Spatio-temporal mapping and modeling of a new forest disease spread using remote sensing and spatial statistics

In central coastal California, a recently discovered pathogen Phytophthora ramorum has been killing hundreds of thousands of tanoak, coast live oak, and black oak trees. This forest disease referred to as Sudden Oak Death (SOD) has attracted attention from the public, government and academia. Monitoring the disease distribution and understanding the disease mechanisms are important for disease control and management. In this dissertation, I developed a spatio-temporal approach to mapping and modeling the SOD spread in California using remote sensing and spatial statistics.; This dissertation seeks to quantify the disease spread over a range of scales using multi-temporal high spatial resolution airborne imagery. The work has three components: multi-temporal image registration, spatio-temporal classification, and spatial pattern analysis of disease dynamics. First, I developed an automated algorithm to register multi-temporal airborne images, which are characterized by complex geometric distortion with respect to one another. In this algorithm, large amounts of evenly distributed control points on regular grids were first derived from area-based methods. The control points with outliers removed were then applied to local transformation models. The results showed that the combination of area-based control point extraction with local transformation models is successful for geometric registration of airborne images with complex local distortion. Second, I developed a spatio-temporal classification algorithm to map mortality patterns from the accurately co-registered multi-temporal images. This algorithm is based on Markov Random Fields and Support Vector Machines and explicitly integrates spectral, spatial and temporal information in multi-temporal high-spatial resolution images. The results indicated that the algorithm achieved significant improvements over non-contextual classifications. Third, I applied both univariate and multivariate spatial point pattern analysis methods to quantify the mortality patterns using the mapped point patterns. The results from the univariate point pattern analysis showed that all the SOD point patterns are significantly clustered at different scales and spatial extents, revealing that the underlying mortality process consists of both first order trend and second order clustering. The results from the multivariate point pattern analysis showed that there exist strong attractions within multi-temporal SOD point patterns and between SOD and its major foliar host, California bay.