PARTNERS: Yale University School of Forestry and Environmental Studies

SUMMARY: In July 2006, research was proposed that would allow for the identification and delineation of individual eastern white pine and eastern hemlock trees on high spatial resolution optical imagery. This research is unique in that it will derive contextual information concerning the morphological patterns within the crown of individual trees using local indicators of spatial association (LISAs). The efficacy of LISAs to detect unique patterns among various overstory genera in dense mixed temperate forests will be assessed using spatially explicit field research. Three phases of research will be conducted to achieve the goals of genus identification of individual trees from high spatial resolution imagery. Phase one will consider errors associated with the geolocation of physical objects such as trees within a forest to their feature object counterpart within the image; phase two will modify existing automatic feature delineation algorithms to parameters specifically suited for coniferous and deciduous crown recognition; and phase three will build context driven classifiers based on models of the spatial associations derived from local indicators.  

STATUS: Ongoing

PROGRESS: Based on research conducted thus far, a software application known as Leaf2Landscape is currently under development. Leaf2Landscape utilizes digital information existing from the leaf to landscape scale in order to provide automatically derived individual tree metrics. High resolution digital images of forest canopies are automatically delineated using one of three possible image segmenting techniques. Additionally, tessellation techniques developed for this application of Light Detection and Ranging (LIDAR) data xyz coordinates can be used to further augment crown delineation techniques. After individual tree crowns are automatically detected and delineated, within crown spatial associations of spectral intensities, crown geometry and second order statistics on high and low pixel clusters are compiled to build an individual tree crown feature space. These feature space portfolios for each crown detected in the digital imagery and augmented with LIDAR data information are used to train support vector machine classifiers to discriminate among tree genera. Supplemental LIDAR data or regressions of image-derived crown areas are used to determine individual tree diameters at breast height.

LINKS: Yale University School of Forestry and Environmental Studies

CONTACT: Andrew Niccolai, Doctoral Candidate, Yale University School of Forestry and Environmental Studies, andrew.niccolai@yale.edu or (203) 432-5411