Masters Thesis

An evaluation of the effectiveness of an unsupervised classification of Landsat imagery

An unsupervised classification of Landsat data covering the Warner Mountain Ranger District of the Modoc National Forest, California was analyzed. Field data were in the form of Timber Stand Improvement (TSI) and Compartment Inventory Analysis (CIA) records. General stand descriptions were determined for four of the six conifer classes using mean values for stand characteristics from these timber stand examinations and spectral signatures of the classes. Discriminant analysis was applied to the average timber stand composition information paired with the six classes, from the Landsat data, identified as "conifer" by the California Department of Forestry (CDF). High standard deviations in each of the measured stand variables and poor assignments of plots to classes (41.2 to 65.2 percent) in the discriminant analysis limited the detail of the class descriptions. The results of this study indicated that unsupervised classification was not adequate for conventional timber type mapping in the study area.