AI-Based Slope Hazard Analysis of 2011 Niigata-Fukushima Heavy Rainfall Disaster in Japan

This article attempts to develop a machine learning based slope hazard assessment method using the case study of the 2011 Niigata-Fukushima heavy rainfall disaster in Japan. The proposed method is constituted of two analysis procedures. The first step is a pointwise hazard assessment by machine learning (Step 1), and the second step is an image analysis (Step 2) to extract collapsed blocks based on the analysis results of pointwise hazard assessment. In Step 1, conventional methods have lacked to capture the accurate topographic information necessary to evaluate the stability of collapsed blocks. Since the topographic information needs to be scaled according to the size of collapsed block, it needs to consider from the microtopography to the macroscopic topography to properly catch the various collapsed blocks. In this article, the moving averages of various topographic indexes are introduced into AI-based slope hazard evaluation and its validity is examined. Furthermore, analyses using overground and underground openness indices were conducted and it was found that slope hazard assessment with very high accuracy is possible. Output of pointwise slope hazard assessment is obtained like a mosaic picture. It is used to evaluate the hazardous area, but difficult to discriminate hazardous collapsible blocks. Therefore, in Step 2, this article attempted to extract dangerous blocks by machine learning of the image outputting the hazard assessment. Case studies showed that collapsible blocks are successfully estimated.