Forest cover maps derived from Landsat Thematic Mapper (TM) imagery have become a popular layer in geographic information systems (GIS). A common classification scheme, the Anderson Level II Deciduous Forest, Evergreen Forest, and Mixed Forest cover classes provides a sufficient documentation of the current inventory on the ground. Using TM imagery in a classification technique (like maximum likelihood) with only spectral information from one date, the three Anderson classes often have degraded accuracy (;A simplified method is developed and tested for its efficacy to improve the Anderson Mixed Forest classification accuracy. Huntington Forest and Heiberg Forest are used as test sites. A spectrally classified image is augmented with digital data representing slope, aspect, elevation, and soil drainage. A spatial model which relates these site factors to forest development, by way of a plausibility index, is implemented such that image pixels whose forest cover class is sufficiently implausible (lowest index score), given the topoedaphic constraints, are assigned to a more plausible forest cover class (highest index score).;Results indicate a small, statistically insignificant, increase in classification accuracy. With Anderson Level II, species with varying site tolerances will be included in the same broad class. A simplified plausibility index may not be enough to overcome this within-class variability. This spatial model may be inappropriate for areas which have undergone management practices that create a forest with some departure from natural growing conditions. |