Machine Learning Based Earthquakes-Explosion Discrimination for Sea of Galilee Seismic Events of July 2018

Discrimination between earthquakes and explosions is an essential component of nuclear test monitoring. However, the discrimination methods currently employed by the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO) are sometimes less effective for regional events. For instance, five seismic events whose epicenters lie near the Sea of Galilee were reported by the CTBTO in July of 2018. Those were relatively strong regional events, observed by primary stations hundreds of kilometers from the epicenter. Notably, three out of those five events were not screened out by the CTBTO as natural events. In this work, a diffusion maps-based discrimination method is configured and applied for the July 2018 Sea of Galilee seismic events. New features are introduced to the method, in order to enhance automation and computational efficiency and facilitate its use in operational settings. In the first of which, waveform segments are selected by relying on calculated arrivals rather than observed arrivals, alleviating the need for detection by a human analyst. In a further extension of the method, the low-dimensional diffusion maps representation from the training set is extended to a test set by means of geometric harmonics, relieving the need for the re-calculation of the diffusion maps coordinates for the entire data set as each new event comes in. Utilizing a network of three stations, we show that this machine learning method classifies as earthquakes all the July 2018 Sea of Galilee seismic events with durational magnitude Md > 2.3. In the context of the CTBT, the method can be used as part of an Expert Technical Analysis in order to aid the State Party concerned to identify the source of specific events.