CRED: A Deep Residual Network of Convolutional and Recurrent Units for Earthquake Signal Detection (1810.01965v1)

Abstract: Earthquake signal detection is at the core of observational seismology. A good detection algorithm should be sensitive to small and weak events with a variety of waveform shapes, robust to background noise and non-earthquake signals, and efficient for processing large data volumes. Here, we introduce the Cnn-Rnn Earthquake Detector (CRED), a detector based on deep neural networks. The network uses a combination of convolutional layers and bi-directional long-short-term memory units in a residual structure. It learns the time-frequency characteristics of the dominant phases in an earthquake signal from three component data recorded on a single station. We train the network using 500,000 seismograms (250k associated with tectonic earthquakes and 250k identified as noise) recorded in Northern California and tested it with an F-score of 99.95. The robustness of the trained model with respect to the noise level and non-earthquake signals is shown by applying it to a set of semi-synthetic signals. The model is applied to one month of continuous data recorded at Central Arkansas to demonstrate its efficiency, generalization, and sensitivity. Our model is able to detect more than 700 microearthquakes as small as -1.3 ML induced during hydraulic fracturing far away than the training region. The performance of the model is compared with STA/LTA, template matching, and FAST algorithms. Our results indicate an efficient and reliable performance of CRED. This framework holds great promise in lowering the detection threshold while minimizing false positive detection rates.

• The paper presents CRED, a hybrid deep residual network combining CNNs and bi-directional LSTMs to enhance earthquake signal detection accuracy.
• It uses a dataset of 500,000 seismograms from Northern California and achieves a 99.95% F-score, outperforming traditional methods like STA/LTA and template matching.
• CRED's robust performance across varying noise levels and geographical regions demonstrates its potential for real-time seismic monitoring and hazard mitigation.

Advanced Earthquake Signal Detection with Deep Residual Networks

The seismic community faces the ongoing challenge of efficiently processing the vast amounts of data generated annually. The paper "CRED: A Deep Residual Network of Convolutional and Recurrent Units for Earthquake Signal Detection" introduces an innovative approach in this regard, offering an advanced method for detecting earthquake signals using a deep neural network. The proposed method, the Cnn-Rnn Earthquake Detector (CRED), aims to address the fundamental issues of sensitivity, robustness, and efficiency in earthquake signal detection by utilizing a combination of convolutional and recurrent units within a residual learning framework.

Methodology

CRED employs a hybrid architecture that integrates convolutional neural networks (CNNs) and recurrent neural networks (RNNs), specifically bi-directional long short-term memory (LSTM) units. The network is designed to learn the time-frequency characteristics of earthquake signals from three-component seismograms. The combination of CNNs and LSTM units allows CRED to efficiently handle both spatial and sequential data features. The CNN layers are utilized for initial feature extraction, significantly reducing the need for extensive feature engineering and preprocessing. The RNN component, specifically the bi-directional LSTMs, facilitates learning temporal patterns and relationships within waveform sequences, while the residual learning framework allows for the construction of deeper networks by simplifying optimization through residual mappings.

Results

The authors trained CRED using a dataset of 500,000 seismograms from Northern California, comprising an equal number of tectonic earthquake signals and noise samples. The network demonstrated impressive accuracy with an F-score of 99.95% during testing. Furthermore, the model was evaluated on semi-synthetic data to assess its robustness across varying noise levels, and it confirmed CRED's superior performance compared to STA/LTA and template matching algorithms. The paper also employed CRED in a different geographical region, Central Arkansas, which highlighted its generalization capabilities as it successfully detected over 700 microearthquakes, including some below magnitude -1.3 ML.

Implications and Future Research

The findings suggest that CRED is a highly promising tool for seismologists, offering significant improvements in sensitivity and efficiency over existing methods. The ability to effectively detect low-magnitude events with minimal false positives has important implications for real-time seismic monitoring and hazard mitigation, particularly in regions susceptible to induced seismicity. The residual learning framework paves the way for future work to incorporate even deeper architectures without the risk of degradation in detection performance.

The generalization ability demonstrated by CRED when applied to data from a different region underscores the potential for broad applicability in diverse seismic settings. Moreover, future efforts could focus on using a more extensive and varied dataset to enhance the generalization and accuracy further. The research could also explore recursive network approaches for dynamic model updating, which would be particularly beneficial in operational settings for real-time earthquake monitoring.

In conclusion, the development of CRED marks a critical step forward in the application of deep learning models to seismology, translating into practical advancements for earthquake detection. The flexibility of this architecture indicates potential scalability and adaptability, offering a robust tool for seismic data analysis in both research and practical applications.