PREDICTING CINNAMOMUMRANDAIENSE HABITAT USING MULTIVARIATE STATISTICAL METHODS WITH GIS

摘要

Randaishan cinnamon trees (Cinnamomum randaiense Hay.), are an evergreen tree and widely distributed in central and southern Taiwan. Many studies have applied a geographic information system with statistical methods to model the habitat of the rare species of plant or animal, but not for the widely distributed tree species. Therefore, the species was chosen as a target for the study. The objective of the study was to predict the potential habitat of the tree species in the Huisun Forest Station in central Taiwan. Elevation, slope, aspect, terrain position, and vegetation indices derived from satellite images were accounted for in tree habitat evaluation and site search. The study developed the decision tree (DT), logistic multiple regression (LMR), and discriminant analysis (DA) models that related known tree sites to habitat characteristics and extrapolated the tree's unexplored sites in the study area. The accuracy of the DT model (95%) is slightly higher than that of the LMR (90%), and accuracies of the first two are much higher than that of the DA (83%); the three models are highly efficient in modeling habitat. Because the DT and LMR models greatly reduce the area of field survey to less than 10 % of the entire study area, they are more suited for predicting the tree's potential habitat. The results also suggest that the vegetation indices derived from SPOT-5 satellite images may not be able to improve model accuracy for widely distributed tree species due to the limitations of spectral resolution and spatial resolution with SPOT-5 imagery. Airborne hyperspectral data and LIDAR data will be used in follow-up studies to improve the model accuracy.