Effects of spectral transformations in statistical modeling of leaf biochemical concentrations

Abstract

The prediction of leaf biochemical concentrations with hyperspectral data is one of latest research directions in hyperspectral remote sensing. Statistical modeling being a convenient and common-used method, spectral transformations are always performed as its preprocess. We discussed several usual transformations including full-band based transformations such as reciprocal, logarithm, and derivative spectra, and one-absorption-feature based transformation: continuum removal. The effects of those transformations on the prediction of C/N were compared using correlation analyses and stepwise regressions. Results show that the effect of continuum removal is the best, which is physically based and not site-specific at all.