A new AI-driven approach is reshaping how scientists read seismic data, helping uncover signals that traditional methods can overlook. By analyzing long-term records from seismic arrays, the system improves the detection of subtle vibrations linked to earthquakes and other underground activity.
In the study, researchers used three decades of measurements from NORSAR and other seismic operators. They tested three strategies: training on each station separately, combining sensor data before training, and allowing the model to determine how to merge the inputs on its own.
The most accurate results came from the method that combined signals before training, which strengthened weak readings and improved detection quality. The most efficient option was the model-led approach, which balanced speed and performance well for real-time monitoring.
The findings suggest that AI can become a powerful partner in seismic observation, especially when researchers need to identify faint activity quickly and reliably. The study also notes that broader training data could improve performance across different regions, particularly for S waves.
As seismic networks grow smarter, this kind of AI support could help build faster, more adaptive monitoring systems for the future.
