2000 年 8 巻 1 号 p. 1-10
The aim of this study is to make a quantitative analysis with respect to landslide occurrence behavior by using GIS. First, simplified ortho correction technique for SPOT images without orbit data is newly developed in this study. Next, brightness correction is carried out to the SPOT images. Finally, tests for algorithms of the automatic detection of landslides were iterated till the best-fit is derived. Comparing with the result from aerial-photo interpretations in the training area, the hit ratio in number of landslides are about 73% in Tanzawa region and 52% in Boso region, respectively. However, error detection mostly occurs at small landslides, and the landslides with over 2000m2 are detected correctly.
The relation between the cumulative number of landslides and the landslide area follows power function, therefore the behavior of landslide occurrence might relate to the model of self-organized critical phenomenon. On the basis of this assumption, it may be inferred that landslide occurrence is complicated interaction of elements. Thus, the prediction of individual landslide occurrences and its magnitude might be impossible by seeking a few parameters. When those relations are examined in detail, we can see the bending point that can divide all data into two segments. Then, the value of threshold is statistically obtained by calculating Akaike Information Criterion (AIC): 7500m2 in Tanzawa and 5100m2 in Boso. Because the bends are observed in both regions, it can be inferred that there is a different mechanism for landslide occurrence between large landslides and small landslides.