Characterization of the diminishing accuracy in detecting forest insect damage over time

Abstract

The goal of this project was to determine the ability to detect forest insect disturbances occurring over a 6 year period using a remote sensing change detection approach. The study area in central British Columbia, Canada, has been experiencing an epidemic outbreak of bark beetles. The actual location and number of trees attacked by mountain pine beetle (Dentroctonus ponderosae Hopkins) were determined by annual surveys using a helicopter and a global positioning system (GPS). In this study, mountain pine beetle red-attack trees, infested between 1995 and 2001, were detected with an enhanced wetness difference index (EWDI), which was created using a 1995 and 2001 Landsat image pair. Red-attack damage was detected with an accuracy (true positive rate) of 74% using all years of the helicopter GPS survey data as validation. Assessments of the classification were subsequently undertaken that compared the EWDI-derived red-attack locations to each year of available validation data. The results of this analysis showed that recent red-attack damage was detected with greater accuracy than older red-attack damage (with an accuracy decline of approximately 15% over the 6 year period). The greatest accuracy was obtained for the most recent 2 years of attack, namely 2000 and 2001, with a red-attack detection accuracy of 81%.