Smoothed Full-Scale Covariance Approximation for Large Spatial Data Sets

• Activity Number: 89
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 8:30 PM to 9:15 PM
• Sponsor: Korean International Statistical Society
• Author(s): Huiyan Sang*
• Companies: Texas A&M University
• Keywords: Covariance approximation ; Gaussian Process ; Spatial statistics ; Kriging

Abstract

With the advent of remote sensing and GPS techniques, the spatial data collection capacity increases dramatically and statisticians nowadays are facing a large number of observations on variables of interest. The growth in data size imposes challenges to classical statistical modeling methods and has driven the innovations of new methods scalable to handle large datasets. This work extends the full scale covariance approximation approach to preserve more information of the residual covariance by accounting for the dependence across blocks of the residual covariance. We show that the the proposed likelihood approximation approach still induces a valid Gaussian process, which allows for doing spatial prediction following the standard kriging method.