SUPERVISED CLASSIFICATION OF LANDSAT THEMATIC MAPPER IMAGERY IN A SEMIARID RANGELAND BY NONPARAMETRIC DISCRIMINANT-ANALYSIS

We used a nonparametric discriminant function in a supervised classifi cation of Landsat Thematic Mapper satellite imagery of a approximate t o 240,000-ha semi-arid region in the Snake River Plains, southwestern Idaho. First, agriculture pixels were classified by distance from the soil baseline and wafer pixels by the thermal band value. Next, succes sive nonparametric discriminant functions were used to separate grassl and and shrubland categories with subsequent classifications of vegeta tion within major classes. Accuracy in separating grass lands and shru blands was 80 percent and remained consistent relative to different th resholds in minimum percent ground cover defining shrublands. Within m ajor grassland and shrubland groups, we achieved 64 percent accuracy i n separating dominant vegetation classes. Distinction between density categories of vegetation based on percent ground cover was not possibl e in our study.