Abstract
The production of oil and gas in the offshore waters of the United States continues to be a major part of US energy extraction activities amounting to just less than a third of total US oil and gas production. However, this industry has been marked by occasional safety and environmental disasters including most famously the Deepwater Horizon explosion and oil spill that resulted in the deaths of 11 workers and the release of more than 130 million gallons of oil in to the Gulf of Mexico. In response, the Bureau of Safety and Environmental Enforcement (BSEE) was created in 2011 to separate enforcement activities from federal lease management activities and reduce the possibility for conflicts of interests and regulatory capture. This paper presents an analysis of the safety and environmental performance of the US offshore oil and gas industry in the years before and after the creation of the BSEE to quantify the changes in the industry record and the level of risk that remains. Recorded events including fires and explosions, spills, and gas releases, collisions, and injuries and fatalities are included in the analysis. The overall level of exposure is estimated based on rig counts and oil and gas production quantities since detailed employment records by facility are not available. Data is sourced from the BSEE, Bureau of Labor Statistics (BLS), and the Energy Information Agency (EIA). In addition to linear regression analysis of trends, this paper presents the results of a random forest-based machine learning investigation of the characteristics of safety and environmental incidents to evaluate the most significant contributors that remain, especially those amenable to control through engineering system design. Facility type, water depth, distance to shore, and time of day or year in the relevant incident reports were included in the input dataset for the random forest model. Results indicate that the overall oil and gas industry has become safer in recent years, though significant risks remain. It is yet unclear whether the BSEE approach bears any responsibility for this change as the data are not yet sufficient to declare the post-2011 period as statistically significantly improved from prior years, though additional data in line with 2016–2017 level of performance would satisfy this condition. The random forest model indicates that increased risk is associated with time of day, quarter of the year, water depth, and distance to shore. Data quality concerns remain present as minor incidents and injuries may be under-reported. BSEE enforcement does not appear to be a direct cause of the noted improvements.
