The purpose of this study was to distinguish landslide areas by using current satellite remote sensing data and GIS, clarify the degree to which landslides can be distinguished and the problems confronted when attempting to use these information sources, and clarify issues for distinguishing landslides in the future through the use of high-resolution satellite data.

1) Satellite image processing and GIS software used in the analysis: TNTmips from MicroImages, Inc.

2) Data used for the analysis:

Grand truth data
Aerial photographs (October 1985, October 1989, May 1994)
Numeric map (1/25,000)
50m mesh elevation data
Satellite data
LANDSAT/TM data (October 1984, September 1995)
SPOT/HRV-XS data (October 1988, October 1995, April 1998)

3) Summary of analysis methodology:

After creating an orthophotograph by using aerial photographs and 50m mesh elevation data, we used this as grand truth data and decided whether there were landslide area threshold values by using a normalized differential vegetation index (NDVI) calculated from satellite data, in distinguishing the landslide areas according to the values.

1) We determined there are no problems for practical use of the data because the accuracy of the distinguished landslide area was approximately 70% for both LANDSAT/TM and SPOT/HRV-XS; the total area of the landslide areas was also nearly the same as the total area on landslide distribution maps interpreted from aerial photographs.

2) Because the NDVI value for distinguishing the landslide value will change depending upon differences in the satellite observation days or satellite sensors, when determining these values it is always necessary to have an orthophotograph from an aerial photograph taken on a date close to the observation date to use as grand truth data.

3) It is possible to improve landslide area discrimination accuracy by estimating the location of shadows from a 50m mesh and taking the shadows into consideration when distinguishing landslide areas, and by using GIS to calculate a channel network from elevation data and excluding this area.

4) Of course, discrimination accuracy will also not be very good if high ground resolution satellite data corresponding to the size of the landslide area is not used.

In mountainous regions particularly, it is more important to use data from time periods when the influence of shadows is at a minimum. Refer to the source literature for ways to determine the discrimination accuracy.

When considering practical application, aerial photographs or orthophotographs must be used to increase the accuracy of landslide discrimination from the use of 20-30m ground resolution data available from LANDSAT satellite images or SPOT satellite images. This means more time will be required for analysis following a disaster.

By using recent high-resolution satellite data, however, it is not necessary to obtain aerial photographs or process orthophotographs. Furthermore, because the ground resolution capacity of high-resolution satellites is high enough compared to the size of most landslides, we believe discrimination using NDVI values makes it possible to attain extremely practical landslide area discrimination.

In order to improve the accuracy of landslide area discrimination or landslide area estimates, higher precision numeric map data for use with GIS and elevation mesh data within several meters detail, must be created in advance of use with GIS.