Predicting spatial hotspot regions

The identification of spatially extreme observations or "hotspots" figures prominently in the application of various forms of statistical analyses. Such analyses do not necessarily capture all suspect or interesting observations. A spatial modeling approach can incorporate possible dependencies in the data. However, in realistic scenarios, a single spatial hotspot is difficult to pin down as it influences or contaminates the surrounding values, thereby obscuring the extreme nature of a particular site. Therefore, we focus on the identification of extreme regions, or areas that are extremely high or low in value. We use several modeling techniques to estimate the placement and size of an extreme region, including a non-spatial model as well as four spatial modeling techniques. An application using dissolved oxygen levels in the Chesapeake Bay demonstrates this approach, and we compare these models using inferential and diagnostic measures.