Identifying ’Redtops’: Classification ofSatellite Imagery for Tracking Mountain Pine Beetle Progression through a Pine Forest

Mountain pine beetles (Dendroctonus ponderosae Hopkins) are a pest indigenous to the pine forests of the western United States. Capable of exponential population growth, mountain pine beetles can destroy thousands of acres of trees in a short period of time. The research reported here is part of a larger project to demonstrate the application of, and evaluate, differential equation models for mountain pine beetle progression through pine forests. The study area is the Sawtooth National Recreation Area in Idaho. To provide input parameters to the mathematical models, and to measure the bark beetle impact (redtopped pines), IKONOS satellite imagery was used to classify the vegetation of the study area. Five classifiers - linear discriminant analysis, quadratic discriminant analysis, k-nearest neighbor discriminant analysis, classification trees and random forests - were applied to raw and transformed multispectral and panchromatic satellite imagery, with and without an elevation variable. Quadratic discrimininant analysis and random forests proved to be the best classifiers as measured by cross-validated error estimates, with overall misclassification rates were about 12% without elevation, and about 5% when elevation was included. Redtops were relatively easy to identify, with misclassification rates of about 5%-6%, but green lodgepole pine and Douglas fir were relatively difficult to discriminat between and had much higher misclassification rates