B0794
Title: Spatial multi-resolution model for forestry data
Authors: Isa Marques - The Ohio State University (United States) [presenting]
Paul Wiemann - The Ohio State University (United States)
Thomas Kneib - University of Goettingen (Germany)
Abstract: Geophysical processes often yield datasets that are spatially irregular, broadcasting a multi-scale character over space where observations are collected at different intensities in different areas. In such cases, models that assume the same dependence structure over the whole space - single-resolution stationary spatial models - can be inappropriate. This is the case for many forestry datasets, in which data is intensively collected in several distanced and relatively small-sized plots. In such cases, within plot data is typically characterized by large spatial ranges, while the spatial range between plots is relatively smaller. We develop a Bayesian spatial multi-resolution technique that models separately local and global processes, while allowing dimensions to interact. The resulting model is non-stationary and performs well for small local datasets, as well as for generally large datasets, through the use of Gaussian Markov random fields. The performance of our model is compared to that of standard single and multi-resolution models, for both simulated and real datasets. The resulting Bayesian model can be extended to fit spatio-temporal data.
