Abstract
Early detection of insect or pathogen infestations in forests would be useful to forest managers who want to make decisions that minimize timber losses. Typical methods of forest reconnaissance to detect infestations have included analysis of multispectral imagery. Multispectral imagery, however, often lacks the sensitivity to detect subtle changes in tree canopy reflectance because of physiologic stress from insects or pathogens. Most hyperspectral imaging has the sensitivity to detect subtle changes in canopy reflectance but lacks high spatial resolution to identify affected trees. Our study examined the use of subcanopy spatial resolution hyperspectral imagery for differentiating Douglas-fir trees attacked by the Douglas-fir beetle. Comparison of the accuracies of step-wise discriminant analysis and classification and regression tree analysis (CART) revealed that CART provided the best separability among tree health classes (93% overall) because of CART's ability to use different band combinations for each class. Predictive accuracy of the CART method was estimated through cross-validation of the dataset using a jackknife resampling technique. Overall classification accuracy was promising (69%), as was classification among healthy and attacked, but still living, trees (50–70%). The results of our study provide support that hyperspectral imagery might be used for detecting and mapping tree stress in Douglas-fir stands. Although the rapid progress of beetle infestation somewhat limited the ability to differentiate among tree stress classes, which might limit the utility of this approach for fast moving infestations, the results were well beyond what might be expected from alternative detection methods. Slower moving infestations would benefit from the use of hyperspectral imagery because a lower percentage of infested trees would be asymtomatic. West. J. Appl. For. 18(3):202–206.
Mapping native and invasive grassland species and characterizing topography-driven species dynamics using high spatial resolution hyperspectral imagery
