Data-Driven Identification of Spatially Coherent Seismic Sources Using Clustering and Probabilistic Modeling

This study presents a data-driven framework for seismic source modeling, suitable for probabilistic seismic hazard assessment in Sabah, Malaysia. Seismicity patterns were classified into five spatially coherent source zones using the K-Means clustering algorithm, providing a structured representation of earthquake distribution. The magnitude completeness of each cluster was determined using the maximum curvature, goodness-of-fit, and entire magnitude range methods. Subsequently, the Gutenberg–Richter relation, maximum likelihood estimation, Aki–Utsu bootstrap, and Monte Carlo simulations were employed to estimate the a- and b-values. The maximum-expected magnitude was computed using a hybrid approach combining Kijko, Weichert, and extreme-value theory formulations, yielding values ranging from 6.28 to 7.17. The results indicate that b-values range from 0.44 to 0.75, consistent with typical crustal seismicity. Depth and spatial analyses reveal that earthquakes are predominantly shallow, with an average depth of approximately 20 km. The derived cluster properties, including centroids, covariance, and spatial geometry, were subsequently integrated into a generalized probabilistic framework. This framework describes the spatial, magnitude (truncated Gutenberg–Richter), and depth distributions of seismic events. The formulation enables the evaluation of the probability density of future seismic events for each cluster. The proposed integrated methodology demonstrates the effectiveness of combining clustering, statistical and probabilistic approaches to develop reliable, reproducible, and updatable seismic source models. The findings provide a robust foundation for refining regional seismic hazard maps and enhancing earthquake-resistant design practices in Sabah and other regions with moderate seismicity.