Real-time determination of earthquake focal mechanism via deep learning

An immediate report of an earthquake’s focal mechanism is crucial for characterizing fault geometry, assessing stress changes, and anticipating aftershock patterns. Here, a deep learning method, Focal Mechanism Network (FMNet), is proposed to determine focal mechanisms in real time from full waveforms. FMNet is trained on 787,320 synthetic samples and successfully estimates focal mechanisms for four Mw ≥ 5.4 events in the 2019 Ridgecrest sequence. Trained on theoretical data, the network learns global waveform characteristics, enabling application in regions with limited historical data. After data arrival and preprocessing, prediction takes under ~200 ms on a single CPU.

Wenhuan Kuang, Congcong Yuan, Jie Zhang