Deciphering Seismogenic Patterns in Hydraulic Fracturing: A Machine Learning Approach in the Southern Montney Play

The burgeoning development of hydraulic fracturing (HF) for unconventional resource extraction has been paralleled by a rise in injection-induced earthquakes (IIEs), posing significant seismic hazards. A critical challenge in mitigating these hazards is the accurate assessment of the seismogenic potential and earthquake productivity of individual HF pads. We addresses this challenge by analyzing over 35,000 earthquakes in the Southern Montney Play (SMP), Western Canada, from 2014 to 2022, and associating them with 357 HF pads.

 

We employed the eXtreme Gradient Boosting (XGBoost) machine-learning algorithm, integrating fifteen geological and operational factors to evaluate their influence on IIE occurrence and intensity. We also utilized Shapley Additive Explanations (SHAP) values for a nuanced interpretation of the model outputs, providing insights into the relative importance and interaction of these factors.

 

Our analysis reveals that the cumulative injected volume and the location of HF pads within the Fort St. John Graben (FSJG) are the primary determinants of seimogenic potential (occurrence of IIE). In contrast, the number of HF stages targeting the Lower Middle Montney formation, cumulative volume from preceding injections, and the HF pad's location within the FSJG predominantly influence the seismogenic productivity (number of IIE). These findings suggest that both operational and geological factors are critical in determining the seismogenic productivity of HF pads. The XGBoost model demonstrated high predictive accuracy (R2 ~0.90), although its performance is constrained by the dataset's size and potential overfitting issues.

 

The study challenges the conventional understanding that proximity to known faults is a major factor in IIE occurrence, instead highlighting the significance of cumulative injection volumes and specific geological settings. The analysis also underscores the complex interplay between various factors, such as the correlation between the location fo the HF pads and the targed formation during HF stimulations, which may influence seismogenic patterns.

 

Overall, our result provides a comprehensive assessment of the factors influencing seismogenic behavior in HF-related IIEs, paving the way for more accurate forecasting of IIE activity levels for individual HF pads in the SMP. The findings have significant implications for seismic hazard assessment and risk mitigation strategies in regions undergoing HF operations. The application of machine learning in this context not only enhances our understanding of induced seismicity but also demonstrates the potential of such techniques in addressing complex geoscientific challenges.