Spatial Factor Models for High-Dimensional and Large Spatial Data: An Application in Forest Variable Mapping

• Activity Number: 410 - High-Dimensional Regression
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Daniel Taylor Rodriguez* and Andrew Oliver Finley and Abhi Datta and Chad Babcock and Hans-Erik Andersen and Bruce Douglas Cook and Douglas C Morton and Sudipto Banerjee
• Companies: Portland State University and Michigan State University and Johns Hopkins Bloomberg School of Public Health and University of Washington and USDA Forest Service and NASA Goddard Space Flight Center and NASA Goddard Space Flight Center and UCLA School of Public Health
• Keywords: LiDAR data; Nearest Neighbor GP; Spatial prediction

Abstract

Gathering information about forest variables is an expensive and arduous activity. As such, directly collecting the data required to produce high-resolution maps over large spatial domains is infeasible. Next generation collection initiatives of remotely sensed Light Detection and Ranging (LiDAR) data are specifically aimed at producing complete-coverage maps over large spatial domains. Given that LiDAR data and forest characteristics are often strongly correlated, it is possible to make use of the former to model, predict, and map forest variables over regions of interest. This entails dealing with the high-dimensional spatially dependent LiDAR outcomes over a large number of locations. With this in mind, we develop the Spatial Factor Nearest Neighbor Gaussian Process model, and embed it in a two-stage approach that connects the spatial structure found in LiDAR signals with forest variables. We provide a simulation experiment that demonstrates inferential and predictive performance of the SF-NNGP, and use the two-stage modeling strategy to generate complete-coverage maps of forest variables with associated uncertainty over a large region in Alaska.