Comparison of convolutional neural networks with matched-filtering for detection of induced seismicity
Bibliographic record
Abstract
The problem of extracting weak earthquake signals from continuous waveforms data recorded by networks of seismic sensors, referred to as earthquake detection, is a challenging and critical task in seismology. Waveform cross-correlation (matched-filtering) is a widely used method to detect weak earthquake signals with waveforms similar to those of known events. Even using this method, earthquake catalogs are often limited and incomplete, so there is a need for a more general detector. Because earthquakes occur infrequently, detection algorithms must be capable of quick processing of months to years of continuous data dominated by noise. To address these challenges, we propose to use convolutional neural networks, a new detection method that utilizes large seismic datasets to perform computationally efficient search to identify events in the continuous data. We set up an algorithm as a supervised machine learning problem and test its performance against matched-filtering based detector on synthetic and Kaybob-Duvernay region data. In this work, I introduce a DuverNet, which is an optimized convolutional neural network for the detection of induced seismicity in the Kaybob-Duvernay production region. First, I test two different convolutional neural network architectures: ConvNetQuake and VGG-Junior. Second, I test two different loss functions: cross-entropy loss and focal loss. My thesis is the first time focal loss is used to tackle class imbalance problem in earthquake detection. Focal loss helps tackling the problem of class imbalance and allows to achieve higher accuracy for convolutional neural network detectors. Third, I compare performance of matched-filtering and DuverNet. After allowing time for appropriate training, DuverNet performs best in computational runtime and memory use. Moreover it delivers superior detection performance compared to the matched-filtering detector. Synthetic data testing illustrates that DuverNet better generalizes to previously unseen events and is found to be better at detecting more events at lower signal-to-noise ratio compared to the matched-filtering method. Finally, my thesis introduces a novel dataset collected by the 6 UC/DSA array stations installed by Nanometrics for the University of Calgary.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".