Classification of tropical trees growing in a sanctuary using Hyperion (EO-1) and SAM algorithm.

摘要

Tropical forests are one of the richest sources of biodiversity and are well known for their ecosystem services. There is a pressing need to monitor the rate and extent of changes in forest cover of countries like India for efficient planning and management leading to sustainable development. Imaging spectroscopy is one of the newer techniques adopted for species-level discrimination. Of the available sensors, spaceborne ones are cost-effective and are more appropriate for monitoring in countries like India. The present study aims at classifying tropical trees using Hyperion (EO-1) and SAM (Spectral Angle Mapper) algorithm. The study was conducted in the Shoolpaneshwar Wildlife Sanctuary (SWS), Narmada District, Gujarat, India. Hyperion data were obtained during October 2006 when the vegetation was lush green. Field survey was done coinciding with data acquisition time. The tree species identified for discrimination were Tectona grandis L., Dendrocalamus strictus Nees., Mangifera indica L., Madhuca indica J. F. Gmel. and Ficus glomerata Roxb. Hyperion data were preprocessed. End-member spectra for each species were selected and used as library spectra for the classification. SAM was performed for the entire spectrum, VIS-NIR region (1-90 bands), SWIR-I region (103-136 bands), SWIR-II region (159-195 bands), 1-10 MNF and 1-15 MNF bands. Overall accuracy assessment (OAA), kappa coefficient and user's and producer's accuracy were calculated. SAM classification with 196 bands (full-spectra) of Hyperion data gave 51% OAA for the five tropical trees selected. The obtained OAA was appropriate looking at the pattern of vegetal cover and also of the sensor used. Partition analysis of the spectrum indicated superiority of VIS-NIR region for classification. SWIR-I and II did not fare well because of the biophysical state of vegetal cover. SAM showed the highest accuracy (59.57%) for spectra of 1-10 MNF bands. Higher accuracy using MNF band combination indicated the potential of MNF transformation to increase classification accuracy of tropical trees by reducing data dimensionality. Our study indicates that homogeneity in the vegetal cover is a critical aspect for classification in the tropical areas. We conclude that SAM is an appropriate method for classifying Hyperion data of the tropics. With the reported densities for Tectona and Dendrocalamus, Hyperion is found to be an appropriate sensor for monitoring.