Adapting Statistical Learning Method for Spatial Applications

Abstract

In this dissertation, we develop new principled applications of statistical learning methods in spatial applications. In the first chapter, we consider a modified regression tree approach allowing for spatial correlation for applications in spatially indexed datasets. In the second chapter, we consider incorporating penalized regression estimators into universal kriging models. In the third and final chapter, we propose a class of flexible, additive regression tree models for joint estimation across multiple domains of interest.