Characteristics of tropical tree species in hyperspectral and multispectral data

Abstract

Remote sensing has been hailed as a promising technology to provide spatially explicit information on tree species distribution. Such information is of high value for ecologists and forest managers, particularly in tropical environments in which it is acquired by costly field inventories performed at the plot level (∼1 ha). Over the last decade, hyperspectral sensors, usually on board airborne platforms, have been successfully employed for tropical tree species classification. Most of the studies focused on the improvement of the classification accuracy, without examining the sources of variability in canopy reflectance that contribute to discriminate among species. Trees of different species usually feature distinct crown structural and chemical characteristics defined by size, shape, leaf density, branching, pigment and nonpigment biochemical constituents, among others. Understanding the role of these characteristics on tree species discrimination is important to improve and optimize the use of hyperspectral data. Here, we show how spectral characteristics of tree species, from highly diverse tropical forests, are related to their chemical and structural properties. These spectral characteristics were retrieved by spectral mixture and feature analysis, using narrow-band data acquired in the optical domain (450–2400 nm). Finally, we assessed how the spectral resolution affects the classification accuracy and spectral separability.