本研究は豪雨時に崩壊リスクの高い箇所を識別する方法を提案する。豪雨時に崩壊した場所に基づいて作成した教師データを用いて, 最も適合したディープニューラルネットワーク (DNN) の構築方法を整理した。予測モデルは既往研究手法よりも12％精度向上を達成した。また従来予測困難であった地域に対して適切な評価を行うことができた。加えて品質の良い教師データを収集した効果でDNNの構造を簡易化することに成功した。この解析方法に基づいた渓流の危険度評価の方法を提案した。

  This study analyzes the influence of surface water and groundwater flows on a landslide in Takumi area, Tomioka City, Gunma Prefecture induced by Typhoon Hagibis on 12 October 2019. Firstly, a surface infiltration model and surface runoff simulation module were integrated into a water table prediction model developed previously. Then the hydrological integration process was validated on a tilted impermeable V-catchment system and trialed on a Plexiglas sandbox system. Finally, the extended model was used to describe the infiltration and exfiltration process, surface water flow as well as groundwater level distribution for a study area in Takumi Village during the Typhoon. According to simulation results, both surface water and groundwater flows have gathered on the landslide sites where the slopes are concave. Surface water and groundwater have interacted with each other to form flows over time and cause the groundwater level in the study area to rise. At the time of the landslide, the groundwater level had reached the slope surface in most of the cells at the collapsed sites. As a result, the safety factor for slopes decreased significantly and landslides occurred. Therefore, both surface water and groundwater flows were thought to be the landslide triggers.