al. (2012).

[2] Several studies reporting unexpectedly high methane leakage from natural gas operations have recently attracted attention and sparked debate [Howarth et al., 2011; Jiang et al., 2011; Cathles et al., 2012]. Pétron et al. [2012](henceforth P12) have now attempted to infer rates of methane emissions from oil and gas operations in the Denver-Julesburg Basin directly from novel measurements of alkane (notably methane and propane) concentrations in air near those operations. They report much higher rates of methane emissions than have been previously estimated through bottom up methods based on industry inventories. Here, we show that their results rely on unsupported assumptions about the molecular composition of vented natural gas. We then use additional observations reported (but not exploited) inP12to estimate the rate of methane leakage without resort to assumptions about the composition of vented methane gas. Our conclusions are consistent with the more modest emissions rates indicated by bottom-up inventories but not with the top-down estimates presented inP12. In addition, our estimates could, in principle, be further constrained by additional observations.

[3]P12analyze alkane concentrations in air samples collected both by the National Oceanographic and Atmospheric Administration (NOAA) Boulder Atmospheric Observatory (BAO) and by mobile air surveys in the Denver-Julesburg Basin area. The former finds a C_{3}H_{8}-to-CH_{4} (C_{3}/C_{1}) molar ratio of 0.104 ± 0.005 for summertime samples (0.105 ± 0.004 for wintertime) originating near oil and gas producing areas, while the latter finds a C_{3}/C_{1} molar ratio of 0.095 ± 0.007. The authors use a C_{3}/C_{1} ratio of 0.1 in their subsequent analyses.

[4] To estimate methane leakage based on their air observations, the authors begin by noting that most observed alkane emissions come from either raw gas venting or condensate tank flashing. They then create two equations. The first describes the vented gas:

vm/p=MpMmxmxp

where v_{m/p}is the basin-average C_{1}/C_{3} molar ratio of vented raw gas, M_{p} = 44 g/mol and M_{m} = 16 g/mol are the molar masses of C_{3}H_{8} and CH_{4} respectively, x_{m} is the mass of methane vented, and x_{p} is the mass of propane vented.

[5] The second relates emissions to observed concentrations of CH_{4} and C_{3}H_{8}:

Mpxm+ymMmxp+yp=am/p

where y_{m} is the mass of methane released by flashing, y_{p} is the mass of propane released by flashing, and a_{m/p} = 10 is the observed ratio of CH_{4} to C_{3}H_{8} in air.

[6] These are solved as follows:

xp=am/pyp−ymMp/Mmvm/p−am/p

xm=vm/pxpMmMp.

[7] The authors use three different values for v_{m/p} to evaluate equations (3) and (4): (1) 18.75, which is the mean value of v_{m/p}used in the Western Regional Air Partnership (WRAP) Phase III inventory of oil and gas emissions in the Denver-Julesburg Basin; (2) 15.43, which is the median of the molar ratios of methane to propane in seventy seven wells studied by theColorado Oil and Gas Conservation Commission (COGCC) [2007] Greater Wattenberg Area Baseline Study (henceforth referred to as the GWA survey); and (3) 24.83, which is the mean of the molar ratios for the same seventy seven wells. For each of these, P12 evaluate x_{p} and x_{m} for each of 16 pairs of Y_{m} and Y_{p}, each of which is based on an observed profile of flashed gas for a single condensate tank. (This data is provided to them by the Colorado Department of Public Health and the Environment (CDPHE); it has been provided to the present author by K. M. Sgamma (Western Energy Alliance, personal communication, 2012).) This gives them minimum, maximum, and average (across all 16 flashing profiles) levels of methane venting for each of the three values for v_{m/p}. The authors also create bottom-up estimates of methane venting based on figures from the WRAP Phase III study.

[8]Table 1 reproduces relevant results from Table 4 of P12. Columns 2 and 3 in Table 4 have been reversed in the original paper, which is corrected here. As emphasized in P12, estimates for methane venting done through the top-down method are much higher than the bottom-up ones.

Table 1. Estimates of Methane Emissions From P12 in Gg/yr

Bottom Up Emissions

Top Down Venting Emissions

Flashing

Venting

Flashing + Venting

v_{m/p} = 18.75

v_{m/p} = 15.43

v_{m/p} = 24.83

Average

11.2

53.1

64.3

118.4

157

92.5

Minimum

4

42

46

86.5

114.7

67.6

Maximum

23

63

86

172.6

228.9

134.9

3. Methodological Limitations

[9] There is, however, no reason given for believing that the three values of v_{m/p} used in P12 actually bracket the possible range of C_{1}/C_{3} ratios that might characterize vented gas. Indeed the results in P12 suggest that the choice of potential values for v_{m/p} may be incorrect.

