Remote sensing of tropical forest tree species composition in Papua New Guinea using multi-band spectral analysis

The rapid decline of rainforests due to human activities was the motivation for this study, which hopes to bring attention to a patch of rainforest in Papua New Guinea (home to some of the most pristine forests in the world). I utilized remote sensing to supplement previous field surveys in an attempt to classify parts of the caldera surrounding Lake Hargy (New Britain, Papua New Guinea) that were not surveyed. Five classification techniques were used, and spectral angle mapper (SAM) was found to be most effective due to the small size of the training samples, followed by ISODATA and decision trees (DTs). The other two classification techniques worked either improperly or not at all. Classification varied by image, which I attributed to seasonal changes in vegetation morphology in the area. In future studies, I hope to use higher resolution imagery to do a more detailed and consistent classification, as well as more (and larger) sample sites to increase accuracy. This will allow me to use other techniques within ArcMap and ENVI that were discussed but never used.