Mechanism of large-scale landslides in residential valley fills induced by earthquake

通过对以往人工地貌变化的调查和以往地震灾害案例的多变量分析表明，填土的深度、宽度、基底倾角、断面形状等形状差异可能是边坡失稳的关键判别因素。在分析中需要考虑触发机制（例如地震）以进行准确的估计，然而，在这种线性多变量分析中很难包括地震参数（量化理论II）。将神经网络分析应用于城市居民区大型填土边坡的失稳评价。开发的神经网络模型，包括两个成因因素（形状的填充，地下水条件，年龄的建设）和触发因素（距离故障，矩的大小，方向故障）进行独立检查对另一个数据集和敏感性分析进行。在2003年地震中的应用结果表明，该模型的命中率超过90%，表明该模型可广泛应用于城市居民区滑坡灾害的预测。有限元分析也证实了这种软填料的衰减效果。根据调查资料和观测结果，提出了该沟谷充填型滑坡的滑动机理为滚动-滑动模型。与传统的二维稳定性分析方法相比，基于滚动-滑动模型的准三维稳定性分析方法能够更好地解释过去地震灾害的历史。这表明，谷底破坏和边墙约束是人工填谷型滑坡的重要机制。

Investigation of past artificial landform changes and multi-variate analysis of case studies of past earthquake disasters show that differences in the shape of fills, such as depth, width, inclination angle of the base, and cross-sectional form, may be the key discriminating factors of slope instability. Triggering mechanisms (e.g. earthquakes) need to be considered in the analysis for accurate estimation, however, it is difficult to include earthquake parameters in such linear multi-variate analysis (quantification theory II). Neural network analysis is applied to assess large fill slope instability in urban residential areas. The developed neural network model including both causative factors (shape of fills, groundwater condition, age of construction) and the triggering factors (distance from the fault, moment magnitude, direction to fault) was independently checked against another data set and sensitivity analysis was conducted. The high hitting ratio over 90 % in the application to the earthquakes in 2003 revealed that the proposed model should be capable of wide application to conduct landslide hazard mapping in urban residential areas.Observation of strong earthquake motion and pore water pressure in valley fill shows that the response on the fill should be smaller than the acceleration on the original ground. This attenuation effect of the soft filling material was also verified by FEM analysis. Based on the inventories and observation results, the roller-slider model was proposed as the mechanism of the valley fill type landslide. The quasi-3D stability analysis based on the roller-slider model should be capable to explain the inventories of the past earthquake disaster compared with the poor result of the conventional 2D analysis. It means that both the failure at the base of the valley and constraint by the side walls should be significant mechanism of the artificial valley fill type landslide.