Modeling Spatially Dependent Forest Diameter Class Distributions Using High-Dimensional Lidar Data

• Activity Number: 104
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Andrew Oliver Finley*+ and Sudipto Banerjee
• Companies: Michigan State University and University of Minnesota
• Keywords: Bayesian ; space-time dynamic models ; predictive process ; forestry

Abstract

Spatially explicit estimates of number of trees by diameter class, referred to as a diameter class distribution, are key pieces of information used in nearly all forest ecosystem assessments. Producing such estimates at the desired fine spatial resolution is often prohibitively expensive due to the large amount of field measurements needed. We explore models that not only capture the correlation between diameter class distributions across locations, but also accommodate the correlation in number of trees between classes within a given location. To capture the between location correlation in diameter classes we consider a multivariate spatial Poisson regression model. One way to introduce dependence across the classes is to assume a Markovian structure across the diameter increments. The forest inventories we consider are high dimensional in both the number of locations where diameter class distributions are observed and also the number of classes within a given distribution. We consider various approaches to dimension reduction and illustrate the proposed models using forest inventory data collected on the USDA Penobscot Experimental Forest, ME.