This week I was directed to a new paper in Science of the Total Environment titled Volatile organic compounds (VOCs) in indoor air and tap water samples in residences of pregnant women living in an area of unconventional natural gas operations: Findings from the EXPERIVA study. The study was cited in a CBC article with the entertaining title: Homes near fracking sites in B.C. have higher levels of some pollutants, study finds. Needless to say this study was jumped on by opponents of BC LNG.

I previously blogged about the challenges of disconnecting theory from data. In that post, I discussed the concept of spurious correlations. Spurious correlations occur when two unrelated observations are incorrectly linked via a statistical analysis. The classic example is the purported link between pirates and climate change. This happens because statistics (especially the non-parametric statistics with small sample sizes used in this article) are prone to false positives. The easiest way to evaluate whether a correlation is likely spurious is to ask a simple question: what is the theoretical link between the two observations?  If no link can be established then it is likely the observed correlation is not real.

As I will demonstrate in this blog post, there is no viable mechanism to justify the relationship (or association as they call it) identified between the observed VOC concentrations and the presence/location of unconventional natural gas (UNG) LNG wells. For those who don’t remember their statistical jargon “associated” is the word you use when your result is not statistically significant but the correlation coefficient is slightly elevated. As I will demonstrate in this blog post, any “association” identified between VOCs and and UNG well proximity in this report is almost certainly spurious.

In the paper, the authors examined indoor air and tap water at the households of 92 pregnant women in the Peace River Valley. One-week indoor air and tap water samples were collected from each home. In addition, the authors used an Oil and Gas Commission database to identify LNG wells in the vicinity of each residence. The results from the air and water sampling was then compared to the LNG well data to see what nuggets might fall out. They had no specific hypothesis, they just threw all the data at the wall to see what the statistics said was relevant. The “Highlights” of the paper were:

Density of UNG wells was associated [my emphasis] with indoor air chloroform, acetone and BTEX.

Density of UNG wells was associated [my emphasis] with tap water trihalomethanes.

The Conclusion includes this line:

Our results also show that even when accounting for the region of residence and/or other known sources of exposure to VOCs, concentrations of acetone, chloroform and total trihalomethanes were associated with UNG well density/proximity metrics

For those unfamiliar with water treatment, chloroform and trihalomethane (THMs) are generated by chlorinated water when it is exposed to organic material. As the organic material reacts with the free chlorine it produces TCMs and/or chloroform. Every household with chlorinated tap water will have these compounds at some concentrations in the air and water with siltier water or water further from the original source generally having higher concentrations than water closer to the source or from pristine clean water.

Now the “associations” identified in the article surprised me because my reading of the fracking literature had never identified “acetone, chloroform or THMs” in fracking fluid or in LNG wells of any sort. Rather, anyone familiar with chloroform knows that it is avoided in most oil and gas uses because of its carcinogenic nature. Its use is typically restricted to labs where it can be handled in fume hoods.

For those who work in the industry, the bible for fracking fluid contents is the EPA’s Analysis of Hydraulic Fracturing Fluid Data from the FracFocus Chemical Disclosure Registry 1.0. The FracFocus chemical disclosure registry provides public disclosure of hydraulic fracturing chemical additives used in more than 55,000 wells by over 600 companies.So what does FracFocus say about “acetone, chloroform and THMs“? Not one of the three even makes an appearance in the document. So what does the paper say?

Hydraulic fracturing wastewaters and produce waters contain a number of VOCs, including acetone, xylenes (Lester et al., 2015) and chloroform(Akob et al., 2015),which are used as powerful solvents by the oil and gas industry (Luek and Gonsior, 2017)

Going to the reference Luek and Gonsior one discovers they say nothing of the sort about acetone or chloroform. Rather chloroform is only mentioned twice with the second instance being where it is identified as among a list of “suspected laboratory and field contaminants and inconsistent with contamination due to hydraulic fracturing activities”. In Lester et al., there is no mention of chloroform or THM and acetone is noted as being used as a cleaning solvent not part of the fracking solution. Akob et al is not a primary reference but rather refers to another report by Hays and Severin which reported finding chloroform in only 1 of 1330 samples in one run and only 2 of 60 well locations where solvents were observed and this was potentially attributed to reuse of fracking water. Acetone was identified as being used in numerous wells as a solvent that occasionally appeared in some wells and not as a standard fracking additive. Put simply, “acetone, chloroform and THMs” are not used in fracking fluids and in BC fracking does not involve the use of municipal water so they wouldn’t be an accidental byproduct. This poses some challenges to the “associations” identified in this article.

With respect to the household water sources the authors detail that 60% of the participants were on municipal water supplies. As the authors note:

Inclusion of Dawson Creek or Fort St John as the city of residence (covariate associated with both water treatment process generating trihalomethanes, and density of UNG wells) in the models did not change the associations between well density/proximity metrics and tap water concentrations of total trihalomethanes.

