Analysis by Wavelet Frames of Spatial Statistics in SAR Data for Characterizing Structural Properties of Forests

• Published in: IEEE transactions on geoscience and remote sensing 2009-02, Vol.47 (2), p.494-507
• Main Authors: De Grandi, G.D., Lucas, R.M., Kropacek, J.
• Format: Article
• Language: English
• Subjects: Applied geophysics; Backscatter; Degradation; Discriminant analysis; Earth sciences; Earth, ocean, space; Exact sciences and technology; Extraterrestrial measurements; Forest structure; Information analysis; Internal geophysics; Linear discriminant analysis; Phased arrays; spatial statistics; Statistical analysis; Statistics; Studies; Synthetic aperture radar; Synthetic Aperture Radar (SAR); Wavelet analysis; wavelet frame
• ISSN: 0196-2892; 1558-0644
• DOI: 10.1109/TGRS.2008.2006183

Spatial statistics (texture) in SAR backscatter data of forested areas bears information on structural and geometric properties that could be useful in mapping forest extent, species type, and stages of regeneration or degradation. Based on a previously published theoretical approach in deriving texture measures from SAR data using wavelet frames, experiments are reported that aim to characterize, from a purely observational point of view, wavelet texture measures' sensitivity with respect to target structural properties and SAR configurations. Suitable analytical tools are introduced to represent dependences in the combined space-scale-polarization domain through signatures that condense information in graphical form. Moreover, class separability, afforded by wavelet texture measures in a supervised classification setting and based on the Fischer linear discriminant analysis, is considered. This paper focuses on two structurally different forest types (tropical rain forest in the Central Africa Congo Floodplain and mixed-species wooded savanna in Queensland, Australia) and uses data from orbital radars, particularly from the Japanese Advanced Land Observing Satellite Phased Arrayed L-band Synthetic Aperture Radar. The analysis indicated that textural information from spatial statistics can provide, in some cases, better class separability in forest mapping with respect to one-point statistics, although spatial resolution in texture products is reduced. However, dependences of texture measures on the polarization state are detected, particularly in forests where a greater diversity of scattering mechanisms occurs.