It is well known that the seismicity in the Groningen area is due to gas extraction. The dynamics of how the site and rate of extraction is related to the magnitude and frequency of seismicity, however, is not understood. In a synergetic collaboration between the Department of Applied Mathematics at the University of Twente and the Department of Earth Sciences at Utrecht University, we propose to develop state-of-the-art tools to comprehensively address this extraction-seismicity causality.

We will infer 3D starting models of the subsurface using full waveform inversion. These 3D models will be used to calculate Green’s functions by randomly changing seismic sources, geometry, stress and elastic structure. In a time-lapse setting, these Green’s functions will be used to train neural networks, which will monitor the changes of the parameters of interest fast and in a fully probabilistic setting. Full waveform inversion and the generation of Green’s functions is computationally very expensive. We will therefore develop a reduced order model within the 3D wave propagation solver. Finally we will develop, based on stochastic geometry, a prediction tool for seismicity using gas production and these monitoring data as input.