Neural Network-Based Risk Assessment of Artificial Fill Slope in Residential Urban Region

Abstract

Destructive urban earthquakes have triggered landslides on gentle residential slopes in Japan. Earthquake-induced slope instability is closely related to artificial landforms, especially valley fill (embankments). The study of artificial landform changes has shown that differences in fill shape, such as depth, width, base angle inclination, and cross-sectional form, may be key discriminating factors in slope instability. The earthquake trigger mechanism must be considered in accurate estimation analysis, but it is difficult to include earthquake parameters in convenient linear multivariate analysis. Neural network analysis is applied to assess large fill slope instability in residential urban areas. The neural network model we developed including causative - fill shape, groundwater, and construction age - and triggering factors - distance from the fault, moment magnitude, and direction to the fault - was checked independently against another dataset and sensitivity was analyzed. Our proposed neural network model should enable us to establish more reliable landslide hazard mapping in residential urban areas, which, in turn, should aid disaster resilient societies in seismically active regions.