Computationally Efficient Spatial Statistics with Stochastic Differential Equations

Abstract

The Matérn covariance is a much used covariance function for spatial modelling using Gaussian random fields. One problem involving this model is that as the number of observations grow larger, the time it takes to perform computations using this model increases sufficiently as to make the model impractical in many cases. In this project an approximate model is presented, for which it is possible to use methods for sparse matrices that are computationally faster than the methods conventionally used.