2014 年広島豪雨による斜面崩壊を制約した地形条件 ―機械学習を利用した評価の試み

抄録

This study examined geomorphological location conditions suitable for the slope failures triggered by heavy rainfall in 19 to 20 August 2014 in Hiroshima using random forest (RF) method, a kind of machine learning. We focused on three geomorphic units with different spatial scales: 1) neighborhood of failure heads, 2) first-order basins, and 3) large basins whose low end reach the base of the mountains. The failure heads mainly located at around the steepest slope near the ridge top. The places where slope type changes from convergent (concave) to straight tend to turn into failure head, possibly reflecting the effect of so-called zero-order shallow valley. The first-order basins elevated near the main divide with steep slopes and with rough textures tend to have slope failures, probably reflecting the susceptibility to the rainfall event. In large basins, the number of surface failures normally increases with the increase of the basins area and the basin roughness. Basin area is one of the most important factors to control disaster risk along the foot of the steep mountains where debris flows originated from failures can reach the entrance of the valley.