An Adaptable Random Forest Model for the Declustering of Earthquake Catalogs

Earthquake catalogs are essential to analyze the evolution of active fault systems. The background seismicity rate, or rate earthquakes that not directly triggered by other earthquakes, relates stressing a crucial quantity for understanding seismic hazards. Determining is challenging because aftershock sequences may dominate rate. Classifying these events in earthquake catalogs—known as catalog declustering—is common practice and most declustering solutions rely on spatiotemporal distances between events, such nearest-neighbor-distance algorithm, widely used various contexts. This algorithm assumes nearest-neighbor distance (NND) follows bimodal distribution related aftershocks. Constraining two distributions distinguish aftershocks from accurately. Recent work often uses linear splitting based NND, ignoring potential overlap populations resulting biased identification sequences. We revisit this problem with machine-learning algorithms. After testing several popular algorithms, we show random forest trained synthetic generated an Epidemic-Type Aftershock Sequence model outperforms approaches k-means, Gaussian-mixture models, support vector machine classification. apply our different catalogs: relocated Southern California center GeoNet New Zealand. Our capably adapts tectonic contexts, highlighting differences productivity crustal intermediate-depth seismicity.