•  
  •  
 

Abstract

The classification and mapping of land cover provides fundamental information about the characteristics, activities, and status of specific areas on the earth's surface. The quality of the final classification is critical in providing accurate information for ecologists and resource managers in decision-making and for developing a landscape-level understanding of an ecosystem. A land cover classification was developed for 5 research watersheds in Garland and Saline counties in Arkansas using 2002 LANDSAT7 Enhanced Thematic Mapper Plus (ETM+) satellite imagery. The supervised classification was based upon 146 training areas identified from reference data and then applied to the imagery using the maximum likelihood classification algorithm. The unsupervised classification used an Iterative Self-Organizing Data Analysis Techniques (ISODATA) algorithm to classify the imagery into 300 spectral classes which then were identified from reference data. Data from 171 field locations were used to assess the accuracy of the final classifications using an error matrix. The supervised classification had an overall accuracy of 74.85% compared to 40.94% for the unsupervised classification. However, the dense canopy pine plantation class, which comprises 10.69% of the total area of the watersheds (1,216.69 ha), was more accurately classified in the unsupervised classification (64.29%) than the supervised classification (43.86%). The unsupervised classification of dense canopy pine plantation was incorporated into the supervised classification to produce a final integrated classification with an improved overall accuracy of 76.61%. We found that, where greater accuracy is desired, both classification methods should be used and the results integrated to utilize each method's strengths.

Download

Share

COinS