Forecasting the Preparatory Phase of Induced Earthquakes by Recurrent Neural Network

Earthquakes prediction is considered the holy grail of seismology. After almost a century efforts without convincing results, recent raise machine learning (ML) methods in conjunction with deployment dense seismic networks has boosted new hope this field. Even if large earthquakes still occur unanticipated, laboratory, field, and theoretical studies support existence preparatory phase preceding earthquakes, where small stable ruptures progressively develop into an unstable confined zone around future hypocenter. The problem recognizing critical importance for mitigating risk both natural induced events. Here, we focus on seismicity at Geysers geothermal field California. We address M~4 identification by developing ML approach based features computed from catalogues, which are used to train recurrent neural network (RNN). show that RNN successfully reveal preparation earthquakes. These results confirm potential monitoring microseismicity should encourage research also predictability