Using hyperspectral satellite imagery for regional inventories: a test with tropical emergent trees in the Amazon Basin

Abstract

Questions: Understanding distributions of tree species at landscape scales in tropical forests is a difficult task that could benefit from the recent development of satellite imaging spectroscopy. We tested an application of the EO-1 Hyperion satellite sensor to spectrally detect the location of five important tree taxa in the lowland humid tropical forests of southeastern Peru. Location: Peru, Departamento de Madre de Díos. Methods: We used linear discriminant analysis with a stepwise selection procedure to analyze two Hyperion datasets (July and December 2006) to choose the most informative narrow bands for classifying trees. Results: Optimal channels selected were different between the two seasons. Classification was 100% successful for the five taxa when using 25 narrow bands and pixels that represented >40% of tree crowns. We applied the discriminant functions developed separately for the two seasons to the entire study area, and found significantly nonrandom overlap in the anticipated distributions of the five taxa between seasons. Conclusions: Despite known issues, such as signal-to-noise ratio and spatial resolution, Hyperion imaging spectroscopy has potential for developing regional mapping of large-crowned tropical trees.