← Back to Sessions

Micro-EQpolarity: Transfer Learning for Microseismic P-wave First-motion Polarity Determination and Its Application in the Western Canada Sedimentary Basin (WCSB)

Description: 

Microseismic focal mechanism solutions (FMSs) are essential for understanding reservoir stress state and rock fracturing during hydraulic fracturing. Focal mechanism inversion based on P-wave first-motion polarity (FMP) has been widely utilized due to the deployment of dense seismic arrays. However, preparing high-quality polarity data is highly labor-intensive and time-consuming, making focal mechanism inversion for extensive microseismic events challenging. While existing artificial intelligence (AI)-assisted workflows, which have been trained on data from moderate to large earthquakes, have demonstrated efficiency and reliability in applications, their performance diminishes when applied to microseismic events. This decline is likely due to differences in signal-to-noise ratio (SNR) and rupture mechanisms between microseismic events and larger earthequakes. Thus, a clear need exists for an AI-based model tailored for microseismic FMP determination.

We propose Micro-EQpolarity, a fine-tuned extension of the recently proposed deep learning model EQpolarity, which leverages transfer learning to determine the P-wave FMP of microseismic events. The model was initially trained on the Southern California Seismic Network (SCSN) dataset and subsequently fine-tuned using the Tony Creek Dual Microseismic Experiment (ToC2ME) - a dataset from western Canada, encompassing a hydraulic fracturing well-pad. Specifically, we manually picked 19,724 FMPs from representative microseismic events to fine-tune the pre-trained model. The resulting model achieves 99.19% accuracy, showing a significant improvement of 29.26% over the baseline pre-trained model. When applied to the full ToC2ME dataset, Micro-EQpolarity successfully determines FMSs of 2,519 events with magnitudes as low as -1.4. This new catalog contains four times more FMSs than previously reported 530 events and reveals four distinct types of FMS. These focal mechanisms enable illuminating fine-scale fault structures (e.g., a well-defined en-echelon structure) at an unprecedented resolution and offer new insights into the complex regional fracture network.

Session: Advances in Reliable Earthquake Source Parameter Estimation [Poster]

Type: Poster

Date: 4/16/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Jiachen

Student Presenter: Yes

Invited Presentation: 

Poster Number: 27

Authors