A Deep-Learning-Based Framework for Focal Mechanism Determination and Its Application to the 2022 Luding Earthquake Sequence

TL;DR

This study tackles automated focal mechanism determination for small-to-moderate earthquakes by coupling a deep neural network polarity detector with HASH grid-search inversion, trained on extensive CSN data. The model achieves high polarity-detection performance and tolerates sizeable arrival-time errors, enabling near-real-time application. When applied to the 2022 Luding sequence, the workflow produced a large set of focal mechanisms consistent with known fault structures and reveals fault-segmented clusters, improving constraints on regional stress and fault geometry. Open-source data and code are provided to promote reproducibility and broad adoption in dense seismic networks.

Abstract

P-wave first-motion polarity plays an important role in resolving focal mechanisms of small to moderate earthquakes (M <= 4.5). High-quality focal mechanism solutions for abundant small events can greatly improve our understanding of regional tectonics, fault geometries, and stress-field characteristics. In this study, we develop an automated focal mechanism determination framework that integrates deep neural networks with P-wave first-motion polarity observations, and apply it to the 2022 Luding earthquake sequence. The model is trained on 12 years (2009-2020) of manually annotated 100 Hz waveform records from the China National Seismic Network, achieving a polarity recall of 97.4 percent and a precision of 98.5 percent. After automatically determining the first-motion polarities, we invert focal mechanisms using the HASH method. The resulting focal mechanism solutions show high consistency with mapped fault structures in the Luding region, demonstrating the reliability and applicability of the proposed workflow.

Figures (7)

• Figure 1: Data statistics for 2009--2020. (a) Distribution of stations (red triangles) and earthquakes (black dots); (b) earthquake magnitude distribution; (c) epicentral distance distribution of stations with Pg-phase picks; (d) epicentral distance distribution of stations with polarity-labeled Pg phases.
• Figure 2: P-wave velocity models used in the grid-search process. The five P-wave velocity structures incorporated into the HASH inversion to enhance robustness of focal mechanism solutions.
• Figure 3: ROC curve for first-motion polarity detection. Receiver operating characteristic (ROC) curve of the polarity-classification model, illustrating performance across different probability thresholds.
• Figure 4: Comparison of focal mechanism solutions for the Luding earthquake sequence with previous work Yang2022LudingProcess.
• Figure 5: Events with large discrepancies from manually determined focal mechanisms. (a) Earthquake that occurred on 5 September 2022 at 19:26:20; (b) earthquake that occurred on 7 September 2022 at 08:34:35. These two events show the largest differences when compared with manual results, primarily due to low-quality first-motion waveforms.