Combined inversion of active and passive seismic data based on optimized data point selection

An accurate velocity model of the subsurface is required for many different applications and settings. Active and passive seismic data can both be used to derive meaningful seismic tomograms, but combining the two data sets remains challenging. In contrast to active seismic data, the hypocentral parameters of passive seismic data are not known and need to be considered during tomographic inversions.

Both, active and passive data sets have individual advantages and disadvantages. Therefore, combining them can be beneficial for obtaining a reliable velocity model. The spatial resolution and coverage of active seismic measurements can be tailored to specific applications and research questions, allowing targeted illumination of selected parts of the volume of interest with often relatively homogeneous ray density. However, the positioning of active sources and receivers is restricted to boreholes or the surface. In contrast, passive seismic data typically covers a larger volume and can illuminate regions that may not be accessible with active seismics, but the ray coverage of passive seismic data is often more heterogeneous, since seismic events commonly occur along fracture zones, leading to clustered source locations.

In our study, we show that a combined inversion of both data sets is feasible and allows their advantages to be exploited. The constraint provided by the active seismic data enables an iterative inversion of passive seismic source locations and seismic velocities using both data sets simultaneously. To combine the two data sets in a balanced and complementary way, we select only a subset of the passive seismic data for the inversion. This is achieved through a QR decomposition of the Jacobian, which ranks the information content of the passive seismic data relative to the information already provided by the active data set, thereby ensuring optimal complementarity. The method also enables an assessment of the appropriate amount of additional data to include in the inversion, helping to avoid overweighting densely sampled grid cells.

We apply this method to data from the BedrettoLab, an underground rock laboratory that serves as a testbed for geothermal injection experiments. From this test site, high-quality active and passive seismic data are available on a hectometer scale. With our approach, we increase the covered volume of the reservoir by a factor of 3 compared to the coverage purely based on active seismic data. We relocate the seismic events after each iteration, taking varying source locations due to the updated velocity model into account in the inversion. We further compare our results with a random data selection of passive seismic data, highlighting the importance of the optimized data selection, to maximize the illuminated volume.