Classification and Predictive Modeling of Plant Communities in the Gorges State Park and Gamelands, North Carolina

Abstract

A method of rapid field assessment and predictive modeling was developed to characterize vegetation communities of the Gorges State Park and Gamelands and to create predictive community maps for the area. This method placed an emphasis on locating rare communities using expert information, existing databases, aerial photography, and random encounters in efforts to provide information to researchers and park personnel about community locations. Approaches for classifying these communities were examined to identify which would provide suitable units for modeling community types. I sampled 102 field locations and assigned their vegetation to 16 different community types. Predictive community maps were generated using discriminant functions incorporating digital terrain data, including elevation, slope, relative slope position, terrain shape index, and weighted landform index. Three sets of discriminant functions were created to meet the different needs of persons interested in using these maps. Photo-interpreted cover classes were also including in the modeling process as filters. Map accuracies ranged from 65% to 75%, with those using only discriminant functions (without filtering) yielding higher accuracies.