Reply to “Reply to ‘Comment on “Hydrocarbon emissions characterization in the Colorado Front Range – A Pilot Study”’ by Michael A. Levi” by Gabrielle Pétron et al.
Corresponding author: M. A. Levi, Council on Foreign Relations, New York, NY, USA. (email@example.com)
 The reply by Pétron et al. (2012a) to my comment, Levi (2012), on their paper, Pétron et al. (2012a), defends their original conclusion: fugitive emissions in an area that they observed in 2008 were likely larger than those indicated by previous estimates. The reply also claims to have now better characterized the uncertainties in their and my approaches. In particular, they question the representative nature of a data set that Pétron et al. (2012) relied on, and that Levi retained, although they do not recalculate their estimates. Pétron et al. (2012) also claim to show that the analysis in Levi (2012) is flawed because, they argue, it implies the physically unrealistic conclusion that fugitive emissions from the natural gas operations that Pétron et al. (2012) studied may be negative. Pétron et al. (2012a) are reasonable to be skeptical of the data set that they highlight. I show in this reply that if that data set is indeed not representative of condensate tank flashing in the area under study, it is impossible to produce the results in Pétron et al. (2012a) that the authors now defend. I also show that the other critique of Levi (2012) offered by Pétron et al. (2012a), i.e., that Levi's analysis has physically unrealistic implications, relies on an incorrect interpretation of the formulas in that comment. Only one of two conclusions is possible: the results in Levi (2012) are correct, or the conclusions in both Levi (2012a) and Pétron et al. (2012a) are unjustified.
1 Introduction and Questionable Assumptions
 Pétron et al.  and Levi  both estimate fugitive methane emissions from natural gas operations by assuming that observed methane and light alkanes in the air sampled by Pétron et al.  came primarily from fugitive emissions at natural gas operations and flashing of condensate tanks. Both juxtapose the same atmospheric observations of the ratio of methane-to-propane with the same range of possible profiles for flashing emissions, replying on profiles from 16 condensate tanks (all reported in Pétron et al. ). Pétron et al.  and Levi  also constrain the possible composition of fugitive natural gas emissions by using the same set of 77 profiles for Colorado natural gas wells (reported in COGCC ), although in different ways.
 Following Pétron et al. [2012a], we use the term “fugitive emissions” rather than “venting” here, to be more general regarding possible emissions sources. We include in “fugitive emissions” all methane, propane, and n-butane emissions upstream of natural gas processing facilities, including from well venting, incomplete combustion of flares, compressor engines, and gathering systems. We assume here (as is done implicitly in Pétron et al. [2012a] and Pétron et al. ) that these all exhibit the same ratios of methane-to-propane and methane-to-n-butane as raw gas, an assumption that we revisit below. (Our definition of “fugitive emissions” does not include products of incomplete combustion, such as formaldahyde, which are not fugitives and are of no consequence to the analysis here.) Pétron et al.  discard data points directly downwind of a natural gas and propane processing plant that leads to strong propane enhancements, eliminating that as a possible source.
 Pétron et al.  assume that the ratio of methane to propane in fugitive emissions is bounded by the mean (24.83) and median (15.43) of that ratio in the 77 wells. Levi  noted that the results in Pétron et al.  are highly sensitive to this assumption and argued that there is no basis for it, given that no claim is made that the 77 wells are representative of either producing or fugitive-emissions-prone wells in the area under study. Levi  noted that a 90% confidence interval for the ratio of methane-to-propane in the 77 well sample—more appropriate given that nothing is known about whether the sample is representative—contains wells with methane-to-propane ratios that range from 8.79 to 61.7. It is observed that, using the methodology in Pétron et al. , this implies a fugitive methane emissions rate that is bounded below by 66 Gg/yr—well below the lower uncertainty bound presented in Pétron et al.  and defended in Pétron et al. [2012a], but consistent with previous estimates—and is not bounded above.
