Abstract
How and to what extent earthquakes repeat remain key open questions with important implications for early warning and hazard forecasting. Using a 19-year waveform dataset and a starting catalog of repeating earthquakes (repeaters) from small segments (<1%) of the Tohoku-Hokkaido megathrust, we quantify rupture repeatability for ∼M3-5.5 repeaters. We observe strong spatial clustering of hypocenters and centroids, widespread and persistent hierarchical fault structures, systematic downdip hypocenter-centroid directivity that may amplify coastal shaking, and unexpectedly limited rupture repeatability among events classified as repeaters. To interpret these patterns, we propose a hierarchical fault model and define “hierarchical repeaters” that link two end-members with similar initiation (hypocenter repeaters) and similar termination (centroid repeaters). Our results highlight geometric controls on rupture recurrence and suggest intrinsic limits on predictability within the spatially restricted, preselected sample analyzed here. If earthquakes are fundamentally self-similar, hierarchical geometric controls may operate across scales, from small repeaters to giant megathrust earthquakes.