Seismic event detection with Fourier Neural Operator

This paper discusses the application of the Fourier Neural Operator (FNO) framework for seismic event detection using its resolution-invariant properties and efficiency in processing regularly sampled data such as seismograms. Using the STanford EArthquake Dataset (STEAD), a comprehensive seismic waveform collection, the study demonstrates the ability of FNO to achieve high accuracy (95\% F1 score) in discriminating seismic events from noise, even with limited training data. The FNO-based approach overcomes the problems of traditional and deep learning methods, such as sensitivity to noise, waveform variability, and computational inefficiency, by leveraging its Fourier domain processing and sample invariance. The results highlight the potential of FNO for near-real-time microseismic monitoring, which will facilitate advances in geophysical exploration and risk management of induced seismicity in energy development projects.