Indiana forest cover mapping based on multi-stage integrated classification using satellite and in situ forest inventory data

Gang Shao, Purdue University

Abstract

Forest species classification through remote sensing data is a complex process, which usually is done either at a coarse level or with low accuracy. This study examines a multi-stage classification algorithm combining supervised and unsupervised classifications to classify forest areas in Indiana. Integrated classification makes the procedures automatic and reduces human errors. Splitting the classification into two steps increases the accuracy with limited ground data. In the first step, in which the Indiana state forest area is classified, the point plug-in classification algorithm is employed, because plenty of ground data are available. In the second step the classifying of the state forest including a surrounding 8km buffer, the ground data are insufficient to process the point plug-in classification approach. In this case, the polygon plug-in classification algorithm is used to realize the extended area classification at the second stage. The resultant land cover map has six tree species (conifer, mixed forest, oak and hickory, mixed oak and hickory/ hardwood, maple and other hardwood). The overall accuracy is 81.93%.

Degree

M.S.

Advisors

Shao, Purdue University.

Subject Area

Forestry|Remote sensing

COinS