[10] There is no overlap between the ranges of possible methane emissions estimated from the bottom up and the top down solely using WRAP III figures. This can only be true if either the choice of v_{m/p}is wrong or if some of the underlying WRAP III figures themselves are incorrect; neither allows one to give credence to this particular top-down estimate.

[11] That said, the top-down estimates based on the GWA survey do not rely on the WRAP III-based assumption about the value ofv_{m/p}. They thus have the potential to provide independent insight into methane venting. However, P12 rely on an assumption that the molar ratio of CH_{4} to C_{3}H_{8} in vented gas is bracketed by the median of that ratio in the 77 wells in the GWA survey (Case 2 above, v_{m/p} = 15.43) and the average of those wells (Case 3, v_{m/p}= 24.83). But the authors make no contention that the 77 wells sampled in the GWA survey are representative of producing wells in Weld Country. Moreover, and most importantly, there is no reason to assume that the typical venting-prone well hasv_{m/p} bounded by the median and mean for all 77 wells.

[12] Indeed the full range of wells sampled show v_{m/p} ranging from 4.11 to 260.2; ninety percent of the wells have v_{m/p} between 8.79 and 61.7. Applying formulas 3–4 above together with lower bound for flashing emissions (reported in P12) yields a lower bound on methane venting emissions of 48 Gg/yr, well below any of the uncertainty ranges reported for the top-down estimates inP12. Moreover, even if one uses the average over the full ensemble of condensate tank flashing profiles reported, instead of the minimum, the estimated lower bound on methane venting emissions is 66 Gg/yr, still outside any of the uncertainty ranges reported for the top-down estimates inP12. Meanwhile, combining the observations of a_{m/p} used in P12 with the full range of v_{m/p}that characterizes potential venting-prone wells yield no upper bound on methane venting emissions. Indeed it is entirely plausible that venting is biased toward wells with either high or lowv_{m/p}, since those tend to characterize different types of production wells (gas and condensate wells, respectively). The upshot is that, absent difficult to support assumptions about the composition of vented natural gas, the top-down methods used inP12 give no new constraints on methane emissions.

4. Constraining Methane Emissions

[13] While P12 use only the observed C_{1}/C_{3} ratio to constrain methane emissions, they note that the observed C_{1}/nC_{4}(methane-to-butane) ratio can be used to do the same thing. In this section, we combine the observed C_{1}/C_{3} and C_{1}/nC_{4} ratios to remove the need to make assumptions about v_{m/p}, and hence better constrain estimates of methane emissions.

where we have defined X_{i} = x_{i}/M_{i} for all species i in order to simplify our equations.In addition, we have two similar constraints related to observed butane levels:

Xm/Xb=vm/b

Xm+YmXb+Yb=am/b

where v_{m/p} is the ratio of methane to butane in vented gas, X_{b} is the number of moles of butane vented, Y_{b} is the number of moles released through condensate tank flashing, and a_{m/p} is the observed ratio of methane to butane in air. We also define a_{b/p} = a_{m/p}/a_{m/b}.

[15] To avoid the assumptions made in P12 about the composition of vented gas, we proceed as follows. Let

vm/p=∑NQNvm/pN

vm/b=∑NQNvm/bN

where N is an index that ranges over all wells, Q_{N} is the fraction of total venting due to well N, vm/pN is the ratio of methane to propane in well N, and vm/bN is the ratio of methane to butane in well N.

[16] C_{1}/C_{3} and C_{1}/nC_{4} are consistently correlated in the 77 wells sampled in the GWA assessment [COGCC, 2007]. Specifically, if

vm/bN=bvm/pN,

we can estimate b=4.15±1.652.43 (95 percent confidence interval). In obtaining these values, we discard one outlying well for which C_{1}/C_{3} (260) and C_{1}/nC_{4} (2277) are much greater than for all other wells. (This observation indicates unusually dry gas for the area under investigation.) One can obtain a slightly better fit, and hence sharper constraints on X_{i}, by introducing a constant term in equation (11). Doing so, however, makes the analysis below considerably more complex and opaque while producing similar results.

