Ayrat Abdullin

Ayrat Abdullin, Umair bin Waheed, Leo Eisner et al.

Seismic phase picking is very often used for microseismic monitoring and subsurface imaging. Traditional manual processing is not feasible for either real-time applications or large arrays. Deep learning-based pickers trained on large earthquake catalogs offer an automated alternative. However, they are typically optimized for high signal-to-noise, long-duration networks and struggle with the challenges presented by microseismic datasets, which are purpose-built for limited time without previously detected seismicity. In this study, we demonstrate how a network-wide earthquake phase picker, the Phase Neural Operator (PhaseNO), can be adapted to microseismic monitoring using transfer learning. Starting from a PhaseNO model pre-trained on more than 57,000 three-component earthquake and noise records, we fine-tune the model using only 200 labeled and noise seismograms from induced events in hydraulic-fracturing settings. The fine-tuned model thus preserves the rich spatio-temporal representation learned from abundant earthquake data, while adapting to the characteristics and labeling conventions of microseismic phases, which are often picked on peaks or troughs rather than onsets. We evaluate performance on three distinct real-world microseismic datasets with different network geometries and acquisition parameters. Compared to the original PhaseNO and a conventional workflow, the adapted model increases F1 score and accuracy by up to 30%, and strongly reduces systematic timing bias and pick uncertainty. Because the adaptation relies on a small, campaign-specific calibration set, the approach is readily transferable to other microseismic tasks where public earthquake data and pre-trained models are accessible. The associated code will be released openly at https://github.com/ayratabd/MicroPhaseNO.

Ayrat Abdullin, Umair bin Waheed, Leo Eisner et al.

Real-time monitoring of induced seismicity is crucial for mitigating operational hazards, relying on the rapid and accurate classification of microseismic events from continuous data streams. However, while many deep learning models excel at this task, their high computational requirements often limit their practical application in real-time monitoring systems. To address this limitation, a lightweight model based on the Fourier Neural Operator (FNO) is proposed for microseismic event classification, leveraging its inherent resolution-invariance and computational efficiency for waveform processing. In the STanford EArthquake Dataset (STEAD), a global and large-scale database of seismic waveforms, the FNO-based model demonstrates high effectiveness for trigger classification, with an F1 score of 95% even in the scenario of data sparsity in training. The new FNO model greatly decreases the computer power needed relative to current deep learning models without sacrificing the classification success rate measured by the F1 score. A test on a real microseismic dataset shows a classification success rate with an F1 score of 98%, outperforming many traditional deep-learning techniques. A combination of high success rate and low computational power indicates that the FNO model can serve as a methodology of choice for real-time monitoring of microseismicity for induced seismicity. The method saves computational resources and facilitates both post-processing and real-time seismic processing suitable for the implementation of traffic light systems to prevent undesired induced seismicity.

Ayrat Abdullin, Denis Anikiev, Umair bin Waheed

Deep neural networks like PhaseNet show high accuracy in detecting microseismic events, but their black-box nature is a concern in critical applications. We apply explainable AI (XAI) techniques, such as Gradient-weighted Class Activation Mapping (Grad-CAM) and Shapley Additive Explanations (SHAP), to interpret the PhaseNet model's decisions and improve its reliability. Grad-CAM highlights that the network's attention aligns with P- and S-wave arrivals. SHAP values quantify feature contributions, confirming that vertical-component amplitudes drive P-phase picks while horizontal components dominate S-phase picks, consistent with geophysical principles. Leveraging these insights, we introduce a SHAP-gated inference scheme that combines the model's output with an explanation-based metric to reduce errors. On a test set of 9,000 waveforms, the SHAP-gated model achieved an F1-score of 0.98 (precision 0.99, recall 0.97), outperforming the baseline PhaseNet (F1-score 0.97) and demonstrating enhanced robustness to noise. These results show that XAI can not only interpret deep learning models but also directly enhance their performance, providing a template for building trust in automated seismic detectors.