Masters Thesis

Change Detection: Using Satellite Imagery To Detect Invasive Shot Hole Borer Infestations Within Southern California Riparian Forest

Two species of invasive ambrosia beetles, collectively referred to as Invasive Shot Hole Borer (ISHB), began to infest Southern California’s native riparian forests in 2015. An estimated 95,791 willows had died from ISHB infestation by 2017, but the current extent and severity of ISHB infestations within San Diego County is relatively unknown. Finding infestations with ground-based monitoring is time consuming and incomplete, whereas use of Landsat data covers the entire County, and in principle could detect and monitor outbreaks much more effectively, provided that infested riparian vegetation produces a consistent spectral signature. The goal of this analysis was to correlate individual Landsat 8 bands and greenness, wetness, brightness and NDVI indices to document ISHB infestations. I used a linear analysis to generate a supervised classifier of riparian vegetation into infestation categories and evaluated performance with accuracy assessment statistics. I generated baseline models from three sources of training data to evaluate variation in spectral data that may normally occur during non-infested years due to landscape and year effects. Baseline models allowed for selection of variables that contributed to differences due to ISHB infestation, rather than natural variation. After eliminating variables that contributed to normal, non-ISHB yearly differences, results suggest differences between non-infested and infested riparian forest were detectable within Band 3 during the Post-flight season (Kappa-Cohen 42.5%). Classifying TRV forest in a single year improved model detection, although this study shows it is still necessary to evaluate previous years landscape scale differences with baseline models to assess performance. I found strong support for detecting differences between riparian forest with high rates of damage, to forest with low rates of damage within post-flight season Brightness (Kappa-Cohen 62.6%), and pre-flight Season Band 7 (Kappa-Cohen 60.2%). Results from a San Diego County wide analysis suggested site spectral wavelengths differed considerably within both baseline and treatment models, and I was unable to isolate diagnostic spectral wavelengths associated with ISHB at the county landscape level. In summary, ISHB infested forest was detectable within unique spectral wavelengths of Landsat 8 imagery. However, baseline models were required to account for landscape and year differences that may normally occur. As a management tool this analysis may be successful in identifying heavily damaged forest, which may not be easily accessed or surveyed, but not to detect early infestations.