AAIR Project: Sub-Task 4.3 Spectral mixture modelling


Contact:

Mário Silvio Caetano

Centro Nacional de Informação Geográfica

Rua braamcamp, 82 - 1 Dto

1250 Lisboa

Portugal

E-mail: mario@ cnig.pt


Description:

This is a methodology for spectral unmixing of tree canopy and understorey spectroradiometer signals and it takes into account non-linear effect of the radiant energy. The capability of the methodology for forest fuel cartography by remote sensing is evaluated.

Inputs

(1) Spectra measurements of mixed scenes in the study area

(2) Spectra characterisation of pure components that may exist in the scene, e.g. soil, shrubs, trees and forest litter.

Outputs

The output of the model consists of a set of proportions of each of the pure surface components, such as pine tree and shrub, existent in the area covered by the spectroradiometer.

Method

The spectral unmixing methodology is based on factor analysis (Huete, 1986), and it includes three steps: (1) decomposition of the spectral data matrix into an abstract eigenspectra matrix and abstract eigenvector matrices, by principal component analysis, (2) determination of the number of physically significant reflecting features present in the data set, by a stepwise procedure and (3) transformation of the abstract factors into physically-based factors, using target rotation (Malinowski, 1989). The methodology that is applied takes into account intimate mixtures of the surface components and consequently, non-linear effects of the radiant energy, i.e. multiple scattering.

The model is tested in a control experiment, where we simulate different proportions of trees, shrubs and soil, and in a real pine stand.

Uses

Determination of proportions of surface components by unmixing of spectral data acquired by spectroradiometers. We test the application of the model for forest fuel cartography.

References:

Malinowski, E.R. (1991), Factor Analysis in Chemistry, 2nd ed., John Wileys & Sons, Inc, New York.

Huete, A.R: (1986) Separation of soil-plant spectral mixtures by factor analysis. Remote Sensing of Environment, 19: 237-251.