The aim of this article is to automatically identify repeating seismic events such as an aftershock sequence by utilizing a machine learning technique named diffusion maps. In previous work, the diffusion maps approach was applied for earthquake‐explosion discrimination and for characterizing explosions by their origin quarries. Diffusion maps, which is a nonlinear dimensionality reduction technique, constructs a low‐dimensional geometric representation of the seismograms. The embedding coordinates capture the intrinsic structure of the seismic signals and analysis is done in this low‐dimensional space. As a preprocessing step, the seismograms are converted to images in the time frequency domain. The approach is demonstrated on an aftershock sequence of the February 2004 Dead Sea earthquake with magnitude ML 5.2. In the first stage, the short‐term average/long‐term average (STA/LTA) detector is applied and then the diffusion maps‐based identification is performed. In a second example, a cross‐correlation detector is applied in the first stage and the proposed algorithm serves as a validation tool for the waveform correlation detector. The obtained results were confirmed by an analyst and compared with other methods. The experimental results demonstrate the potential and strength of the diffusion‐maps‐based approach, as the identification process can be carried out with no need of master templates for detecting new aftershocks.

You do not currently have access to this article.