A statistical forest reflectance model

Abstract

The variability of forest reflectance among hemiboreal forests can be described with a few basis functions. Five basis functions describe almost 98% of variability of directional reflectance spectra in the optical spectral domain (400-1700 nm) in forest stands at the top of a canopy in nadir. A statistical forest reflectance model (SFRM) was developed, the input parameters of which are the forest parameters measured in the course of regular forest inventory. Nadir spectral reflectance of a forest stand is expressed in the SFRM as a linear combination of basis functions, the weights of which are linear combinations of the 15 stand parameters in the forest inventory database. Multiple correlations of the weights on the forest inventory parameters are determined separately for pine, spruce, and broadleaf forests. The basis functions are found from low altitude airborne measurements over managed forests in southeastern Estonia, where a forest management database is available. The model was validated against more than 3000 spectral signatures of forest stands from Sentinel-2 Multispectral Imager (MSI) measurements over a test site in southeastern Estonia. In most cases, the model predicts the forest reflectance spectrum at nadir with a relative error about 20-40%. The errors of reflectance values are less than 0.02 in most cases. The sole exception is the reflectance of broadleaf stands, which in near infrared bands of Sentinel-2 MSI is overestimated by 0.02-0.05.