Keystone XL Spill Risk: A Reanalysis of the Environmental Impact Statement - Guest Blog by David Malitz
Posted April 24, 2013
This is a guest blog post by David Malitz, Ph.D. David is a consultant living in Austin, Texas with over 30 years of experience in statistical analysis.
As a statistician, I expected the Keystone XL draft supplemental environmental impact statement (SEIS) to address the risk of oil spills if the pipeline were built. More specifically, if an 875 mile pipeline were built which carried as much as 800,000 barrels per day of heavy crude, what would be the risk of spills such as the 20,000 barrel spill in Michigan’s Kalamazoo River or the recent spill in Mayflower, Arkansas? Surely, we should be able to estimate the degree of such risk using historical data.
The SEIS partially addresses such spill risks, but in a way that seemed to me to be oddly incomplete. The report discusses (in Section 4.13.2 and Appendix K) an analysis by Keystone of data available from the Pipeline and Hazardous Materials Safety Administration (PHMSA), a federal agency which is part of the Department of Transportation. PHMSA’s database contains information about reported pipeline spill incidents including the size of the spill (in barrels) and the type of equipment which was the source of the spill (mainline pipe, tanks, mainline valves, and other discrete elements). The SEIS used data from 2002 through 2012 to calculate the historical risk associated with pipeline operation and used these data to present information regarding the distribution of spills by size (small, medium, and large) and the prevalence of spills by cause (e.g., internal corrosion, incorrect operations, etc.).
What interested me was not what the report said, but what it didn’t say. Tables in Appendix K showed the calculated rate of spills per pipeline mile per year (per “mile year”) as well as the average size of those spills. This tells us the rate (and volume) of spills per mile year. So couldn’t we just multiply this rate (and volume) by the length of the proposed pipeline (875 miles) to compute the average number of spills per year and the average size of those spills?
The SEIS didn’t take this step and it seems an obvious gap. Imagine buying carpet for your living room and knowing both the cost per square yard of the carpeting and the size of your room. Wouldn’t you want to multiply these two factors together to find out the total cost of carpeting your room? Of course you would, but Keystone XL’s SEIS fails to take this simple step which would tell us the environmental cost (or, at least, one of the environmental costs) of the proposed pipeline.
I performed this exercise and detailed my methods and results in a comment that I submitted to the Department of State (you can read my paper here). For this analysis, I used only statistics that were reported in the SEIS. I didn’t utilize any outside data sources, nor did I verify the accuracy of Keystone’s analysis of the PHMSA database. I also didn’t resort to data manipulation that was more sophisticated than addition and multiplication.
This analysis showed that based upon reported historical industry experience, we would expect about 1.9 spill incidents per year from the 875 mile proposed pipeline, with an average total spill volume per year of 805 barrels (almost 34,000 gallons). About 1/8 of these incidents on average (0.126) would be “large” (the SEIS classified spills of 1,000 to 20,000 barrels as “large”).
Over a longer time span, say a decade, we would expect about 19 spill incidents with an aggregate spill volume of about 8,000 barrels, enough to fill about half of an Olympic-sized swimming pool. We would expect about 1.3 of these spills to be “large,” which means that on average we would expect a “large” spill to occur about once every 8 years or so. Clearly, based upon reported historical industry performance, spills in general and large spills in particular would not be a rare occurrence for the proposed pipeline.
Since the statistics reported in the SEIS and in my reanalysis are entirely dependent upon the PHMSA database, it is reasonable to ask how accurate the PHMSA data are. A data quality assessment that was performed by PHMSA itself in 2009 details serious problems with the data that are reported by industry to the agency. The agency outlines many deficiencies with the data, including underreporting of incidents. A root problem for data unreliability is described by PHMSA as follows:
Most of our data collection relies on third-party reporting from regulated companies. This is convenient, and it goes directly to the source. It also introduces serious biases and gaps in the data we collect. Despite the best intentions and professionalism, the regulated industry has an institutional bias (and probably a liability aversion) in determining the causes, circumstances, and consequences of failures. Accident investigations—the limited number that we do—have shown some significant differences between what a company reports and an objective view of these events. Reports from companies also reflect large numbers of blanks and “unknown” data, particularly in the most serious cases—exactly where it is most critical that we have good data. (PHMSA, page C-12, italics in original).
Thus, it is likely that due to underreporting the spill rates that are reported in the SEIS and in my reanalysis are understated to an unknown degree.
Comments are closed for this post.