[H-TT15] Geographic Information Systems and Cartography

[HTT15-P02] Assessment of Earthquake-Induced Landslide Risks in the Greater Tokyo Area, Japan, Using Transfer Learning

*Chenzuo YE¹, Yuting TANG², Takashi OGUCHI³ (1.Department of Natural Environmental Studies, Graduate School of Frontier Sciences, The University of Tokyo, 2.Department of Complexity Science and Engineering, Graduate School of Frontier Sciences, The University of Tokyo, 3.Center for Spatial Information Science, The University of Tokyo)

Keywords:Earthquake-Induced Landslide Risk Assessment, Transfer Learning, Seismic Landslide Susceptibility, GIS

This study proposes an innovative approach for predicting earthquake-induced landslides in the Greater Tokyo Area, utilizing advanced transfer learning techniques. Given the area’s vulnerability to seismic activities, the research aims to enhance landslide predictive models using rich datasets from significant earthquakes. The methodology centers on a pre-trained Multilayer Perceptron (MLP) model, initially developed with extensive data from the Nepal earthquake in April 2015. This model is notable for its high accuracy and robust performance. To validate the effectiveness of this approach, the study employs landslide data from the 2008 Wenchuan earthquake in China and compares it with the Nepal earthquake data. By creating a seismic landslide susceptibility dataset for both earthquakes, the research adapts the well-trained MLP model to the unique environmental and seismic conditions of the Greater Tokyo Area using transfer learning principles. The outcome is a predictive framework specifically tailored for the area, aiming to enhance the identification of areas at high risk of seismic landslides and support targeted disaster prevention and mitigation strategies.
This research demonstrates the potential of transfer learning in bridging data gaps and adapting models across regions with diverse seismic histories. It offers a scalable and efficient solution for assessing landslide risks in earthquake-prone areas, particularly relevant to risk management. The validation process illustrates that transfer learning can significantly streamline model training, maintaining a minimal decrease in accuracy compared to building models from scratch. This highlights the adaptability of transfer learning in geoscience and its promising application in global disaster risk management frameworks.