Stress-based forecasting of induced seismicity with instantaneous earthquake failure functions: Applications to the Groningen Gas Reservoir.
The Groningen gas field is a natural laboratory to test stress-based forecasting models of induced seismicity due to the detailed knowledge of the reservoir geometry and production history, as well as the availability of surface subsidence measurements and high quality seismicity data. A specific feature of that case example is the exponential rise of seismicity that was detected nearly 30 years after the onset of production. In this study, the subsurface is represented as a homogeneous isotropic linear poroelastic half-space subject to stress changes in three-dimensional space due to reservoir compaction and pore pressure variations. The reservoir is represented with cuboidal strain volumes. Stress changes within and outside the reservoir are calculated using a simple convolution with semi-analytical Green functions. The uniaxial compressibility of the reservoir is spatially variable and constrained with surface subsidence data. Coulomb stress changes are maximum near the top and bottom of the reservoir where the reservoir is offset by faults. To assess earthquake probability, we use the standard Mohr-Coulomb failure criterion assuming instantaneous nucleation and a non-critical initial stress. The distribution of initial strength excess, the difference between the initial Coulomb stress and the critical Coulomb stress at failure, is treated as a stochastic variable and estimated from the observations. We calculate stress changes since the onset of gas production. The lag and exponential onset of seismicity are well reproduced assuming either a a generalized Pareto distribution of initial strength excess, which can represent the tail of any distribution, or a Gaussian distribution, to describe both the tail and body of the distribution. This representation allows to test if the induced seismicity at Groningen has transitioned to the steady-state where seismicity rate is proportional to the stressing rate. Our results indicate that the system has not yet reached such a steady-state regime. The forecast is robust to uncertainties about the ability of the model to represent accurately the physical processes. It does require in particular a priori knowledge of the faults that can be activated. The method presented here is in principle applicable to induced seismicity in any setting provided deformation and seismicity data are available to calibrate the model.