 Pétron et al. [2012a] do not respond to this critique directly. They do, however, address it indirectly in the context of arguing that other results in Levi  are implausible. They observe that in the WRAP Phase III inventory [Bar-Ilan and Morris, 2012] of oil and gas activities in the area under study, “49% of the total VOC fugitive emissions were attributed to pneumatic devices and 32% to unpermitted fugitives emissions from various pieces of equipment at well pads.” They then noted that, in 2009, Colorado implemented regulations requiring low-bleed controllers on all valves in the area under examination, and infer that high-bleed valves “were likely used across the oil and gas field” during the period that Pétron et al.  studied (in 2008). From this, they conclude that “unpermitted fugitive emissions were also likely ubiquitous in the Denver-Julesburg Basin in 2008.”
 There are two main problems with this reasoning. The fact that regulations were imposed in 2009 does not imply that emissions were uniformly distributed (“ubiquitous”) among different types of wells prior to that year. (Regulations often impose uniformity on previously uneven practices.) More important, as pointed out in Levi  but not addressed in Pétron et al. [2012a], Pétron et al.  showed that assuming that the WRAP Phase III figures are correct implies that fugitive emissions in the Denver-Julesberg Basin are far lower than the alarmingly high ones that Pétron et al.  estimate using their top-down methods. Indeed a central conclusion in Pétron et al.  is that bottom-up estimates of methane emissions based on the WRAP Phase III figures are far too low (which suggests that something is missing from them). The implication is that either the WRAP Phase III figures are incorrect, in which case they should not be used to defend the conclusions in Pétron et al. , or they are correct, in which case they contradict the conclusions in Pétron et al.  that are defended in Pétron et al. [2012a]. Neither possibility reinforces the conclusions in Pétron et al. .
2 Summary of Critique of Levi 
 Unlike Pétron et al. , Levi  makes no assumptions about the ratio of methane to propane in fugitive natural gas emissions. Instead, it uses observations of the atmospheric propane-to-n-butane ratio reported in Pétron et al. , together with a 95% confidence interval for the propane-to-n-butane ratio in fugitive natural gas emissions for the 77 wells also used by Pétron et al. . This allows it to constrain estimates of fugitive methane emissions. Its results are consistent with previous bottom-up studies but not with the top-down estimates in Pétron et al. .
 Pétron et al. [2012a] do not directly challenge any of the physical reasoning behind the analysis in Levi . Instead they claim to show that the approach taken in Levi  implies physically implausible results, and hence is problematic. They then infer that the supposedly implausible results may be partly explained by the fact that “the full range of possible flashing emission profiles and the total VOC flashing source are likely not captured by the set of 16 flashing emission estimates used by Pétron et al.  and Levi .” Some skepticism regarding the representative nature of the flashing profiles is reasonable (although Pétron et al. [2012a] do not justify the conclusion that they are “likely” wrong). We show below, however, that their critique of the remainder of the methodology in Levi  is unsound. Moreover, applying the critique of the 16 flashing emission profiles in Pétron et al. [2012a] to the methodology in Pétron et al.  severely undermines their own original results.
3 Physical Realism of Levi 
 Levi  derives formulas for fugitive emissions of methane, propane, and n-butane given atmospheric methane-to-propane and methane-to-n-butane ratios; assumptions about the relationship between propane and n-butane concentrations in fugitive-emissions-prone wells derived from the data set of 77 wells; and a range of possible emissions from condensate tank flashing (which are constrained to be linear combinations of the profiles considered by Pétron et al. ). Pétron et al. [2012a] do not directly challenge the physical logic of any of the analytical steps in Levi , nor do they challenge the atmospheric concentration ratios used in Levi , or the assumptions that Levi  makes about gas that enters the atmosphere as fugitive emissions. They do note in passing that Levi  does not discuss the possibility that there is a third major source of n-butane emissions, other than fugitive emissions and flashing, that is colocated with, and proportional to, existing sources of propane emissions. For there to be a third source matching this description, there would need to be a common driver of all fugitive and flashing emissions that also drove significant n-butane emissions from this third source. This same source could not generate significant propane or methane emissions; otherwise, it would invalidate the critical assumption in Pétron et al.  that there are only two main types of sources of such emissions in the area under study. Pétron et al. [2012a] offered no suggestions for what this highly unusual source might be. That said, it is possible in principle that incomplete combustion of raw gas in flares and compressors could result in emissions that have different ratios of methane-to-propane and methane-to-n-butane from those characteristic of either raw gas venting or condensate tank flashing. If such fractionation is occurring strongly and at significant scale, then the two-source assumption used in Pétron et al.  and retained in Levi  and Pétron et al. [2012a] is incorrect. This could, in principle, fatally undermine the results in both papers.
