Many types of industrial activity require the subsurface injection of fluids. Seismic activity induced by fluid injection has become a topic of increasing concern in recent years. In order to mitigate this issue, models capable of simulating injection-induced seismicity are required. In this paper we outline two contrasting model approaches: a statistics-based approach and a numerical modeling approach. We apply both model types to different case studies, and evaluate the advantages and drawbacks of each model type.
Fluids are injected into the subsurface to hydraulically stimulate tight formations, dispose of produced water from conventional reservoirs or flowback water after hydraulic stimulation, and dispose of CO2 to mitigate anthropogenic climate change. Injection commonly results in small-magnitude (ML < 0) "microseismic events", which are located with geophone arrays either in boreholes near to the injection or dense near surface arrays that use stacking to boost signal levels. However, occasionally larger "felt" (i.e. of sufficient magnitude to be felt by local populations) events have also been triggered by injection.
The industry generally wishes to avoid such events: even if events do not cause damage, they tend to foment public opposition, leading to increased regulatory oversight or even moratoria on their operations. In particular, traffic-light schemes (TLS) are being used to regulate the industry, where operations must cease if events are triggered above a given magnitude threshold. Given the potential cost of exceeding such a threshold – for example being required to cease all further operations in a well – there is a need for
operators to develop methods to mitigate induced seismicity from their operations, allowing them to identify sites at which a larger event might occur, and to moderate their activities accordingly (by reducing injection rates, for example).
Broadly speaking, models for induced seismicity fall into two categories: physical and statistical. Physical models numerically simulate the processes occurring in a reservoir, while statistical models use microseismicity observed during operations to date to forecast future seismicity. In this paper we present an example of each type of approach, and discuss their respective strengths and weaknesses.