EQAnomalyNet: Unsupervised Seismic Phase Picking via Transformer Autoencoder (opens in new tab)

Seismic phase picking is a critical geophysical process for earthquake monitoring and early warning. The exponential growth of continuous waveform data from modern seismic networks renders traditional manual analysis increasingly time-consuming and labor-intensive. Here, we introduce EQAnomalyNet, an unsupervised Transformer-based autoencoder that learns to reconstruct continuous seismic waveforms, thereby detecting phase arrivals by identifying reconstruction anomalies without labeled data. ...