An expert system was developed to integrate all useful spatial information and help the interpreters determine the landslide and shaded areas quickly and accurately. The intersection of two spectral indices, namely the normalized difference vegetation index and the normalized green red difference index, as well as the first principle component of the panchromatic band, is employed to automatically determine the regional thresholds of nonvegetation and dark areas. These boundaries are overlaid on the locally enhanced image and the digital topography model to closely inspect each area with a preferred viewing direction. The other geospatial information can be switched on and off to facilitate interpretation. This new approach is tested with 2 m pan-sharpened multispectral imagery from Formosat-2 taken on August 24, 2009, for several disaster areas of Typhoon Morakot. The generated inventory of landslide and shadow areas is validated with the one manually delineated from the 25 cm aerial photos taken on the same day. The production, user, and overall accuracies are higher than 82%, 85%, and 98%, respectively. The fall in production and user accuracies mainly comes from the differences in resolution. This new approach is as accurate as the general approach of manual delineation and visual interpretation, yet significantly reduces the required time.
No takes yet. Share an insight, caveat, or question.
Cheng‐Chien Liu (2015) studied this question.