Classification of landslide surfaces using fully polarimetric SAR: examples from Taiwan

Landslides typically denude forested hillsides, changing dominant scattering from multiple- to single-bounce. SAR polarimetry can be used to determine the dominant scattering mechanism of the terrain on a per pixel basis. Thus, hillslopes affected by landslides may be distinguished easily from adjacent, unaffected hillslopes. Based on the polarimetric eigenvalue decomposition and the relative amplitudes of scattering matrix terms, dominant scattering mechanisms can be determined. These mechanisms - odd- or single-bounce (bare surface), double- or even-bounce, and volume (diffuse) scattering - describe the total radar returns for each pixel. The modified approach of Kim and van Zyl (2001) include the radar vegetation index, pedestal height and entropy, all representative of forest. Our analysis follows Kim and van Zyl to reveal that at L-band, the landslide scar areas of the Tsaolong and Mt. Jou-Feng-Err mega-slides are dominated by single-bounce scattering and the surrounding forested regions are dominated by volume scattering. In addition, fully polarimetric SAR is sensitive to reforestation in tropical jungle environments as demonstrated by comparison with a panchromatic Indian Remote Sensing (IRS) image. Inversion of SAR polarimetry for terrain scattering mechanisms offers a new approach for landslide mapping that has significant operational advantages over traditional methods that rely on optical sensors for rapid response to and management of landslide disasters.