Abstract for presentation at The 13th Australasian Remote Sensing and Photogrammetry Conference

Integration of vegetation indices for automated landslide recognition

  • Wing Yip Lau, School of Surveying and Spatial Information Systems, The University of New South Wales, Australia
  • Dr Linlin Ge, School of Surveying and Spatial Information Systems, The University of New South Wales, Australia
  • Dr Xiuping Jia, School of Information Technology and Electrical Engineering, The University of New South Wales,Australian Force Defense Academy, Australia

    • Landslide assessment using optical remotely sensed data often relies on evaluating the change of surface greenness resulted from slope failure, using, for example, Normalised Difference Vegetation Index (NDVI) or Principal Component Analysis (PCA). However, this approach cannot fully utilise the surface change information related to landslides. We developed a landslide recognition procedure by integrating change information of surface greenness with surface brightness and surface wetness. These change components were derived from Vegetation Indices (VIs). The accomplishment of this analysis was fulfilled by two change detection techniques.
    One was image differencing based approach. NDVI, Kauth-Thomas transformation, Modified Soil Adjusted Vegetation Index Two and Normalised Difference Water Index were generated first, and their differencing results were then stacked. Each of combinations contained three change components representing brightness, greenness and wetness. The other change detection algorithm was bi-temporal linear data transformation, whereby multitemporal Kauth-Thomas (MKT) transformation was adopted. This transformation generated three surface change components related to brightness, greenness and wetness.
    Two Landsat TM images (Path/Row: 90/84) acquired on 30/7/1998 and 31/8/1998 were used for this study. The landslide mapping was conducted using a Maximum Likelihood classification (MLC) technique. The best landslide mapping performance was yield by the image differencing method using brightness and wetness components of Kauth-Thomas transformation and NDVI. Its omission error (i.e. the percentage of actual landslide pixels which were not detected) and commission error (i.e. the percentage of change pixels identified which were not landslide) were 14.4% and 3.3%, respectively, with a strong agreement (KHAT = 88.8%). While MKT transformation method did not provide the best mapping performance in this study, this technique is easy to be implemented since it is free of radiometric normalisation and requires less computational time.

    Conference Organiser - ICMS Pty Ltd