Spatial Prediction Using Multivariate Data Structures

• Activity Number: 525
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Alix I. Gitelman*+ and Xuan Che and Kathryn Irvine
• Companies: Department of Statistics, Oregon State University and Oregon State University and USGS
• Keywords: Bayesian belief networks ; structural equation models ; spatial prediction

Abstract

Bayesian belief networks and structural equation models are used increasingly in ecology to explore the multivariate structures in biological systems. We examine the predictive ability of these models using a large dataset of Pacific Cod presence/absence and catch in the Bering Sea. Here, we consider predictive ability in the sense of predicting one "response" of interest, and also in terms of a evaluating a "network of predictors" that is meaningful for the biological system. We compare prediction from Bayesian belief networks and spatial structural equation models to those from a model in Che (2012) that uses graphical modeling techniques and a Gaussian Copula transformation to incorporate spatial dependencies for non-Gaussian components of the multivariate system-particularly the binary presence/absence information. Implementation and computational issues are discussed and compared using both the Cod data and simulations.