At a similar point in the P12 analysis, the authors continue by evaluating X_{m} for the maximum, minimum, and average values of Y_{p} and Y_{m} over their ensemble of 16 condensate tank flashing profiles, thus obtaining a range of estimates for X_{m}. In the present case, though, one finds that for all but one set of flashing profiles, the implied X_{p} (based on equation (13)) is negative. Thus, in order to understand the full range of possible venting rates, we first need to determine the space of Y_{m}, Y_{p}, and Y_{b} for which X_{m}, X_{p}, and X_{b}are all non-negative. (We always assume, as inP12, that Y_{m}, Y_{p}, and Y_{b} are obtained by some linear combination of the 16 flashing profiles used in P12.) Specifically, we need to determine the sets of Y_{m}, Y_{p}, and Y_{b} that maximize and minimize implied X_{m}.

[18] We find that X_{m} is maximized for Y_{m} = 0.51, Y_{p} = 0.32, and Y_{b} = 0.17. The similar values that minimize X_{m} depend on a_{b/p}. We find that for observations using the mobile lab (a_{b/p} = 0.490), X_{m} is minimized for Y_{m} = 0.56, Y_{p} = 0.33, and Y_{b} = 0.16, while for observations using the BAO (a_{b/p} = 0.447), X_{m} is minimized for Y_{m} = 0.58, Y_{p} = 0.33, and Y_{b} = 0.16 (detailed justifications for these figures are in Appendix A).

[19]Equation (15) now allows us to calculate the range of most likely values for X_{m}, and hence x_{m}. (We present no expected value within this range because we have no way of determining which values of Y_{i} are most likely.) We also estimate uncertainties (95 percent confidence intervals) in the maximum and minimum values for these ranges by propagating known uncertainties in b, a_{m/p}, and a_{b/p}. Uncertainties in a_{m/p} are given in Table 3 of P12. Table 3 of P12 also reports uncertainties for a_{b/p}, but these exclude systematic uncertainty of as much as 20 percent (total) due to provisional calibration of the equipment used to measure n-butane concentrations (G. Pétron, personal communication, 2012); we combine both sources of uncertainty in our estimates. The uncertainty forb reported above ( b=4.15±1.652.43) is for a single well; the uncertainty for a sample with a large number of wells will be lower unless we assume that all wells are of the same profile. We estimate uncertainties both in the conservative case where all venting emissions come from wells with one consistent profile, and for the more realistic case where 100 different profiles are represented among wells that vent significantly. This is still somewhat conservative but is more likely to be more realistic, and reduces uncertainty in b by a factor of 10. Since b is only weakly correlated with a_{m/p} — their correlation coefficient is 0.24, or 0.19 if we exclude wells drilled in the Sussex zone, which are rare — this is still much weaker than the assumptions made in P12 that wells that vent significantly have certain values of a_{m/p}. The results are summarized in Table 2 and Figure 1.

Table 2. Revised Estimates of Methane Emissions in Gg/yr

Mobile Lab

BAO

Expected

Realistic Errors

Conservative Errors

Expected

Realistic Errors

Conservative Errors

Maximum

52.5

+15.9/−10.9

+19.5/−11.0

58.8

+20.1/−12.8

+62.2/−13.8

Minimum

46.4

+4.4/−3.9

+4.5/−3.9

49.4

+2.9/−9.6

+28.5/−9.9

[20] With the exception of the combination of BAO observations and highly conservative uncertainty estimates, all of the inferred methane emissions rates are consistent with those derived from accepted bottom-up inventories, but inconsistent with the top-down estimates reported inP12. Indeed the method used here places considerably tighter constraints on methane emissions than previous ones have. The one exception is in the case of observations at the BAO using highly conservative uncertainty estimates: there, there remains a very small chance that annual methane venting emissions are greater than 118 Gg/yr. It is most likely, though, that this simply indicates that observations at a single point (the BAO) are insufficient to tightly constrain possible methane emissions across the entire Denver-Julesburg basin.

5. Conclusion

[21]P12infer from air measurements of methane-to-propane ratios that methane leakage from oil and gas operations in Weld County, Colorado, is considerably higher than previously believed. However, this inference is based on assumptions about the molecular profile of vented natural gas that lack support. Using observed methane-to-propane and butane-to-propane ratios, both of which are reported inP12, we have made independent estimates of methane emissions that do not rely on assumptions about the composition of vented gas. These estimates are largely consistent with previous bottom-up predictions of methane emissions from oil and gas operations. The coincidence of bottom-up and new top-down estimates reported here for estimates using the mobile lab, as well as the modest uncertainties in methane leakage inferred from those observations, also indicates the potential value of carefully monitoring alkane concentrations in air near oil and gas operations, particularly through sampling across entire areas of operations. Additional observations, including statistically meaningful samples of flashing emission profiles from condensate tanks, could be used to further constrain estimates of methane emissions. Moreover, the prominent role of uncertainty ina_{b/p} in the analysis suggests that repeating the observations reported in P12but with more careful calibration of n-butane measurements could further constrain estimates of alkane venting from oil and gas operations.

