Deep Compositional Spatial Models

  • 2020-04-03 16:05:52
  • Andrew Zammit-Mangion, Tin Lok James Ng, Quan Vu, Maurizio Filippone
  • 0

Abstract

Spatial processes with nonstationary and anisotropic covariance structure areoften used when modelling, analysing and predicting complex environmentalphenomena. Such processes may often be expressed as ones that have stationaryand isotropic covariance structure on a warped spatial domain. However, thewarping function is generally difficult to fit and not constrained to beinjective, often resulting in `space-folding.' Here, we propose modelling aninjective warping function through a composition of multiple elementalinjective functions in a deep-learning framework. We consider two cases; first,when these functions are known up to some weights that need to be estimated,and, second, when the weights in each layer are random. Inspired by recentmethodological and technological advances in deep learning and deep Gaussianprocesses, we employ approximate Bayesian methods to make inference with thesemodels using graphics processing units. Through simulation studies in one andtwo dimensions we show that the deep compositional spatial models are quick tofit, and are able to provide better predictions and uncertainty quantificationthan other deep stochastic models of similar complexity. We also show theirremarkable capacity to model nonstationary, anisotropic spatial data usingradiances from the MODIS instrument aboard the Aqua satellite.

 

Quick Read (beta)

loading the full paper ...