Estimating Oil Spill Occurrence Rates: A Case Study

About 25% of U.S. oil production comes from about 3400 active oil platforms in the Outer Continental Shelf (OCS) areas of the Gulf of Mexico, so estimating the rate of oil spills is important. This article reports on a portion of a study sponsored by the U.S. Minerals Management Service (MMS) that focused on extrapolating from the data on past operations in the Gulf of Mexico to potential future operations in the Alaska OCS areas of the Beaufort and Chukchi Seas. That extrapolation required the development of statistical models for the GOM that advanced current practice. Important differences from published MMS work include exact Poisson confidence intervals, exact binomial confidence intervals, detailed analyses for the exposure variables of pipeline mile-years and platform-years, the use of the larger spill data set of spills exceeding 50 barrels to estimate spill rates at higher thresholds, and the inclusion of more recent data (through 2005). A declining rate of platform spills is statistically verified, so that platform results are generally based on spills 1990 to 2005, while pipeline results are based on data from 1972 to 2005. It is suggested that some of the techniques may be applicable to other problems with a limited number of occurrences spread over decades.