Microseismic source localization using Fourier Neural Operators

Microseismic event localization involves determining the location of small earthquakes or tremors, which can provide valuable insights into subsurface geological structures, the presence of natural resources, and the potential for future earthquakes. While a number of machine learning approaches have been proposed over the years; however, they are plagued by their generalization ability and in their capability for real-time monitoring. We propose a Fourier Neural Operator (FNO) based method that demonstrates high accuracy with limited training data to address the need for real-time event detection. The FNO model is trained using first arrival traveltimes on observation locations used as the input to predict the traveltime field in the entire computational domain. The minimum of the predicted traveltime field yields the hypocenter location. We demonstrate that the use of FNOs overcomes limitations associated with the recently developed data-driven and physics-based machine learning methods for hypocenter localization, and provides a reliable approach for real-time microseismic monitoring.