Landform classification method using self-organizing map and its application to earthquake damage evaluation

This paper presents a supervised classification method using a self-organizing map (SOM) to classify typical landforms based on a land cover map and a digital elevation model (DEM). The proposed method classified the landform of Kobe City in Japan into hill, plateau, fan and reclaimed land. These classified landforms were adopted for an earthquake damage evaluation of the 1995 Hyogoken Nanbu earthquake in Kobe. First, an amplification value for each landform was assumed in order to calculate the peak ground velocity (V_max) of the ground motion in the study area. Next, using empirical prediction methods based upon the relationship between V_max and the number of wooden houses damaged by past earthquakes, we can calculate predicted number of collapsed wooden houses. As a result, we can obtain damage distribution map that corresponds with the actual damage recorded following the 1995 earthquake