A stochastic sampling-filtering framework for probabilistic reconstruction of paleo-earthquake ruptures

Reconstructing paleo-earthquake ruptures from sparse geological observations is inherently non-unique because evidence of past surface ruptures is commonly preserved only at isolated sites. We develop a stochastic sampling-filtering framework that represents this ambiguity as an ensemble of rupture scenarios consistent with geological, chronological, and mapped fault constraints. Candidate ruptures are sampled in timing, center position, and length and retained only when they satisfy these constraints. Applied to the Yanzhi Shan Fault, the framework identifies eight rupture-age density modes and a persistent southeastward trend in the spatial density of admissible ruptures. Criteria-based enumeration yields broadly similar first-order temporal concentrations, while prior-sensitivity tests show that filtering reduces differences among ensembles generated from contrasting priors. These results indicate constraint-induced convergence rather than prior-distribution independence. The framework therefore extracts observation-supported statistical rupture patterns while preserving uncertainty associated with non-unique rupture timing and geometry.