Scalable Filtering Methods For High-Dimensional Spatio-Temporal Data
Abstract
We propose a family of filtering methods for deriving the filtering distribution in the context of a high-dimensional state-space model. In the first chapter, we develop and describe in detail the basic method, which can be used in a linear case with Gaussian data. In the second chapter, we show how this method can be extended to incorporate non-Gaussian observations and non-linear temporal evolution models. We discuss how two algorithms, the multi-resolution decomposition and the incomplete Cholesky decomposition, can be used to quickly update the filtering distribution at each time step of the filtering procedures.
Citation
Jurek, Marcin Piotr (2020). Scalable Filtering Methods For High-Dimensional Spatio-Temporal Data. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /192341.