The Bayesian Gaussian mixture model with nearest-neighbor distance (BGMM-NND) algorithm: A new earthquake clustering method and its application to the Sichuan–Yunnan Block

We propose a robust earthquake clustering method: the Bayesian Gaussian mixture model with nearest-neighbor distance (BGMM-NND) algorithm. Unlike the conventional nearest neighbor distance method, the BGMM-NND algorithm eliminates the need for hyperparameter tuning or reliance on fixed thresholds, offering enhanced flexibility for clustering across varied seismic scales. By integrating cumulative probability and BGMM with principal component analysis (PCA), the BGMM-NND algorithm effectively distinguishes between background and triggered earthquakes while maintaining the magnitude component and resolving the issue of excessively large spatial cluster domains. We apply the BGMM-NND algorithm to the Sichuan–Yunnan seismic catalog from 1971 to 2024, revealing notable variations in earthquake frequency, triggering characteristics, and recurrence patterns across different fault zones. Distinct clustering and triggering behaviors are identified along different segments of the Longmenshan Fault. Multiple seismic modes, namely, the short-distance mode, the medium-distance mode, the repeating-like mode, the uniform background mode, and the Wenchuan mode, are uncovered. The algorithm's flexibility and robust performance in earthquake clustering makes it a valuable tool for exploring seismicity characteristics, offering new insights into earthquake clustering and the spatiotemporal patterns of seismic activity.