Performance of orbital remote sensing in the detection of landslides in southwest Coastal British Columbia

The utility of satellite imagery for landslide detection in southwest coastal British Columbia was evaluated by interpretation of orbital imagery (SPOT, IKONOS, and QUICKBIRD) and comparison with large-and medium-scale aerial photography. Satellite imagery was enhanced by simple, repeatable digital image processing techniques (contrast stretches and transformations, merging of images of different spatial resolutions, composition of 3-band composite images, and band ratioing). More numerous processing options were available for IKONOS and QUICKBIRD imagery than for SPOT due to the multispectral nature of the former. Examples of satellite imagery are provided to demonstrate detection of obvious, small, and relict landslides, identification of detail on large landslides, and interpretation problems caused by topographic shadows. Approximations of the smallest landslides visible under ideal illumination and viewing conditions were 1700 m 2 for 5 m SPOT panchromatic imagery, 1000 m 2 for 2.5 m SPOT panchromatic imagery, and 175 m 2 for panchromatic-sharpened multispectral IKONOS and QUICKBIRD imagery. Insufficient perception of morphology at these sizes, however, prevents absolute identification of the features as landslides. In the absence of stereoviewing, greater minimum landslide sizes than those indicated above for detection are required for landslide identification. The exact landslide size varies with illumination conditions, viewing geometry, and contrast between the landslide and the surrounding terrain. Detail visible on IKONOS and QUICKBIRD imagery of large landslides included large debris blocks and gullies on slope surfaces. Relict landslides were identifiable on multispectral imagery mainly due to increased green and near infrared spectral reflectance of colonizing vegetation. Topographic shadowing was found to hinder landslide interpretation on both aerial photographs and satellite imagery. This problem was severe for SPOT panchromatic imagery, but was slightly reduced by image contrast transformation. Shadow problems were significantly reduced on IKONOS and QUICKBIRD imagery by band ratioing using the red and near infrared bands. Some considerations important in landslide detection using remotely sensed imagery are also discussed. Re-vegetation masks landslides and can prevent detection of landslides where significant time has passed between time of failure and time of imaging. One of the most important factors affecting the ability of a sensor to detect landslides is spatial resolution. All other conditions being equal, the minimum size of a feature detectable from remotely sensed imagery decreases as spatial resolution increases. As sensor resolution increases, however, the area covered by a single image decreases, reducing the synoptic coverage provided by the imagery. Based on interpretation of satellite imagery and …