[S-CG61] Dynamics in mobile belts

[SCG61-06] Nonlinear Coupling of 3D Velocity Heterogeneity and Seismic Potential in the Japan Arc decoded by Machine Learning

*Zhang Chunjie¹, Hikaru Iwamori¹, Aitaro Kato¹, Thomas Yeo², Usui Yoshiya¹ (1.Earthquake Research Institute , The University of Tokyo, 2.Graduate School of Science and Technology, University of Tsukuba)

Keywords:Seismic Veolocity Gradient, Machine Learning, Earthquake occurence, Japan Arc

Seismic velocity structure in the crust and uppermost mantle provides critical constraints for understanding earthquake generation mechanisms. However, quantitative characterization of the nonlinear relationships between these velocity architectures and seismicity remains challenging, owing to the geological complexity and subjectivity inherent in conventional interpretations. Here, we integrate three-dimensional P-wave (Vp) and S-wave (Vs) velocity models into ensemble machine learning frameworks to systematically investigate how velocity magnitudes, gradients, and localized heterogeneities govern seismogenesis across the heterogeneous Japan Arc. SHAP interpretability analyses identify Vp values of approximately 5.5–6.5 km/s and Vs values of 3.4–3.8 km/s near the cut-off depth as key discriminators of earthquake occurrence. In particular, contrasting vertical gradients (positive in Vp yet negative in Vs) display a strong correlation with regional seismicity, whereas positive east–west Vs gradient transitions show a distinct association with elevated seismicity. Complementary physics-based modeling of geofluid distribution and thermal structure supports these data-driven insights, illuminating the nonlinear interplay among fluid accumulation, thermal regime, and velocity structure. By integrating data-driven and physics-based methodologies, the multi-paradigm approach offers a quantitative framework for assessing seismogenic potential via crust and uppermost mantle velocity diagnostics, shedding light on the earthquake triggering mechanisms in the Japan Arc.