Mapping the forests of Jefferson Proving Ground, Indiana using remote sensing, geographic information systems and classification trees

Classification trees were tested as a potential methodology for increasing the thematic detail of forest maps for the Indiana Gap Analysis Project. The study area used to test the methodology was Jefferson Proving Ground in southeastern Indiana. Multi-temporal Landsat TM data were integrated with digitized elevation and soils data using the classification tree approach. Results indicate that the integration of TM and GIS data provides greater classification accuracy than using either of these data sources independently. Classification trees also resulted in higher classification accuracies than those obtained with a maximum likelihood approach.;The research indicates that the major limitation of the classification tree approach to practical applications in forest mapping is the large learning sample size necessary to achieve reasonable classification accuracies. Out of a population of 56,114 individuals, a learning sample size of at least 31,823 cells was necessary to achieve a 100% forest cover type classification accuracy using only TM data. A learning sample size of at least 25,582 cells was necessary to achieve a 100% forest cover type classification accuracy using TM and GIS data in conjunction.