News

Seminar

Covariance Tapering for Large Spatial Datasets

Ben Shaby
Postdoctoral Fellow, SAMSI (Statistics and Applied Mathematical Sciences Institute)

Date: Wednesday, November 4th, 2009
Location JEC 3117 - 11:00 a.m. to 12:00 p.m.

Abstract:

Spatial data is data for which the locations at which the observations are made are informative. This type of data plays a central role in areas like epidemiology, ecology, and climate science. Often, we like to think of spatial data as being a function of a realization of a Gaussian random field. Likelihood-based statistical analysis of Gaussian random field data is computationally expensive, and the site of relevant datasets is growing faster than our ability to carry out these computations. This challenge can be mitigated by introducing sparsity into the covariance matrix in a principled way and using fast sparse matrix algorithms. In this talk I will describe this "tapering" procedure and investigate the statistical properties of the resultant estimators.

Last updated: October 13, 2009