Microseismic Monitoring with the Quake Neural Operator

Overview This repository provides the official implementation of the Quake Neural Operator (QNO). The QNO is a deep learning framework for automated microseismic monitoring that performs end-to-end earthquake detection and location directly from continuous seismic waveform data without explicit phase picking or phase association. Unlike conventional seismic monitoring workflows that treat detection and location as separate sequential tasks, QNO jointly learns signal classification and source characterization within a unified multi-task learning framework. QNO enables the construction of earthquake catalogs directly from continuous waveform data and maintains robust performance under low signal-to-noise conditions where clear phase onsets may be unavailable. The framework consists of a classification task for earthquake detection and a regression task for earthquake location and origin time. For the detection task, QNO predicts the probabilities of earthquake signals and noise at each station. For the regression task, the model estimates the earthquake location and origin time directly from waveform data. Similar to the Phase Neural Operator (Sun et al., 2023), QNO leverages neural operator architectures to model complex spatiotemporal wavefield patterns. Specifically, Fourier Neural Operators (FNOs) extract temporal features from individual station waveforms, while Graph Neural Operators (GNOs) capture spatial relationships across stations through a message-passing framework. Repository Structure This repository is organized as follows: . ├── model.py # QNO model definition ├── utils.py # Dataset and utility functions ├── train.py # Training script ├── evaluation.py # Evaluation & visualization ├── data/ # Sample data ├── output_training/ # Example training outputs ├── output_evaluation/ # Example evaluation outputs ├── requirements.txt # Python dependencies ├── LICENSE └── README.md