Item

Abstract

LFE’s regularly occur as dense groups or bursts which propagate along the subduction zone, within a belt like structure at 20-40 km deep. Deep-learning algorithms can model complex systems using large temporal and spatial datasets. We test two hypotheses that can explain what mechanism controls the migration of bursts. First, we assume that small patches of ductile material are responsible for the cascade propagation of bursts via stress diffusion. Thus, the variations in propagation fronts are controlled by structural heterogeneities along the path. Thus, we can obtain a map of their geographical location. The second model implies fluid diffusion propagation, which is described as parabolic propagation. Also, migration patterns of fluid induced micro-earthquake swarms are related to how close from failure a region is. Thus, the propagation is controlled by the accumulated and released stress within each cycle, not by the location of the heterogeneities. If the structural heterogeneities are responsible for the variations in the migration of burst, the heterogeneities within the tremor belt are corresponding to a consistent increase/decrease in the rupture velocity or diffusivity. For stress/fluid diffusion-mediated transport, such heterogeneities do not control the propagation pattern, but rather the stress does. If the area is under high stress, the migration shows accelerating front, while the front is mostly diffusive if the stress state is low. Using a PINN algorithm, we test which model best describes the migration of LFE swarms to contribute to the understanding of the heterogeneities across Nankai subduction zone for producing better forecasting models.