The on-going mountain pine beetle outbreak in British Columbia has reached historic proportions. Recently, management efforts at the local level shifted from exhaustive mapping of the infestation, to detection and mitigation of sites with minimal levels of infestation, creating an operational need for efficient and cost-effective methods to identify red-attack trees in these areas. High spatial resolution remotely sensed imagery has the potential to satisfy this information need. This paper presents the unsupervised classification of 4 metre IKONOS multispectral imagery, for the detection of mountain pine beetle red-attack, at sites with minimal infestation (< 20% of trees infested). A 4-metre buffer (analogous to a single IKONOS pixel) was applied to the red-attack trees identified on the IKONOS imagery in order to account for positional errors. When compared to the independent validation data collected from the aerial photography, it was found that 70.1% (lightly infested sites) and 92.5% (moderately infested sites) of the red-attack trees existing on the ground were correctly identified through the classification of the remotely sensed IKONOS imagery. These results demonstrate the operational potential of using an unsupervised classification of IKONOS imagery to detect and map mountain pine beetle red-attack at sites with minimal levels of infestation. Key words: mountain pine beetle, remote sensing, accuracy assessment, IKONOS, red-attack
A pixel shape index coupled with spectral information for classification of high spatial resolution remotely sensed imagery
