| In response to a local and global need to quantify the current and future economic and ecological viability of forest landscapes, many countries and agencies are investing in long-term forest inventory and monitoring initiatives. As a result, the volume, complexity, and availability of forest inventory data is rapidly increasing. Given these data, researchers and resource analysts are asking a diverse array of important and complex questions that often require summarizing uncertainty at many levels (e.g., spatial, non-spatial, and temporal). While traditional approaches to data analysis are often ill-equipped to address these analytical needs, Bayesian modeling approaches enable the incorporation of rich spatial dependence structures and offer a highly flexible inferential framework capable of providing geostatistical methods to address complex data structures and multifaceted research questions. Further, recent advances in Markov chain Monte Carlo methods have contributed enormously to the popularity of spatial hierarchical models. This dissertation presents methodological advances in Bayesian hierarchical geostatistical modeling of multi-source forest inventory data. Within a model-based framework, multi-source forest inventory methods couple georeferenced forest inventory plot data with remotely sensed imagery and auxiliary variables to improve prediction of forest attributes and associated measures of uncertainty. The proposed methodologies focus on: (1) multivariate analysis---to address the common need to predict simultaneously several spatially dependent forest attributes of interest; (2) multi-resolution analysis - to expand modeling options for spatially dependent observations taken from forest inventory cluster plots; and (3) non-Gaussian response variable analysis---to permit spatial process modeling of binary response variables (e.g., forest and non-forest) through the addition of random spatial effects to generalized linear models. The proposed methodologies are illustrated with analyses that use remotely sensed variables and point-reference data from the USDA Forest Service Forest Inventory and Analysis program and other spatially explicit forest inventory datasets. |
