Forest Harvest Mapping along the Appalachian Trail using Supervised Decision Tree Classifiers and Landsat Imagery

David Potere, Boston University

Multitemporal Kauth-Thomas transforms of thirty-two NASA GeoCover Landsat images provide inputs to a supervised decision tree classification of 1990-2000 forest harvest along the Appalachian Trail. A 16km-wide corridor centered on the Trail encloses a 3.8 million ha study area—the largest fine-resolution decision tree change detection to date. Including both single-date and multitemporal Kauth-Thomas components during image interpretation and classification significantly improved efficiency. Nested-hierarchical segmentation based on six-band, two-date imagery improved map accuracy and accelerated the editing process.

Between 1990-2000 the Appalachian Trail corridor lost 75,115 ha to forest harvest, 2.5% of total forest cover. A novel corridor segmentation and normalization scheme facilitated interpretation and analysis of the spatial distribution of these harvests, promising to improve the effectiveness of both the National Park Service and the Appalachian Trail Conference as they strive to preserve the Trail for future generations against the pressures of development and logging.

Abstract Author(s): David Potere