Statistical Controls on Induced Seismicity

By | November 27, 2018

Statistical Controls on Induced Seismicity

Conference paper by Yunjie Wen, Saurabh Sinha, Rafael Pires De Lima, and Kurt Marfurt

Seismicity in Oklahoma has shown a sharp increase since 2010 and is mostly attributed to wells used to dispose wastewater from hydraulically fractured production wells (McClure, Gibson, Chiu, & Ranganath, 2017). Many studies are conducted so far to include / eliminate various causes and solutions to this problem. The recent studies in this research area (Holland, 2011), show a general consensus on the main cause of the seismic events (earthquakes) as the high volume disposal in the Arbuckle formation causing the critically stressed faults in the basement rock to fail.

Some of these studies include the modeling and simulation of the physical processes whereas some studies delve into the statistical analyses of the relationships between disposal wells and induced seismicity. However, most of the previous attempts on the statistical analysis of this dataset use a more qualitative view of the problem (Langenbruch & Zoback, 2016) instead of quantifying the impact of various parameters such as injection rate, volumes, pressures, etc. Other work like Gogri et al. (2017) uses geo-modeling and simulation approach, but this constraint the modeling to a smaller section due to limitations on the seismic data extent.

In this work, we use various data analytics methods to quantify the impact of different injection well parameters and rock properties on the earthquake event magnitude and intensity. Our models show that hierarchical and K-means clustering are able to group the wells into clusters that conforms with the earthquake event density.

Including more clusters in our analysis refined the results but in general, four clusters are enough to capture the trends in our dataset. Fuzzy clustering, which is a soft clustering yields good results only after number of clusters exceed five.

For predictive modeling part, Gradient boost regression and random forest work better than least absolute shrinkage and selection operator (LASSO), elastic nets and linear models.


The complete study has been published by the SPE/AAPG/SEG Unconventional Resources Technology Conference, July 23, 2018 in Houston, TX.

Sinha, S., Wen, Y., De Lima, R. P., & Marfurt, K. (2018, August 9). Statistical Controls on Induced Seismicity. Unconventional Resources Technology Conference. doi:10.15530/URTEC-2018-2897507