Appendix A:: Supplementary Material

[22] Estimating methane emissions requires that we determine the sets of Y_{m}, Y_{p}, and Y_{b} that maximize and minimize implied X_{m}.

[23] Denote the constituent emissions for the sixteen flashing profiles used in P12 as Y_{m}^{L}, Y_{p}^{L}, and Y_{b}^{L}, where L is an index that ranges from 1 to 16, and Y_{i}^{L} is rate of emissions of species i due to flashing that one would observe if all flashing emissions came from condensate tanks with the profile of tank L. The values for Y_{i}^{L} are given in Table A1. We have

Yi=∑LPLYiL

where P_{L} is the fraction of condensate tanks that generate flashing emissions with the same profile as that of tank L in the reference ensemble. To determine the set of P_{L} that maximizes implied X_{m}, note from equation (15) that X_{m} is linear in Y_{m}, Y_{p}, and Y_{b}. We thus have

Xm=∑LPLXmL

where X_{m}^{L} is X_{m} evaluated for Y_{i} = Y_{i}^{L}. Substituting the values of Y_{i}^{L} into (A1) reveals that X_{m}^{14} > X_{m}^{L} for all L ≠ 14, which implies that X_{m} is maximized for P_{14} = 1 and P_{L} = 0 for L ≠ 14. This corresponds to Y_{m} = 0.51, Y_{p} = 0.32, and Y_{b} = 0.17, all in Gmol/yr.

Table A1. Flashing Profiles For Reference Tank Ensemble

Tank Number

Y_{m}

Y_{p}

Y_{b}

1

1.537

0.424

0.107

2

0.369

0.498

0.173

3

0.551

0.476

0.168

4

0.787

0.383

0.135

5

0.235

0.446

0.145

6

0.611

0.411

0.079

7

0.501

0.398

0.147

8

1.034

0.355

0.095

9

1.357

0.393

0.120

10

0.810

0.378

0.109

11

0.271

0.396

0.146

12

0.749

0.38

0.125

13

1.122

0.396

0.125

14

0.507

0.322

0.167

15

0.352

0.463

0.171

16

0.427

0.544

0.168

[24] To determine the set of P_{L} that minimizes implied X_{m}, note from equation (13) that X_{p} is linear in Y_{b} and Y_{p}. We thus have

Xp=∑LPLXpL

where X_{p}^{L} is X_{p} evaluated for Y_{i} = Y_{i}^{L}. Substituting the values of Y_{i}^{L} into A3 reveals that X_{p}^{14} > 0 and X_{p}^{L} < 0 for all L ≠ 14. In order to have X_{p} > 0, then, we must have P_{14} > 0. In addition, for any choice of Y_{b} and Y_{p} such that implied X_{p} > 0, we can lower the implied X_{p} and X_{m} by lowering P_{14} and increasing any of those P_{L} for which X_{m}^{L} < 0. This implies that X_{m} will be minimized for a set of P_{L} such that X_{p} = 0, or Y_{b} = Y_{p} − a_{b/p}.

[26] Define RL=XmL−Xm14/XpL−Xp14 for all L ≠ 14. Note that R_{L} is maximized for L = 8. We now show that X_{m} is minimized only if P_{L} = 0 for all L ∉ {8,14}. To do that, assume that we have some set of P_{L} than minimizes X_{m}. For any K ∉ {8,14}, decreasing P_{L} by Δ while increasing P_{8} by ΔXpK−Xp14/Xp8−Xp14 and P_{14} by ΔXp8−XpK/Xp8−Xp14, where Δ is an arbitrarily small positive number, leaves X_{p} > 0. It does, however, decrease X_{m} by Xp14−XpK/R8−RK. This implies that X_{m} could only have been a minimum if P_{L} was zero for all L ∉ {8,14} in the first place.

[27] We thus know that X_{m} is minimized for some P_{L} such that P_{8} and P_{14} are nonzero and P_{L} = 0 for all other L. As noted above, this minimum will occur as X_{p} approaches zero. We can thus calculate P_{8} and P_{14} that minimize X_{m} for each possible value of a_{b/p}. For observations made using the mobile lab (a_{b/p} = 0.490), this is obtained for P_{8} = 0.10, P_{14} = 0.90 (Y_{m} = 0.56, Y_{p} = 0.33, Y_{b} = 0.16). For observations using the BAO (a_{b/p} = 0.447), this is obtained for P_{8} = 0.14 and P_{14} = 0.86 (Y_{m} = 0.58, Y_{p} = 0.43, Y_{m} = 0.16).