Data-Driven Prediction of Peak Ground Acceleration from Seismic Waveforms

Data-Driven Prediction of Peak Ground Acceleration from Seismic Waveforms
The identification and rapid estimation of earthquake parameters, such as Peak Ground
Acceleration (PGA), are critical components of earthquake monitoring and Earthquake Early
Warning (EEW) systems. As seismic waves propagate through the geological media, their
interaction with subsurface layers possessing varying elastic and damping properties leads to
significant variability in observed ground motion. These local site effects strongly influence
PGA values, for instance if the site is composed of soft-sediments the amplification within the
ground motion is more prominent than that of a rocky terrain or very firm sediments.
In this study, we investigate the application of deep learning techniques to model the nonlinear
relationships between incoming seismic signals and the resulting PGA. The proposed model
architecture may be considered a prototype that can be integrated into operational EEW
systems, enhancing the timeliness and accuracy of ground motion predictions and thereby
supporting more effective emergency response and risk mitigation strategies.