Merging Spectral Classes Based on Spatial Association (Neighbouring Class Frequency) Improves Correspondence between Spectral Classes and Air Photo-mapped or Ground-mapped Habitat Classes in a Mountainous Area of Southern British Columbia.

摘要

Correspondence is determined between a wildlife habitat inventory map derived from air photographs and ground samples (Timberland Consultants) and a spectral map derived from an unsupervised classification of Landsat TM imagery (GeoSense Consultants). A low degree of correspondence (Cramer's C = 0.205) exists between these two fundamentally different types of classifications. However, based on the Melody Index of Association (IA) between ground-based and spectral-based classes, certain spectral and habitat classes are found to be significantly associated. A better agreement of classified image with habitat map is obtained by using spatial context to agglomerate spectral classes into the same number of classes as on the wildlife habitat map. Neighbouring Class Frequency is an automated approach that identifies spectral classes that have a high frequency of association. Levels of correspondence between the habitat classification and spectral classes merged by the Neighboring Class Frequency procedure are greatly improved (Cramer's C = 0.8653).