Rapid Strong Earthquake Magnitude Estimation Based on Near-Field High-Rate GNSS Data Using Deep Learning

Strong earthquakes cause severe casualties and economic losses. Accurate and rapid magnitude estimation can enable timely emergency response and effectively mitigate earthquake disasters. Current mainstream algorithms rely on broadband seismic or strong-motion data, but during strong earthquakes, near-field clipping and a limited P-wave time window lead to significant underestimation of magnitudes. High-Rate Global Navigation Satellite System (HR-GNSS) data, which directly records ground displacement, can overcome the near-field clipping issue and holds significant potential for magnitude estimation in moderate-to-large earthquakes. With the application of deep learning in seismology, combining HR-GNSS data with deep learning can enable fast and accurate magnitude estimation. By simulating numerous earthquake ruptures and their HR-GNSS displacement waveforms, we trained a deep learning model for near real-time magnitude estimation. Results show high accuracy for moderate earthquakes (Mw 6.3+) in single-fault scenarios. In multi-fault scenarios, without requiring prior source location knowledge, the model quickly estimates magnitudes for major earthquakes (Mw 7.0+) near predefined faults based on GNSS displacement waveforms. Testing on the 2019 Ridgecrest earthquake sequence indicates that within 30 s, the estimated magnitudes stabilize close to the actual magnitudes with an error of less than 0.15 magnitude units. Compared to the peak displacement (Pd) method based on broadband seismic waves and the HR-GNSS regression method, our approach offers superior timeliness and accuracy without needing prior epicenter location information. Uncertainty tests show that for large earthquakes (Mw 7.2+), the method achieves 99% accuracy within 40 s. Overall, the model performs best for Mw 7.2+ events, retains limited applicability for Mw 6.3–7.2 events, and is less stable below Mw 6.3.