Since 1999, we have been doing predictive mapping of mountain pine beetle (MPB) attack using multi-temporal Landsat (LS) data in central British Columbia (BC). In 1999, spectral unmixing of LS data was done over 19 map sheets (194,000 ha) east of Tweedsmuir Park. An overall prediction accuracy of 71.5% for MPB attack at the polygon level was achieved. Then, in 2001, 234,000 ha were predictively mapped in near Williams Lake, BC, using tasseled-cap transformations and spectral unmixing. The LS data were from July 1999 and August 2000. Independent ground verification in 2001 by 3 consultants indicated a prediction accuracy of 69.8 for MPB attack at the pixel level, and 86.8% for forest health issues. In 2002, MPB predictive mapping was done over 1.8 million hectares for 6 forest companies for their operational planning using data from July 1999 and June 2002. In 2002/2003, 600,000 ha were again predictively mapped for operational applications using LS data from August 2000 and 2002. In March 2003, accuracy and errors of omission and commission for the pixel level mapping of MPB attack were assessed. Preliminary results indicated that accuracy varied with Biogeoclimatic zone/variant level. Accuracy was best in the SBPSmk variant at 82%, and worst in the IDFdk3 at 65% correct. Errors of omission with 1 pixel buffering ranged from 18% to 52% and were attributed to missing single trees and green attack. The accuracies achieved using remote sensing were adequate to meet operational planning needs of the forest companies at substantially reduced costs.
Species 1: Coleoptera Scolytidae Dendroctonus ponderosae (mountain pine beetle)
Keywords: bark beetle attack, spectral unmixing
Back to Section F Symposium: Applications of Remote Sensing in Entomology
Back to Section Symposia
Back to The 2003 ESA Annual Meeting and Exhibition