CHANGE DETECTION USING THE GRAMM-SCHMIDT TRANSFORMATION APPLIED TO MAPPING FOREST MORTALITY

Coordinate scores of appropriately rotated multitemporal TM pixel vectors directly measure scene change. A successful transformation requires initial identification of a subset of multispectral space containing all unchanged pixel vectors, and subsequent definition of axes which measure the displacement of pixel vectors from this space. Carefully selected pixel vectors guide the Gramm-Schmidt orthogonalization process to produce such a transformation. A Gramm-Schmidt analysis of a multitemporal Thematic Mapper image of the Lake Tahoe Basin produces components which correspond to identifiable scene characteristics, are easily interpretable, and correlate highly with field measurements of conifer mortality.