 Pétron et al. [2012a], in any case, focused instead on the use the 16 condensate tank flashing profiles, first used in Pétron et al. , in Levi . They write that “Levi's calculation implies that only one condensate tank flashing profile (tank #14) out of the set of 16 in the CDPHE data set results in nonnegative propane and butane fugitive fluxes…. For the other 15 modeled flashing composition profiles, Levi's calculations require that fugitive emissions must remove propane and n-butane to match the atmospheric ratios.”
 This is incorrect. No tank profile “results” in anything; the relationships in question are only arithmetic. To be certain, if one assumes that all flashing emissions come from a single tank profile, then for 15 of the 16 profiles, the implied fugitive emissions of propane and n-butane are indeed negative. However, there is no reason to believe that flashing emissions are due to a single tank profile. The correct implication to take from the exercise in Pétron et al. [2012a] is that, as shown in Levi , it is not plausible that all flashing emissions come from the same profile. Indeed it would be quite surprising if they did.
 Pétron et al. [2012a] continue by arguing that “Levi's minimum fugitive methane emission estimates rely on a combination of the tank #14 positive estimate and a negative contribution from the estimate associated with tank #8. Given that we are dealing with source processes [leaks] that must be zero or positive [emitting to the atmosphere, not removing mass from the atmosphere], any single fugitive emission has to be positive to be considered part of the solution.” Based on this, Pétron et al. [2012a] rejected the remaining analysis in Levi .
 Their reasoning is incorrect: it attributes physical causation (“removing mass from the atmosphere”) to an arithmetic manipulation. Given a set of constraints on atmospheric concentrations of methane, propane, and n-butane, and a set of possible values of emissions from certain sources that contribute to these, Levi  estimates emissions from other (unknown) sources by requiring that the combination add up to the observed concentrations. There is nothing unusual or unphysical about the fact that higher emissions from some sources will imply lower emissions from others, given constraints on their total. These lower emissions are the inappropriately labeled “negative” or “removed” emissions that Pétron et al. [2012a] have identified. Pétron et al. [2012a] insisted this is unacceptable, and thus concluded that there is something fundamentally flawed in the approach in Levi . There is no reason to do so.
4 Condensate Tank Flashing Profiles
 Pétron et al. [2012a] do, however, reasonably point out that the results in Levi  indicate that fugitive emissions are biased toward dry gas wells. In part for that reason (though in part because they are skeptical of the “negative” propane and butane fugitive emissions) they reasonably suggest that the 16 flashing profiles used in Pétron et al. [2012a] and retained in Levi  may be unreliable. (This is in part why Levi  noted that “statistically meaningful samples of flashing emission profiles from condensate tanks” would be valuable.) The implication is that the results in Levi , which rely on the flashing profiles in question, are likely incorrect.
 This critique is inconsistent with the defense of the ultimate results in Pétron et al.  by Pétron et al. [2012a]. In Pétron et al. [2012a], the formula for fugitive methane emissions contains a factor equal to the difference between the mass of propane from flashing emissions (scaled by the atmospheric ratio of methane to propane) and the mass of methane from flashing emissions (equation (3)). However, Pétron et al. [2012a], after calling the tank flashing profiles relied upon in Pétron et al.  into question, suggest no new constraints on tank flashing emissions. If the flashing profiles used in Pétron et al.  are unreliable, there is now no way to calculate that paper's estimates of fugitive methane emissions, which depend vitally on the flashing profiles.
 One can only conclude that either the flashing profiles are reasonably representative—in which case Pétron et al. [2012a] have presented no reason to question the results in Levi —or the flashing profiles are unrepresentative, in which case neither Pétron et al. [2012a] nor Levi  have any basis to report reliable estimates of fugitive methane emissions. In either case, the results reported in Pétron et al.  are without foundation. Because the flashing profiles of condensate tanks in the area under study have likely changed since Pétron et al.  collected their data in 2008, this part of the debate is unlikely to be resolved definitively. Debate and observations should focus on rigorously understanding what is happening today through multiple observational and analytical methods. Several data collection efforts that could enable this are currently underway [EPA, 2012].