Thesis

Using remote sensing to predict invasive plant species distributions in wetlands

Exotic plant invasion is a major environmental and ecological concern and is a particular issue for wetland ecosystems. I present statistical models that predict the locations of three exotic invasive plants (Arundo donax, Eucalyptus species (Eucalyptus globulus, Eucalyptus calmodulensis), and Tamarix ramosissima) that invade wetland areas throughout San Diego County based on their spectral signatures. I used three images that differed in their spectral resolutions and spectral coverage: Color-infrared (1m pixel size, infrared, blue and green bands), high resolution true color imagery (lm pixel size, red, blue and green bands), and hyperspectral Landsat imagery (30m pixel size, blue, green, red, near infrared, (2), mid infrared, and thermal infrared bands). For each invasive plant, three well-known multivariate statistical analyses, Discriminant Function Analysis (D.F.A.), Quadratic Discriminant Function Analysis (Q.D.F.A.), and CART, were used to identifY the models that best separated invasive plants from surrounding vegetation. A predictive accuracy analysis was preformed for each model by predicting which points should contain the invasive species based on their spectral values, then comparing these predictions to the actual presence or absence of the species. The best model for both Arundo and Eucalyptus species was obtained from the Q.D.F.A. using spectral values calculated from a combination ofNAIP and Landsat wavebands. CART using spectral values obtained from Landsat imagery produced the best results for Tamarix. Past studies show that plant species do in fact have distinct spectral signatures however; further investigation of classification techniques for this study is needed in order to create a more successful predictive model for each invasive plant species. Key-words: Discriminant Function Analysis, CART Model, hyperspectral imagery, invasive plants, predictive model, spectral signatures.

Relationships

Items