An Emulator Approach to Upscaling an Individual-Based Model of Tree Growth for Learning About Tree Traits Affecting Forest Dynamics

• Activity Number: 293
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jarrett Barber*+ and Darren Gemoets and Kiona Ogle and Michael Fell
• Companies: Arizona State University and University of Wyoming and Arizona State University and Arizona State University
• Keywords: Bayesian ; computer experiments ; forest ecology ; large datasets ; meta models ; surrogate models

Abstract

We want to use our individual-based model (IBM) of tree growth to learn about growth and forest dynamics on large scales. For each tree, 25 traits (e.g., specific leaf area, wood density) are input into the IBM to determine a trajectory of radial growth. Allocation of carbon is constrained by allometric relationships so that form and function help to ensure realistic growth. Our objective is to inform IBM inputs with prior information on species-specific traits and with data on tree radii obtained from 1000s of plots representing >100 species on a regional scale. To address the computational impracticality of the IBM at this scale, we create a computationally tractable emulator of tree growth. Space-filling designs create a manageable number of input parameter points for which the IBM returns corresponding outputs. The emulator is then fit to these output "responses" and input "covariates", and we use the fitted emulator to learn about tree traits; emulator parameter posteriors are used as distributional constants to account for emulator approximation. We illustrate our approach using literature-derived trait information and tree data from USFS Forest Inventory and Analysis plots.