Massive earthquake detection using matched filter and fingerprinting techniques

Abstract:

Seismology data analysis is becoming a challenge due to the exponential growth of continuous data being stored. In this study we present and compare two methods to massively detect earthquakes: the matched filter and fingerprinting. We have tested matched filter over several study zones of interest: in the Western part of Mexico (the Jalisco Block and nearby zones) to study general seismic activity using more than 2000 templates, in the Isthmus of Tehuantepec in Southern Mexico to track aftershocks from the September, 2017 Mw8.2 earthquake, and in the North Pole to study seismic activity mainly caused by ice cracking, or ice-quakes. We have demonstrated the accuracy of this technique especially detecting low amplitude signals hidden in the noise and coming out when we stack the resulting correlation coefficients over multiple stations. We are now testing fingerprinting, a technique much more efficient computationally, where we focus on extracting a fingerprint of the waveform for several templates in the frequency domain by compressing the resulting spectrogram. We then apply a hash function to create a hash table for our automatically selected templates. Finally we search on all the data streams of the different networks we applied matched filter to match similar fingerprints.