Land Cover-level Vegetation Mapping using AVIRIS
Hyperspectral data have proven more effective for vegetation mapping than multispectral data due to their rich spectral content. This chapter discusses 4-m Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) imagery was applied to map land cover-level vegetation types in the central Greater Everglades using modern machine learning classifiers and object-based mapping techniques. It examines the value of hyperspectral imagery for wetland mapping, classification results were compared with the application of multispectral Ikonos imagery. After data dimensionality reduction, the Minimum Noise Fraction imagery was segmented to produce image objects for object-based mapping. The reference objects of AVIRIS were first converted from polygon features to point features in ArcGIS, and then the point features were spatially joined to the segmentation polygons of Ikonos. AVIRIS hyperspectral imagery produced acceptable accuracy from the Random Forest and Support Vector Machine classifiers. The Kappa value describes the proportion of correctly classified validation samples after random agreement is removed.