string(18) "[Energy-Resources]"

Development of an Induced Seismicity Susceptibility Framework and Map for NEBC using an Integrated Machine Learning and Mechanistic Validation Approach

Lead Researcher(s):  E. Eberhardt

Key Researcher(s):  Afshin Amini, Ali Mehrabifard

Project ID:  2019-014

Key Research Organization(s):  University of British Columbia

Project Location:  Northeast BC

Strategic Focus Area:  Energy-Resources


This project combined multiple data sets in a machine learning and advanced numerical analysis review, together with laboratory rock data and numerical simulations, to model the relationship between natural gas hydraulic completions, geology and seismic activity in the Montney Play region in British Columbia’s Northeast Region, and produced induced seismicity susceptibility maps.

The project was one of a series of four research projects started in December 2019 to further investigate how and why, in certain circumstances, earthquakes can be caused by hydraulic fracturing and wastewater disposal during natural gas development. In-kind support for the project was provided by geoLOGIC.

View Related Projects

Research Statement

Numerous factors influence the potential for the hydraulic fracturing and wastewater disposal processes used for natural gas development to cause earthquakes, a phenomenon known as induced seismicity.

Since 2012, Geoscience BC has been funding induced seismicity research to better understand when, where and why induced seismicity occurs and to inform industry, government and community decisions so that the likelihood of future induced seismicity can be reduced.

This project developed induced seismicity susceptibility maps for the Montney Play, an area within the Western Canadian Sedimentary Basin in BC’s Northeast Region that contains some of North America’s most significant natural gas deposits. These maps can be used by regulators and the natural gas sector to help assess risk from hydraulic fracturing and wastewater disposal operations.


This Energy project fits under Geoscience BC’s Strategic Objective of Facilitating Responsible Natural Resource Development and our goal to:

  • Maintain joint research with partners examining seismicity induced by hydraulic fracturing in northeastern BC to provide new science to better understand induced seismicity, mitigate risks and further improve regulation and industry practices.

Specifically, this project:

  • compiled relevant public data from oil and gas wells (e.g. well location, completion metrics, pressure data and geology data);
  • tested various machine learning techniques to assess the data sets;
  • conducted numerical simulations to confirm relationships and feed into the machine learning algorithms;
  • identified and improved understanding of the major factors controlling induced seismicity in the Montney Play; and
  • identified areas needing more focused research.


Benefits of this project include:

  • a public database of geological and operational data;
  • machine learning and advanced numerical analysis methodologies for peer review and research;
  • induced seismicity susceptibility maps;
  • results from numerical simulations supporting results from susceptibility mapping; and
  • potential to decrease the number of felt seismic events.

Location Details

The Montney Play is in BC’s Northeast Region, in the territories of Treaty 8 First Nations.

What was found?

Eight machine learning algorithms were tested across four sets of machine learning analyses, with top performing models selected for interpretation:

  1. a classification feature importance analysis, based on a combined data set of all geological and operational features at each hydraulic fracturing well;
  2. a classification susceptibility analysis, based on a restricted data set limited to geological features and those operational features that apply to susceptibility;
  3. a classification susceptibility mapping analysis, based on the restricted data set of geological and susceptibility‐relevant features interpolated to a 2.5 km grid spacing; and
  4. a regression severity analysis, based on a combined data set of geological and operational features that included the maximum magnitude of the induced seismicity events associated with each well.

Feature importance analysis showed that, in general, geological features ranked higher in importance than operational features.

Regarding induced seismicity susceptibility, the depth to the top of the basement was identified as the most important predictor of a well being seismogenic: a shallower depth to the basement increases the chance of a well being seismogenic. Other key factors included the magnitude and orientation of the sub-surface stress field.

Machine learning results were used to generate induced seismicity susceptibility maps using a 2.5 km grid resolution. Overall, the locations of susceptible areas correlate with the historical location of induced seismicity events.

Together, the machine learning and numerical modelling results suggest that the conditions that best serve as an indicator for the potential of a large induced seismicity event (Mw ≥ 4), are the combination of operating within a strike‐slip far‐field stress regime, in stiffer rocks, with little early, significant seismicity (of Mw > 1 or 2) with increasing injection time. Similarly, in the context of a Traffic Light Protocol, the detection of a Mw 2 event might be more concerning when the formation targeted by the injection, or adjacent to it, is stiffer and if operating within a strike‐slip far‐field stress regime.

The report provides a number of recommendations and guidelines to improve modelling and improve use of machine learning to generate seismicity susceptibility maps.