A Convolution Model for Earthquake Forecasting Derived from Seismicity Recorded During the ST1 Geothermal Project on Otaniemi Campus, Finland

We analyse and model the spatio-temporal evolution of seismicity induced by hydraulic stimulations at 6.1 km deep over 50 days during an EGS development on Aalto University's Otaniemi campus. We use the records from surface accelerometers and borehole seismometers installed at depth between 0.3 and 2.7km. The data were processed using Machine Learning techniques for phase detection (the Generalized Phase Detection), a back-projection technique for phase associations and location (Quake Migrate). Relative locations were refined using a cross-correlation techniques. The procedure yielded a catalog of ~70,000 events, including ~10,000 events which could be located precisely. Local magnitudes between range between -1.5 and 2. We investigate how the seismicity relates in time and space to the injection history. We find that the evolution of the seismicity in time can be successfully represented using a simple convolution model. The parameters of the model are calibrated from the seismicity data themselves, in particular the Omori-like decay of seismicity observed during shut-in periods. We use a simple physics-based simulator which assumes linear-pore pressure diffusion and earthquake nucleation governed by a standard Coulomb failure model to assess the validity of the convolution model approximation. This approach could in principle be used for time-dependent seismic hazard assessment, to feed a traffic light system or eventually for optimization and control during stimulation or operation of a geothermal well.