Machine Learning Delivers High-Resolution Earthquake Risk Maps for Tokyo’s Vulnerable Soils

Tokyo, one of the world’s most densely populated megacities, sits atop a seismically active zone where the risk of devastating earthquakes is constant.

Among the most destructive effects is soil liquefaction—a process where intense shaking causes water-saturated soil to lose its strength and behave like liquid. Traditional hazard maps, often based on limited data and low-resolution grids, struggle to capture Tokyo’s complex subsurface conditions. A new study led by Professor Shinya Inazumi of Shibaura Institute of Technology introduces a machine learning–driven framework that dramatically improves the precision of liquefaction hazard maps, offering critical insights for earthquake preparedness and urban resilience.

phys.org

1. Seismic vulnerability: Tokyo faces persistent earthquake threats, with soil liquefaction posing one of the most severe risks to infrastructure.
2. Historical context: Liquefaction damage was evident in the 1995 Great Hanshin-Awaji Earthquake, the 2011 Great East Japan Earthquake, and the 2024 Noto Peninsula Earthquake.
3. Limitations of traditional maps: Existing hazard maps rely on sparse borehole data and simple geostatistics, producing coarse grid results (500 m or larger).
4. Urban complexity: Tokyo’s reclaimed lands, river floodplains, and soft soils require finer-scale, more accurate hazard assessments.
5. Machine learning solution: Researchers used artificial neural networks (ANNs) to integrate geotechnical data into high-resolution 3D hazard models.
6. Extensive dataset: The framework analyzed 13,926 borehole records, one of Japan’s largest AI-based geotechnical datasets.
7. Model accuracy: The ANN outperformed traditional methods, predicting soil types and N-values (measures of soil density and strength) with high precision.
8. Detailed risk mapping: The new maps identified high-liquefaction-risk zones at a 200 m grid scale, with clarity surpassing official maps.
9. Critical findings: Localized vulnerable areas, such as Koto Ward, were successfully highlighted, where conventional maps often fail.
10. Global applicability: The method offers a scalable model for megacities worldwide, improving urban planning, foundation design, disaster prevention, and public awareness.