Automatic classification of vegetation and land degradation with large-scale, color infrared, aerial imagery

The use of aerial, color infrared, digital imagery was examined as a tool for monitoring land degradation from training on the Fort Bliss military reservation in New Mexico. Principal components analysis and raster correlation matrices were utilized to interpret image variance. High color band correlation and a predominance of the spatial aspect of imagery was determined. A color space conversion and low pass filtering in frequency space were employed as pre-classification image processing techniques to enable vegetation patch delineation. Ecologically significant classes relating to vegetative disturbance and recovery stages were automatically extracted with an unsupervised classification algorithm. Range ecology models of Seral Stages and Increaser/Decreaser/Invaders were used to label classification results.