[H-GM04] 地形

[HGM04-06] Landslide Susceptibility in Hiroshima, Japan: Impact of Slope Angle and Lithology Using GIS and 2014 Rainfall Event Data

*Kanchana Kumari Mallika Achchillage¹、Tsuyoshi Wakatsuki²、Chiaki Oguchi³ (1.PhD student, Graduate School of Science and Engineering, Saitama University, 255 Shimo-Okubo, Sakura-ku, Saitama, 338-8570, Japan、2.Chief Researcher, Storm, Flood, and Landslide Research Division, National Research Institute for Earth Science and Disaster Resilience, 3-1 Tennodai, Tsukuba, Ibaraki, 305-0006, Japan、3.Associate Professor, Department of Civil and Environmental Engineering, Saitama University, 255 Shimo-Okubo, Sakura-ku, Saitama, 338-8570, Japan)

キーワード：Landslide susceptibility, Logistic regression, Frequency ratio, GIS-based analysis, Rainfall-induced landslides

Landslides induced by intense rainfall during monsoons, typhoons, and extreme weather events are significant natural hazards in Japan. Understanding the key factors influencing their occurrence is crucial for effective hazard assessment and risk reduction. Hence, this study examines the primary factors contributing to landslide occurrence and develops a predictive model for landslide susceptibility in Hiroshima City, specially following the 2014 rainfall event. A total of six hundred landslides were analyzed, with four key factors identified based on prior research: slope angle, curvature, height, and lithology. Furthermore, logistic regression (LR) and frequency ratio (FR) methods were applied to quantify the relationship between these factors and landslide occurrence. To ensure an unbiased model, 1,800 non-landslide points were systematically selected through GIS-based random sampling, stratified selection, and spatial constraints. Moreover, a landslide susceptibility map was generated using a GIS-integrated statistical approach, categorizing the study area into various risk levels. Then, the model’s performance was assessed using the receiver operating characteristic (ROC) curve, which yielding an area under the curve (AUC) of 0.780 and an accuracy of 0.761. The results of the logistic regression analysis indicate that slope angle is the most influential factor in landslide initiation, with a strong positive coefficient of 5.876. This highlights its critical role in the occurrence of landslides. Lithological factors, mainly granite (1.945) and pyroclastic rock (1.483), also contribute significantly to landslide susceptibility. In contrast, profile curvature (-0.221) and height (-0.009) showed minimal impact. Also, the frequency ratio analysis corroborated the strong correlation between steeper slopes and landslides occurrence, with granite identified as the lithology most prone to landslide. Conversely, no landslide events were recorded in alluvium, chert, limestone, and granodiorite, indicating their lower susceptibility. These findings enhance assessing and mitigating landslide hazard and offer practical implications for disaster risk management in Hiroshima. Although the model demonstrates moderate predictive accuracy, future research is essential to explore factor interactions, incorporate additional variables, and conduct field validation to enhance reliability. Notably, it is imperative to carry out independent validation of this predictive model, especially considering the rainfall event that occurred in 2018. However, the resulting susceptibility map is valuable resource for future disaster risk management and mitigation efforts in Hiroshima. Similarly, this methodology may also apply to other regions in Japan, extending its relevance.