SCEC Award Number 15004 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Statistical Study of Induced Seismicity from Wastewater Disposal in California from mid 1980s to Present
Investigator(s)
Name Organization
Mark McClure University of Texas at Austin
Other Participants
SCEC Priorities 2f, 2b, 2d SCEC Groups EFP, CS, CSEP
Report Due Date 03/15/2016 Date Report Submitted 03/03/2016
Project Abstract
In this study, we develop a statistical method for identifying induced seismicity from large datasets and apply the method to decades of injection and seismicity data in California and Oklahoma. The method is designed to be robust against a variety of statistical challenges. For example, injection and seismicity are nonrandomly located and spatially and temporally clustered, complicating the quantification of statistical significance and creating the potential for confounding factors to induce spurious relationship. The study regions are divided into gridblocks. A longitudinal study design is used, seeking associations between seismicity and wastewater injection along time-series within each gridblock. A statistical relationship is used that is flexible enough to describe the seismicity, which have high kurtosis and temporal correlation. In each gridblock, the maximum likelihood estimate is found for a model parameter that relates induced seismicity hazard to total volume of wastewater injected each year. A Gibbs sampler is used to infer the joint probability distribution of nuisance model parameters and marginalize them in order to calculate the likelihood values. Likelihood ratio tests are used to assess statistical significance in each gridblock and each state. In California, the analysis does not find a statistically significant relationship between wastewater disposal and seismicity. In Oklahoma, the analysis finds with very high confidence that the recent increase in seismicity is associated with wastewater injection. Our method could be applied to other datasets, extended to identify risk factors that increase induced seismic haz-ard, or modified to test alternative statistical models for natural and induced seismicity.
Intellectual Merit So far, the statistical approaches that have been applied to understanding induced seismicity from large datasets have been lacking in rigor. Statistical approaches to induced seismicity are still having to fight for respect among geoscientists, who prefer more conventional approaches, or simpler (but non-rigorous) statistical techniques. Geophysicists can’t take on induced seismicity alone- they need to bring together a broader cross-section of experts, including engineers and statisticians!
In this project, we developed a new method for study induced seismicity that overcomes many of the difficulties that make induced seismicity so challenging to describe statistically. We applied to 14 years of Oklahoma data and 34 years of California data, rigorously quantifying statistical confidence in whether induced seismicity (due to wastewater disposal) has taken place, both on the state and local level. This method could be extended to apply to other regions or to look for risk factors that increase the probability of induced seismicity.
Broader Impacts In 2015, Hornbach et al. wrote a paper identifying a clear correlation between injection and seismicity in Azle, TX. In response the Texas Railroad commission held a hearing to study whether the injection wells should be shut down. They concluded that the seismicity correlation with injection was suspicious, but it might be a coincidence.
This experience makes clear the danger of merely inspecting the data and drawing conclusions. It is subjective. Methods that rigorously quantify statistical significance could go a long way towards shooting down skepticism from regulators, businesses, and the public. Furthermore, we hope that our method could be built-upon in the future to seek risk factors that increase the probability of induced seismicity, which would enable injection practices to be modified to limit hazard.
Exemplary Figure Figure 2: Left panel: cumulative number of earthquakes (bright red representing > 50); middle panel: cumulative water injected (bright red representing >108 bbl); right panel: p-value for test model including induced seismicity (bright red representing near zero; blue representing 0.2 or higher). Maps are con- structed using the Mercator projection. Oklahoma City is shown with a green star.