Probabilistic centroid moment tensor inversion and rupture analysis of incuded seismic events in the Groningen gas reservoir

For more than a decade, earthquakes induced by natural gas extraction have been a significant societal concern in Groningen, the Netherlands. Their occurence underlines the importance of understanding earthquake source characteristics and the development of suitable and accurate characterization methods. In this study we (i) estimate the source characteristics of ten relatively strong induced events (magnitude > 2 M_L) and (ii) analyse the rupture propagation of these events. The determination of the source characteristics (aspect i) involves the estimation of centroid-moment tensors (CMT) by means of an iterative workflow based on the Hamiltonian Monte Carlo algorithm. Importantly, this approach allows us to quantify the uncertainties of the model parameters (hypocenter, origin time, and moment tensor components) as the Markov Chain asymptotically approaches the posterior probability of these model parameters. The Bayesian inference problem is paired with geological prior knowledge of the Groningen subsurface (i.e., a detailed 3D velocity model and the know fault geometry). For the rupture propagation analysis (aspect ii), we employ the Empirical Green's Function method and exploit the dense sampling of the wavefield in Groningen resulting from the extensive seismic network in the region. This analysis allows us to estimate the directivity and speed of the ruptures, as such giving insight into the kinematics of the ten selected earthquakes.

 

We find that the lateral coordinates of the estimated centroids (posterior means) are consistent with the available Groningen fault map. Furthermore, the depths are mainly distributed in the vicinity of either the top or bottom of the gas reservoir. In terms of source mechanisms, the earthquakes are predominantly explained by double-couple sources featuring normal faulting. After conversion of the mean of the ensemble of moment tensors to strike, dip, and rake, we obtain values consistent with the known fault geometry. As for the rupture propagation analysis, our results indicate that these earthquakes display a relatively minor directivity effect. In spite of the minimal effect, however, the rupture directions are mostly consistent with the strike derived from both the MTs and the available fault map.