Combining Deep Learning With Physics Based Features in Explosion‐Earthquake Discrimination

Kong, Qingkai; Wang, Ruijia; Walter, William R.; Pyle, M. L.; Koper, Keith D.; Schmandt, Brandon

doi:10.1029/2022gl098645

Cited by 34 publications

(22 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Koper (2020) stressed the importance of lowering the detection threshold of explosions using the International Monitoring System (IMS) global seismic network at local and regional distances. Kong et al (2022) used local augmented waveforms and P-S ratios from 90 explosions in Northwest United States to train a deep learning model as an explosion-earthquake discriminator with high rates of success. However, their model did not perform well when applied to other regions.…”

Section: Discussionmentioning

confidence: 99%

“…However, there are opportunities to test how to create improved performance of the algorithm such as including synthetic tests or augmented waveforms to improve the diversity and size of the training datasets. Kong et al (2022) increased their explosion training explosion data from 8,502 to 178,059 traces by using data augmentation. For our case, data augmenting to create a larger data set would allow us to train station-distance dependent models to test performance compared to the global model.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Global Nuclear Explosion Discrimination Using a Convolutional Neural Network

Barama,

Williams,

Newman

et al. 2023

Geophysical Research Letters

View full text Add to dashboard Cite

Using P‐wave seismograms, we trained a seismic source classifier using a Convolutional Neural Network. We trained for three classes: earthquake P‐wave, underground nuclear explosion (UNE) P‐wave, and noise. With the current absence of nuclear testing by countries that have signed the Comprehensive Test Ban Treaty, high quality seismic data from UNEs is limited. Even with limited training data, our model can accurately characterize most events recorded at regional and teleseismic distances, finding over 95% signals in the validation set. We applied the model on holdout datasets of the North Korean test explosions to evaluate the performance on unique region and station‐source pairs, with promising results. Additionally, we tested on the Source Physics Experiment events to investigate the potential for chemical explosions to act as a surrogate for nuclear explosions. We anticipate that machine‐learning models like our classifier system can have broad application for other seismic signals including volcanic and non‐volcanic tremor, anomalous earthquakes, ice‐quakes or landslide‐quakes.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Global Nuclear Explosion Discrimination Using a Convolutional Neural Network

Barama,

Williams,

Newman

et al. 2023

Geophysical Research Letters

View full text Add to dashboard Cite

show abstract

“…Although deep learning can automatically extract useful features from seismic waveforms without any knowledge of physics, it still has some limitations. For instance, features obtained may not have clear physical meaning, or they may not be applicable in new regions (Kong et al, 2022). In addition, they may implicitly pay close attention to features such as event location or timing while ignoring truly salient components of the seismic waveform, which may lead to a decrease in accuracy in practical applications.…”

Section: Introductionmentioning

confidence: 99%

Combining Convolutional Neural Network with Physics- Based Features in Shallow and Intermediate-depth Earthquake Discrimination

Zheng,

Lin,

Jin

et al. 2023

Preprint

View full text Add to dashboard Cite

It is crucial for earthquake early warning (EEW) to distinguish earthquakes of various focal depths accurately and reliably. However, this task is a significant challenge due to the difficulty in interpreting the underlying physical mechanisms of earthquakes of different focal depths. In this study, we proposed an algorithm that combines a convolutional neural network with physics parameter-based features (CNN-PP) to discriminate between shallow and intermediate-depth earthquakes. A total of 3586 earthquakes in Japan recorded by the K-NET and KiK-net strong-motion seismograph networks from 2003 to 2020 were collected and processed as research data; 38081 Three-channel acceleration seismic records were obtained by station record interception, baseline correction and quality screening along with other pre-processing procedures. Among them, 26644 and 11437 records were used as the training and the test dataset, respectively. The test results show that the CNN-PP model outperforms the CNN model in discriminating shallow and intermediate-depth earthquakes. In addition, we test the CNN-PP model with the seismic events (M ≥ 3) that occurred in Japan in February 2022, and the results confirmed that this model has good performance in discriminating earthquakes of varying magnitudes. The CNN-PP model can effectively discriminate shallow and intermediate-depth earthquakes and has great application potential in EEW.

show abstract

“…Overall, studying the black-box nature of DL model, implies interpreting how the data are fitted for some predefined task by using a specific DL architecture. In recent years rich set of various techniques has been developed for the purpose of interpreting a prediction process behind DL models (Barredo Arrieta et al 2020;Roscher et al 2020;Samek et al 2021;Ras et al 2021;Linardatos et al 2021;Kong et al 2022). It is highly crucial to recognize if DL model failed to represent training data and sometime sole prediction value is not enough to alert the user of the problem.…”

Section: Introductionmentioning

confidence: 99%

Interpreting convolutional neural network decision for earthquake detection with feature map visualization, backward optimization and layer-wise relevance propagation methods

Majstorović

Giffard‐Roisin

Poli

2022

Geophysical Journal International

View full text Add to dashboard Cite

Summary In the recent years, the seismological community has adopted deep learning (DL) models for many diverse tasks such as discrimination and classification of seismic events, identification of P- and S- phase wave arrivals or earthquake early warning systems. Numerous models recently developed are showing high accuracy values, and it has been attested for several tasks that DL models perform better than the classical seismological state-of-art models. However, their performances strongly depend on the DL architecture, the training hyperparameters, and the training datasets. Moreover, due to their complex nature, we are unable to understand how the model is learning and therefore how it is making a prediction. Thus, DL models are usually referred to as a “black-box”. In this study we propose to apply three complementary techniques to address the interpretability of a convolutional neural network (CNN) model for the earthquake detection. The implemented techniques are: feature map visualisation, backward optimisation and layer-wise relevance propagation. Since our model reaches a good accuracy performance (97 per cent), we can suppose that the CNN detector model extracts relevant characteristics from the data, however a question remains: can we identify these characteristics? The proposed techniques help to answer the following questions: How is an earthquake processed by a CNN model? What is the optimal earthquake signal according to a CNN? Which parts of the earthquake signal are more relevant for the model to correctly classify an earthquake sample? The answer to these questions help understand why the model works and where it might fail, and whether the model is designed well for the predefined task. The CNN used in this study had been trained for single-station detection, where an input sample is a 25 seconds three-component waveform. The model outputs a binary target: earthquake (positive) or noise (negative) class. The training database contains a balanced number of samples from both classes. Our results shows that the CNN model correctly learned to recognize where is the earthquake within the sample window, even though the position of the earthquake in the window is not explicitly given during the training. Moreover, we give insights on how a neural network builds its decision process: while some aspects can be linked to clear physical characteristics, such as the frequency content and the P- and S- waves, we also see how different a DL detection is compared to a visual expertise or an STA/LTA detection. On top of improving our model designs, we also think that understanding how such models work, how they perceive an earthquake, can be useful for the comprehension of events that are not fully understood yet such as tremors or low frequency earthquakes.

show abstract

Combining Deep Learning With Physics Based Features in Explosion‐Earthquake Discrimination

Cited by 34 publications

References 39 publications

Global Nuclear Explosion Discrimination Using a Convolutional Neural Network

Global Nuclear Explosion Discrimination Using a Convolutional Neural Network

Combining Convolutional Neural Network with Physics- Based Features in Shallow and Intermediate-depth Earthquake Discrimination

Interpreting convolutional neural network decision for earthquake detection with feature map visualization, backward optimization and layer-wise relevance propagation methods

Contact Info

Product

Resources

About