Feature Extraction from Seismic Waveforms for Earthquake-Explosion Discrimination

Jeongung Woo, Alamgir Hosain, Nathan Maier, & Ting Chen

Submitted August 30, 2026, SCEC Contribution #15511, 2026 SCEC Annual Meeting Poster #140

Accurate classification of seismic source types is an important task for seismic monitoring. While classical methods such as the P-to-S ratio provide a simple and general standard for discrimination, recent deep-learning methods can improve classification performance but may be less stable in general-purpose applications because of variations in source characteristics, crustal heterogeneity, and training data availability. In this study, we propose a fingerprint extraction method to examine whether complex seismic data can be compressed into compact features while retaining source-relevant information for discrimination. Labeled earthquake and explosion waveforms from multiple regions across the United States are converted to time-normalized spectrograms based on their theoretical P- and S-wave arrival times. Each spectrogram is then standardized, transformed using a two-dimensional Haar wavelet transform, and binarized to retain only the dominant time–frequency features. UMAP, a dimensionality-reduction technique, maps the seismic fingerprints into a low-dimensional space reflecting similarities among individual samples, indicating that source-related characteristics remain identifiable after substantial simplification. The spectrogram compression proposed in this study provides a compact and visually accessible representation of seismic source differences and may support the review of data provisionally classified using existing tools. Furthermore, the feature extraction preserves physically meaningful information, including relative timing, frequency content, and component-dependent waveform structure, and provides sparse feature representations for computationally efficient classifier development.

Citation
Woo, J., Hosain, A., Maier, N., & Chen, T. (2026, 08). Feature Extraction from Seismic Waveforms for Earthquake-Explosion Discrimination. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology