Remote Sensing Techniques for Classifying the Habitat of an Endangered Bird Species on Mauna Kea

Abstract

This study investigates whether high-spatial resolution satellite imagery (4-meter IKONOS imagery) is sufficient to differentiate between two Hawaiian forest species--Sophora chrysophylla (mamane) and Myoporum sandwicense (naio). These two tree species are critical to the survival of Loxioides bailleui (palila), an avian species that is listed as endangered under the 1973 Endangered Species Act. This study utilizes four types of supervised classification (maximum likelihood, mahalanobis, parallelepiped, and minimum distance) in association with texture filters and vegetation indices, to classify mamane and naio. The study also utilizes two different forms of decision tree analysis (classification and regression tree analysis and random forest analysis) in an attempt to differentiate between the two tree species. In doing so, the purpose was to find the best classification technique and to compare and contrast the results obtained by each classifier. Masked maximum likelihood classification of the IKONOS image produced the best result out of the six classifiers, returning a value of 79.40%. The addition of a fifth band calculated from the soil-adjusted vegetation index (SAVI), further increased the accuracy of MLC to 80.57%. Low spectral separability between mamane and naio and spatial constraints associated with the small crown size of mamane in comparison to the 4-meter resolution of the imagery, likely contributed heavily to the observed results. It is questionable whether these results will be of value for long-term mamane and naio conservation efforts, however, it is hoped that they will provide a baseline for future studies utilizing hyperspectral and airborne lidar technologies.