PoViT-UQ: P-wave polarity and arrival time determination using vision transformer with uncertainty quantification

Abstract

Determining earthquake focal mechanisms is essential for understanding fault geometry and the stress field in the Earth's crust. When focal mechanisms are estimated using P-wave first-motion polarities, accurate polarity determination is critical. In recent years, deep-learning-based polarity-determination models have been developed. However, the estimation of focal mechanisms using P-wave polarities is often not robust. When automating this process using deep learning models, it is crucial to identify and utilize only those polarity predictions that the model infers with high accuracy and low uncertainty. In this study, we propose a novel deep learning model, PoViT-UQ, that combines a Vision Transformer (ViT) with Monte Carlo Dropout (MCD) to estimate high-precision initial P-wave polarity classification and arrival time detection with uncertainty quantification. Using seismic waveform data sampled at 100 and 250 Hz, the model classifies polarities into three classes (Up, Down and Noise) and simultaneously estimates P-wave arrival times. The results showed a classification accuracy exceeding 98 per cent and a standard deviation of 0.027 s in the arrival time estimation using the 250 Hz model. By integrating MCD, we evaluate prediction uncertainty and apply an interquartile range threshold of ≤ 0.15 to improve the accuracy of focal mechanism estimates. Validation using aftershock data from the 2016 Central Tottori Earthquake confirmed that our approach contributes to efficient and high-precision focal mechanism estimates. Our model advances automated initial P-wave polarity determination and enables reliable data selection based on uncertainty quantification.