Shared Nearest Neighbor Based Classification of Earthquake Catalogs in Spatio-temporal Domain

A magnitude dependent space-time correlation of earthquake events (foreshock and aftershocks occurred before and after the mainshocks respectively) require a classification approach to obtain an unbiased/uncorrelated estimation of seismicity (Background events; occurred due to regular movement of tectonic plates). In this paper, a shared nearest neighbor (SNN) based approach is introduced to classify events from an earthquake catalog in space, time and energy (magnitude) domain. The space-time separated mainshocks (events with high intensity in Richter scale) are considered the cluster centroids in this paper. Temporal zones are determined based on the cluster centroid with a single iteration distance algorithm. A space-time shared neighborhood criterion is incorporated to find the foreshock-aftershocks (clustered events) related to the respective cluster centroid for in each temporal zone. Earthquakes have higher magnitude are combined to space-time clusters and rest are considered as part of background seismicity. The proposed method is applied on California and Japan earthquake catalogs and obtained classification results are interpreted in terms of the epicenter plot, space-time plot, lambda and cumulative plots for true, clustered (forshock-aftershocks) and unclustered (backgrounds) events. The background seismicity has linear cumulative rate with respect to time and small deviation from the mean in lambda plot reveals the better performance of the model.