Advancing seismic monitoring with operator learning

Seismic monitoring is essential for understanding earthquake processes and mitigating associated hazards. Traditional deep learning methods in seismology often rely on single-station inputs or fixed seismic acquisition geometries, limiting their robustness and scalability in dynamic seismic monitoring systems. Recent developments in operator learning have enabled new approaches to seismic monitoring, which address these limitations by leveraging the full spatiotemporal information contained in multi-station waveform data and generalizing across diverse seismic network configurations. This work reviews three recent contributions that illustrate these approaches: (1) the Phase Neural Operator (PhaseNO), an algorithm for multi-station picking of seismic arrivals; (2) the Location Neural Operator (LocNO), a full-waveform earthquake location algorithm; and (3) an accelerated correlation-based time-reversal imaging method that uses the Fourier Neural Operator for efficient wavefield backpropagation. Together, these methods demonstrate that operator learning can enhance the accuracy, adaptability, and computational efficiency of seismic monitoring across varying geologic settings and network configurations.