Hotspot Identification for Massive Spatial Data Sets

• Activity Number: 519
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Joshua French*+
• Companies: University of Colorado Denver
• Keywords: hotspots ; exceedance regions ; massive spatial data ; confidence regions ; climate ; environment

Abstract

Hotspots or exceedance regions are the regions in a spatial domain where the response variable exceeds some threshold. Hotspot identification is important in public health, geology, climatology, and many other fields since these regions may indicate outbreak of disease, toxic conditions, regions of extreme risk, etc. Only recently has methodology been developed to confidently identify hotspots, but this methodology breaks down under the weight of large data sets. Methodology is presented for constructing a confidence region containing the entire true hotspot of a spatial process when the observed data set is massive. Modeling the responses using a spatial mixed effects model (a class of mixed models with flexible, nonstationary spatial random effects) allows for efficient prediction and estimation of prediction error. This model can be further exploited to allow for efficient conditional and unconditional simulation, which can be combined with existing methodology to obtain the desired confidence regions. The new methodology is applied in identifying the hotspots of a large climatological data set from the National Center for Atmospheric Research.