Earthquake activity as captured using the network approach

Earthquakes are a major threat to nations worldwide. Earthquake detection and forecasting are important and timely scientific challenges, not only for their obvious social impacts, but also because they reflect the actual degree of understanding of the physical processes controlling seismic event occurrences. Here, we propose an alternative approach for evaluating and understanding the dynamics of seismic events. The approach is based on the phase between the waveform signals of many stations, enabling detecting the evolution of relatively small magnitudes, down to Mw 1.3. We constructed a time-evolving network in which the network nodes are the stations, while the links are the level of correspondence between the stations’ signals. The links’ weights are quantified using the following statistical methods: cross-correlation, synchronization, mutual information, and coherence. Each of these methods reflects a different aspect of the phase relations between the waveforms of different stations in a given time window. We then developed global measures to study the properties of the time-evolving network of seismic activity. The global measures include the leading eigenvalues of the network links, the number of links above a certain threshold, and k-means clustering. We show that the network and its corresponding global measures vary significantly during seismic events. The results are based on detailed waveform station data and detailed catalogs from Southern California; our analysis focused on 27 mainshocks, during which we examined one-day data prior to the occurrence of the mainshock, as well as one hour of data following it. Among all the measures we investigated, we found that the coherence measure using the k-means clustering procedure exhibits the best performance. This technique correctly identifies earthquake events with magnitudes larger than 2.5 and exhibits moderate performance for weaker earthquakes with magnitudes larger than 1.3.