Python code applying deep learning techniques to strong motion records for estimating Vs30, a parameter representing the average shear-wave velocity in the top 30 meters of soil. This study explores whether strong motion records contain useful information for Vs30 estimation and whether DL-based methods can effectively utilize them. The paper introduces a large-scale strong motion record collection, AFAD-1218, which contains over 36,000 strong motion records from Türkiye.
This study uses the AFAD-1218 dataset, a comprehensive collection of over 36,000 strong motion records from Turkey. These records were obtained from Turkey's national strong-motion network operated by the Disaster and Emergency Management Authority (AFAD). The dataset spans a wide range of seismic events and regions, providing a rich resource for deep learning applications.
Characteristics of the AFAD-1218 dataset: Number of Records: 36,418 Sampling Rate: All sampled at 100 Hz Duration: Varying event durations from 5 to 300 seconds SNR: Ranging from a few dBs to 100 dB. Signals with SNR values lower than 25 dB were eliminated during the training process. Geographic Coverage: Nation-wide strong ground motion stations across Turkey Features: Includes ground acceleration time series and metadata, such as event magnitude, epicenter location, and station coordinates.
- This experiment adjusts seismic signals around the P-wave arrival time.
- Uses P-wave arrival time as additional input information.
- EQT dataset is used for both training and testing.
Main script for evaluating the model.
Includes functions for data processing and evaluation setup.
Defines the CNN-LSTM model architecture.
Evaluates sample data.
This project is licensed under the MIT License.