Editors’ Highlights are summaries of recent papers by AGU’s journal editors.
Source: Journal of Geophysical Research: Solid Earth
Deep learning has been well-developed to handle tasks that are difficult to execute using algorithms, yet are easily understood through human experience, making it a powerful tool for automating tasks. In seismology, the most successful application of deep learning has been the automatic detection of seismic phases with an efficiency unprecedented by previous algorithms. This advantage is also beneficial for earthquake early warning systems.
In a new study, Lin et al. [2024] used tremendous amounts of synthetic waveforms, generated from 10,000 simulated earthquakes, to train a deep learning neural network (M-Large) to predict ground shaking intensity using only a portion of real-time waveforms. The trained neural network achieved a significant improvement in warning time (about 40 seconds for earthquakes with Modified Mercalli Intensity (MMI) greater than 4) using high-rate global navigation satellite system (HR-GNSS) data. This research also suggests that the earthquake scaling relationship is utilized by the neural network to predict rupture parameters.
Citation: Lin, J.-T., Melgar, D., Sahakian, V. J., Thomas, A. M., & Searcy, J. (2023). Real-time fault tracking and ground motion prediction for large earthquakes with HR-GNSS and deep learning. Journal of Geophysical Research: Solid Earth, 128, e2023JB027255. https://doi.org/10.1029/2023JB027255
—Han Yue, Associate Editor, JGR: Solid Earth
