Deep Sequence Models for Predicting Average Shear Wave Velocity from Strong Motion Records

TL;DR

This work addresses estimating the time-averaged shear-wave velocity $V_{s30}$ at strong-motion stations in Türkiye by leveraging a CNN-LSTM hybrid that captures spatial and temporal features from three-component strong-motion records. It extends prior CNN-based approaches by modeling sequential information, evaluating how P/S-wave arrival information and signal segmentation affect performance, and applying transfer learning with a large AFAD-derived dataset. The best results come from a transfer-learning setup that incorporates auto-annotated P/S information, achieving improved accuracy across site classes and highlighting the value of leveraging prior feature representations for regional generalization. The study advances site characterization for seismic hazard assessment by demonstrating a data-driven pathway to better predict $V_{s30}$ where direct measurements are unavailable.

Abstract

This study explores the use of deep learning for predicting the time averaged shear wave velocity in the top 30 m of the subsurface ($V_{s30}$) at strong motion recording stations in Türkiye. $V_{s30}$ is a key parameter in site characterization and, as a result for seismic hazard assessment. However, it is often unavailable due to the lack of direct measurements and is therefore estimated using empirical correlations. Such correlations however are commonly inadequate in capturing complex, site-specific variability and this motivates the need for data-driven approaches. In this study, we employ a hybrid deep learning model combining convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to capture both spatial and temporal dependencies in strong motion records. Furthermore, we explore how using different parts of the signal influence our deep learning model. Our results suggest that the hybrid approach effectively learns complex, nonlinear relationships within seismic signals. We observed that an improved P-wave arrival time model increased the prediction accuracy of $V_{s30}$. We believe the study provides valuable insights into improving $V_{s30}$ predictions using a CNN-LSTM framework, demonstrating its potential for improving site characterization for seismic studies. Our codes are available via this repo: https://github.com/brsylmz23/CNNLSTM_DeepEQ

Figures (6)

• Figure 1: Local site classes at the AFAD strong motion stations
• Figure 2: Unrolled-in-time representation of the proposed CNN+LSTM sequence model.
• Figure 3: A 4-channel input structure highlighting the arrival times of P and S waves
• Figure 4: Visualizations of the training losses (left-most column), $V_{s30}$ prediction errors (middle column) and $V_{s30}$ percentage prediction error (right column) of experiments (from top to bottom): a) $\beta_{auto,PGA}$, b) $\alpha_{man,PGA}$, c) $\alpha_{auto,P,15sec}$, d) $\alpha_{auto,PGA}$, and e) $\gamma_{ps,auto}$. All experiments are repeated with varying hyperparameters. The bold line represents the average of the results, with the transparent shaded regions indicating the maximum and minimum ranges.
• Figure 5: Stations with $V_{s30}$ data measured according to the K-means clustering method (585) stations divided into clusters (regions) (colors represent different clusters).