Accessing a Geographic-Object Based Image Analysis (GEOBIA) Workflow for Identifying and Classifying Individual Trees by Species in an Oak Savanna Woodland
Kenya Creer, California State University, Long Beach, Department of Geography
Through the acquisition, processing, and analysis of earth imagery, remote sensing has been at the forefront of mapping and monitoring biodiversity. While remote sensing has shown potential for mapping vegetation communities at a wide range of scales, success in accurate species-level classification has been uneven under the per-pixel paradigm. With the ascension of Geographic Object-Based Image Analysis (GEOBIA), concepts and tools have emerged and re-emerged to encourage the integration of expert knowledge with appropriate data and methodologies, allowing for more detailed land cover mapping. In particular, knowledge of plant phenology and morphology combined with segmentation concepts have been useful in species-level classification of complex environments, like woodland-grasslands. However, in an oak savanna woodland, where tree canopy structures are complex, species level classification of individual trees crowns (ITCs) is not guaranteed. Using well-established methods for identifying individual trees, isolating tree crowns, and classifying tree crowns by species, presents many challenges for analysts who study sites differ from those where these methods developed and gained fame. Often, much tweaking is necessary to make these workflows viable, for while the study goals are the same, the context is not. This study will apply and access a local maxima detection and region growing technique – a popular GEOBIA workflow for ITC mapping - to help identify individual oak trees by species at a study site on River Ridge Ranch, Tulare County, California.