Abstract
Identifying the arrival times of seismic P-phases plays a significant role in real-time seismic monitoring, which provides critical guidance for emergency response activities. While considerable research has been conducted on this topic, efficiently capturing the arrival times of seismic P-phases hidden within intensively distributed and noisy seismic waves, such as those generated by the aftershocks of destructive earthquakes, remains a real challenge since most common existing methods in seismology rely on laborious expert supervision. To this end, in this paper, we present a machine learning-enhanced framework based on ensemble learning strategy, EL-Picker, for the automatic identification of seismic P-phase arrivals on continuous and massive waveforms. More specifically, EL-Picker consists of three modules, namely, Trigger, Classifier, and Refiner, and an ensemble learning strategy is exploited to integrate several machine learning classifiers. An evaluation of the aftershocks following the MS 8.0 Wenchuan earthquake demonstrates that EL-Picker can not only achieve the best identification performance but also identify 120% more seismic P-phase arrivals as complementary data. These findings clearly validate the effectiveness, efficiency, flexibility and stability of EL-Picker.
Introduction
In seismic monitoring, primary waves (P-waves) are the first signals to arrive at a station, providing a crucial window for early warning before more destructive waves occur. Accurately "picking" these arrival times is essential for determining earthquake location and magnitude. However, during intense aftershock sequences, the volume of data and the presence of high-frequency noise make manual picking unsustainable and traditional automated methods unreliable. Existing algorithms like STA/LTA often struggle with a trade-off between sensitivity and false alarm rates.
The EL-Picker Framework
To address these challenges, we developed EL-Picker, a three-stage framework that combines traditional signal processing with advanced machine learning.
The framework operates through three interactive modules:
- Trigger: Utilizes a broadband phase-picking algorithm (FilterPicker) with a low threshold to ensure high recall, capturing nearly all potential seismic events.
- Classifier: For each triggered window, the system extracts over 600 features across various domains (amplitude, spectral, etc.). These features are processed by an ensemble of nine machine learning models (e.g., SVM, Random Forest, KNN). A meta-model then aggregates these outputs to distinguish real P-waves from noise.
- Refiner: Confirmed P-waves undergo arrival time fine-tuning using the Akaike Information Criterion (AIC) and multi-station cross-validation to eliminate outliers.
Experimental Evaluation
EL-Picker was evaluated using datasets from the MS 8.0 Wenchuan earthquake (12,000+ labeled arrivals) and the MW 6.0 South Napa earthquake. The training process involved a robust set of negative samples, including S-waves and mechanical noise, to ensure the classifier could effectively handle real-world interference.
Results and Discussion
The framework demonstrated superior performance compared to deep learning baselines. On the Wenchuan dataset, EL-Picker achieved high precision and recall, maintaining stability across different window lengths.
Key findings include:
- Sensitivity: EL-Picker identified 120% more seismic P-phase arrivals than were present in the expert-labeled catalog. Expert re-examination confirmed that over 90% of these additional picks were valid low-magnitude aftershocks.
- Geological Personalization: Analysis showed that different machine learning models within the ensemble performed optimally at different stations, likely due to varying geological conditions (e.g., mountainous vs. plains).
- Efficiency: The Trigger module successfully filters out 99.92% of non-event data, allowing the system to run in real-time on standard hardware.
Conclusion
EL-Picker bridges the gap between sensitive automated triggers and accurate human inspection. By leveraging ensemble learning, it provides a robust solution for real-time seismic monitoring, especially during high-activity aftershock sequences. Future work will focus on extending this ensemble approach to S-phase picking and full-scale deployment in national seismic networks.
Summary updated