Investigation of leaf biochemistry by hierarchical foreground/background analysis

Abstract

A hierarchical procedure was developed for quantitative estimation of foliar chemistry from remote reflectance spectra. The authors based their analysis on a new methodology called Hierarchical Foreground and Background Analysis (HFBA) that derives sequentially a series of weighting vectors which simultaneously extract important discriminant features, in this case, leaf anatomy and chemical concentration at different levels of detection from the spectral information. In this study, they focused on the application of detecting carbon, cellulose, nitrogen concentrations, and water content. The goal of the derived vectors is twofold: 1) create a robust detection and classification system of constituent materials and 2) create a good information packing system that minimizes extraneous undesired interference, like noise, in the analysis. In their study, two data sets were examined: a fresh leaf (FL) data set, LOPEx, and a dry leaf data set, Blackhawk Island (BH), WI. The authors tested the robustness of the derived vectors with four other data sets: fresh leaf data from Jasper Ridge Biological Preserve (JRBP), Santa Monica Mountains (SMM), CA and dry leaf data from two ACCP sites Howland and Harvard Forest. The results support the robustness of the HFBA system and demonstrate an advantage in classification accuracy as well as in predicting the biochemical composition (subsequent levels) over classical forms of analysis that ignore effects of the nonlinear variation that contribute to reflectance at different (subpixel and spectral) scales, HFBA primarily deals with the spectral scaling issue.