According to the authors the relationship between the fracking and water quality is

It is known that an increase in organic matter entering a water treatment play may lead to an increase formation of trihalomethanes (Xie, 2003). Interestingly, wastewaters generated during hydraulic fracturing contains high concentrations of dissolved organic matter… Surface and groundwater contamination events linked to UNG operations have been documented in the scientific literature…Furthermore, increases in the concentration of trihalomethanes have been observed in drinking water facilities in Pennsylvania, potentially because of the high levels of dissolved organic matter in hydraulic fracturing wastewaters discharged into surface water resources (EPA, 2016). Given these documented events, it is therefore possible that the density and proximity of UNG wells is contributing to the high concentrations of tap water total trihalomethanes in the EXPERIVA study.

So here is the thing. The authors evaluated the proximity of the dwellings to local LNG wells. The analysis did not consider the proximity of LNG wells to municipal supply sources. The City of Fort St. John gets its potable water from 5 shallow wells in Taylor on the Peace River. Dawson Creek gets its water from the Kiskatinaw Watershed northeast of Tumbler Ridge. The residences (and associated LNG wells) are not related in any way to the water source locations. The authors don’t explain how an LNG well, 2.5 km from a dwelling, can affect tap water that is supplied via a piped municipal water supply sourced 175 km away (in the case of Dawson Creek) to that dwelling.

So let me put this all in one paragraph, to understand the logic underlying their identified “association”. Dawson Creek gets its water from a watershed 175 km to the north of the city. The water is pumped to a modern and highly advanced multi-stage water treatment facility. The water is then pumped into a utility system that transports the water to individual dwellings. If the “association” was real it would mean that a fracked LNG well which was installed sometime in the last couple decades, located something like 2.5 km from the dwelling, generated enough organic material during its original installation to have affected said water to the extent that the dwelling’s water is now generating excess THMs and chloroform. In order for the “association” to be correct the organic material generated from that well either migrated the 175 kms to the water system’s originating watershed or migrated overland to overwhelm the Dawson Creek water treatment plant and thus caused the water to generate excess THMs and chloroform. Does anyone else imagine that a better explanation for this fairy tale would be “spurious correlation”?

And it gets worse. As I noted earlier, LNG wells don’t generate “acetone, chloroform or THMs” so how are these compounds getting to the indoor air from the LNG wells? There is no plausible mechanism by which an LNG well can cause an indoor air problem kilometers away…indoors no less!!! The observed VOCs in the indoor air are, however, typically associated with the tap water concentrations but as we have discussed, the tap water is not related to proximity to any LNG wells in the study.

This post is getting a bit long so I will simply highlight some other gems from this article.

The report relies on self-reporting by the participants. Self-reporting without auditing is known for its lack of reliability. Consider that according to the paper:

It is important to note that other household products containing chlorine (e.g., toilet cleaners, bleach, detergents) can lead to VOC emissions indoor (Odabasi, 2008). In EXPERIVA, only one participant confirmed storing this type of household products in their residence.

If I read that correctly, it appears to say that of the 92 households surveyed only one stored household cleaning products in their house? I can go around my house and find three different types of household cleaning products that include chlorine (toilet cleaner, laundry bleach and Vim household cleaner) yet only one of the 92 households had any chlorine cleaners of any kind? That seems…unlikely???

With respect to air sampling the study includes zero duplicates and zero regional background samples. Recognize that the authors are attributing indoor air concentrations of VOCs to outdoor sources (LNG wells). Therefore these concentrations should be higher in the outside air…but the researchers chose to collect no samples to compare indoor versus outdoor air concentrations, nor did they collect duplicates from different locations in the same houses.

Were I preparing a report on indoor air for a regulator the absence of a background sample or duplicates would result in my sampling regime to be deemed deficient and I would be sent back to reproduce my work. That the peer reviewers failed to address this issue is simply problematic.

The report also doesn’t consider how old the LNG wells are or when they were installed. The VOCs considered in the study were all quite volatile and would be subject to rapid deterioration in the natural environment. The VOCs would only be an issue immediately after a well was fracked so a well installed a year prior would simply not generate these VOCs. So even if these VOCs were found in the fracking water [they aren’t] their presence would only be expected for a very short time after installation. That consideration was never included in the analysis.

I think I can stop here because from the information I have provided above there is simply no way anyone could credibly argue that the associations observed in the article are real. The critical VOCs identified as being “associated” with “UNG well density/proximity metrics” do not appear in fracking formulations in BC and the” UNG well density/proximity metrics” used to generate the association between tap water and dwellings does not account for the actual source of the drinking water source for 60% of the sample locations. The UNG locations used to generate the “association” were incorrect. How could that result in a real association?