Landsat Computer-Aided Analysis Techniques for Range Vegetation Mapping

Abstract

Landsat computer-aided analysis techniques were used to map the sagebrush-grass vegetation of northern Nevada. A final Landsat digial clacation resulted in 14spectral classes representing 8 range plant communities. Classification accuracy for all sample plots was 86.4%, with individual class accuracies ranging from 77.8 to 95.4%. Classification methods included supervised, unsupervised, and guided clustering techniques using a maximum likelihood classifier. Rangeland inventory techniques have been subject to question and controversy since the beginning of range management. The problems include cost, adequate trained manpower, the requirement to inventory vast areas and the obtaining of an adequate sample. Remote sensing techniques have often been suggested and promoted for doing basic range inventories. The repetitive availability, relatively low cost per unit area, and digitized format of the Landsat data make such information of potential interest to range managers. This study was designed to measure the success of Landsat computer-aided analysis techniques for range vegetation mapping in Northern Nevada. Landsat digital data has seen only limited application on rangeland despite its potential for providing large quantities of vegetation mapping data at reasonable cost. Resolution limitations of the digital data (.42 ha or approximately 1.12 acres) along with complexity, diversity, and heterogeneity of range vegetation have tended to discourage its use. Several researchers have evaluated the application of Landsat digital data for mapping range and arid land vegetation (Daus 1975, Maxwell 1976, Tueller et al. i978, Everitt et al. 1979, Todd et al. 1980, Everitt et al. 1981). Maxwell (1976) inventoried vegetation types, range condition, and green biomass on grasslands in Colorado using a supervised classification technique. He concluded that Landsat was a very useful inventory tool on grasslands. Tueller et al. (1978) used Landsat digital data to map various arid land vegetation types in Australia. Todd et al. (1980) classified various densities of pinyon-juniper on two different geologic types on the Shivwits Plateau in Arizona. Misclassification was the result of low canopy and high bare ground cover. Bonner and Morgart (1980) described an operational procedure for arid land vegetation inventories and the sampling units required for accurate classification. Recently Everitt et al. (1981) used digital pattern recognition techniques and a maximum likelihood ratio classification and found a highly significant correlation (r2= 0.997) between air-photo and computer-estimated area of 5 land use categories for a June Lanasat-2 scene. Conditions were not significant for an August overpass, suggesting the importance of selected dates to reduce misclassification. Brush, mountain shrub/juniper, conifer, meadow, rock or bare ground and water were readily identified on rangeland near Susanville, Calif. (Daus 1975). Classification problems occurred at ecotones and areas that contained mixes of vegetation types. Areas of low canopy cover were difficult to classify because of the spectral dominance of soil background. Sub-class classification problems Authors are graduate research fellow and professor, Division of Renewable Natural Resources, University of Nevada Reno. This manuscript is published with approval of the Director, Nevada Agriculture Experiment Station as Journal Series No. 568. Manuscript received June 21, 1982. occurred in the big sagebrush communities with high proportions of bitterbrush, rabbitbrush, and other sagebrush species.