In 2010 the Deepwater Horizon oil rig, located in the Gulf of Mexico, exploded. The resulting oil spill was the largest maritime oil spill in US history. Toxins were released into the Gulf region, both water and air. Over 100,000 individuals participated in the clean-up effort, and a major concern is the impact on their health from exposure to toxins from the oil, and the chemical dispersants used during the clean-up. Understanding the nature of exposure is one component in the evaluation of any potential health effects upon the clean-up workers. A large degree of the clean-up was focused along the coastal regions of the Gulf. Over 100,000 measurements of the concentration of fine particulate matter (PM2.5) were taken along the Gulf Coast between April and September of 2010. Such large datasets overwhelm traditional spatio-temporal models, so a reduced rank spatial model (RRSM) is necessary. In such a model, one specifies a set of knot locations to model though. Often these knot locations are chosen arbitrarily. In this work we develop a method of objectively selecting the knot locations for a RRSM and apply it to the PM2.5 data. The methodology developed is efficient and flexible enough to model any data indexed by location, or by location and time.
This is joint work with Shyamal D. Peddada.