Deep Learning Peak Ground Acceleration Prediction Using Single-Station Waveforms

Predicting the peak ground acceleration (PGA) from the first few seconds after the P-wave arrival time is crucial in estimating the ground motion intensity of the earthquake. The early estimation of PGA supports the earthquake-early warning (EEW) system to generate the warning. Here, we propose to use the vision transformer (ViT) to predict the PGA using 4-s three-channel single-station seismograms, i.e., 1 s prior to the P-wave arrival and 3 s subsequent to the arrival. The ViT can significantly extract remarkable information from the data resulting in superior prediction performance. The core layer of the ViT is the multihead attention (MHA) network which highlights the significant features of the input data. We train and evaluate the proposed algorithm using the Italian earthquake waveform data, where the proposed algorithm shows a promising result. The proposed ViT network utilizes an augmentation strategy to improve the learning ability of the model. Our proposed method is compared to the benchmark deep learning (DL) methods and empirical ground-motion models (GMMs) and outperforms all of them. The proposed algorithm can even predict the PGA accurately using only 2-s data after the P-wave arrival time. The proposed ViT architecture can also be integrated into a PGA classification framework. Finally, the proposed algorithm is tested using real-time data and shows accurate results, indicating its applicability in real-time monitoring.