Empowering Machine Learning Forecasting of Labquake Using Event‐Based Features and Clustering Characteristics

• Published in: Journal of geophysical research. Machine learning and computation Vol. 1; no. 2
• Main Authors: Karimpouli, Sadegh, Kwiatek, Grzegorz, Ben‐Zion, Yehuda, Martínez‐Garzón, Patricia, Dresen, Georg, Bohnhoff, Marco
• Format: Journal Article
• Language: English
• Published: 01.06.2024
• Subjects: catalog‐driven features; clustering analysis; earthquake; forecasting; labquake; machine learning
• DOI: 10.1029/2024JH000160
• ISSN: 2993-5210

Following recent advances of machine learning (ML), we present a novel approach to extract spatiotemporal seismo‐mechanical features from Acoustic Emission (AE) catalogs to empower ML‐based forecasting. The AE data were recorded during laboratory stick‐slip experiments on granite samples cut by rough faults. Based on the features computed for a past time window, a random forest (RF) classifier is used to forecast the occurrence of a large magnitude event (MAE > 3.5) in the next time window. Event‐based features allow us to associate informative time‐space characteristics to each feature and nearest‐neighbor clustering analysis enables us to separate background and clustered seismicity and train individual models. The results show that the separation of AEs enhances the forecasting accuracy from 73.2% for the entire catalog up to 82.1% and 89.0% if background and clustered events are used separately. The presented new approach may be upscaled for applications to forecast tectonic earthquakes. Plain Language Summary It is widely discussed to use machine learning (ML) attempts to predict seismic events generated during rock deformation experiments in the laboratory. To improve these predictions, one needs to either refine the model or provide it with more detailed and informative physically understandable input data. In our study, we examine Acoustic Emission (AE) catalogs from laboratory experiments involving repetitive slips of the rough fault surfaces observed on three Westerly granite samples. By analyzing AE catalog data of past AE activity using a moving time window, we employ a random forest classifier, a type of ML tool, to forecast the likelihood of a significant earthquake happening in the future time period. What makes this approach different is that we introduce a new method to calculate features related to both location and timing of seismic activity. Our findings reveal that separating earthquake catalog into background and clustered seismicity is crucial for improving forecasting accuracy. Regarding an accuracy of 73.2% for the whole catalog, we achieved an enhanced accuracy of up to 82% for background events and 89% for clustered events. We discuss how the insights gained from this study can be scaled up for forecasting tectonic earthquakes in real time. Key Points We present an event‐based approach to extract spatiotemporal seismo‐mechanical features according to event time and location of labquakes Nearest‐neighbor analysis allows separating background and clustered events and defining new topological features of the clustered families Implementing individual classifier models for background and clustered populations significantly improves labquake forecasting