Improved Classification of Forest Vegetation in Northern Wisconsin Through a Rule-Based Combination of Soils, Terrain, and Landsat Thematic Mapper Data

Abstract

Abstract Landcover classifications which use evidential and hierarchical decision rules to combine satellite image, soils, and terrain data were compared to standard Landsat Thematic Mapper (TM) based maximum likelihood classifications (MLC) under northern Wisconsin forest conditions. Maximum likelihood classifications employed a visible, near-infrared, and mid-infrared band of TM image data, while a rule-based (combined) approach was used to integrate these data with soil texture information, terrain position, and soil-plant relationships for landcover classification. Comparisons were made for classifications by two analysts for each of two 300 km2 study areas. Accuracies from the combined data method were higher than the MLC method for all classifications. Accuracies averaged 89% for the combined method and 73% for the standard method when areas were classified into 13 landcover classes corresponding to Anderson levels II and III. Pairwise differences of Cohen's Kappa values for the two classification approaches were significantly different (P < 0.10) for all combinations of study area and analyst. For. Sci. 38(1):5-20.