Spectrally Driven Classification of High Spatial Resolution, Hyperspectral Imagery: A Tool for Mapping In-Stream Habitat

Streams represent an essential component of functional ecosystems and serve as sensitive indicators of disturbance. Accurate mapping and monitoring of these features is therefore critical, and this study explored the potential to characterize aquatic habitat with remotely sensed data. High spatial resolution, hyperspectral imagery of the Lamar River, Wyoming, USA, was used to examine the relationship between spectrally defined classes and field-mapped habitats. Advantages of this approach included enhanced depiction of fine-scale heterogeneity and improved portrayal of gradational zones between adjacent features. Certain habitat types delineated in the field were strongly associated with specific image classes, but most included areas of diverse spectral character; spatially buffering the field map polygons strengthened this association. Canonical discriminant analysis (CDA) indicated that the ratio of the variability among groups to that within a group was an order of magnitude greater for spectrally defined image classes (20.84) than for field-mapped habitat types (1.82), suggesting that unsupervised image classification might more effectively categorize the fluvial environment. CDA results also suggested that shortwave-infrared wavelengths were valuable for distinguishing various in-stream habitats. Although hyperspectral stream classification seemed capable of identifying more features than previously recognized, the technique also suggested that the intrinsic complexity of the Lamar River would preclude its subdivision into a discrete number of classes. Establishing physically based linkages between observed spectral patterns and ecologically relevant channel characteristics will require additional research, but hyperspectral stream classification could provide novel insight into fluvial systems while emerging as a potentially powerful tool for resource management.