Benchmarking earthquake location algorithms: A synthetic comparison

A priori assessment of the expected location accuracy of a sensor network is typically done through inversion of the traveltime spatial gradients. This approach assumes that the applied location algorithm successfully recovers the global minimum of the objective function. However, even for accurate and precise phase picks, complexity in the velocity model and limitations in the network layout may inhibit the finding of a global minimum. The location algorithms may end up in a local minimum instead. We compare the location performance of various objective functions and minimization approaches. Although most of the analyzed location approaches mostly lead to good location results, none of the analyzed approaches recovered the correct location for all event locations. This implies that (microseismic) event locations estimates include an additional inherent error, which is linked to the applied location approach. This aspect is often neglected when interpreting event locations. Our site comprises a sensor network of two vertical strings, in a 1D velocity model of the Groningen gas field (The Netherlands), complicated by two thin, very high-velocity layers. For a series of synthetic event locations, we calculate arrival times, add picking errors, and then feed these synthetic picks into a set of different location routines. We also determine a novel way to analyze this approach-dependent location error of the sensor network for the given velocity model and a set of event locations: We compare the distances between a set of assumed event locations to resulting locations in the target region of the subsurface. From the cumulative distribution function of that mislocation distance, we determine the 1 sigma and 2 sigma confidence distances for each method. This results in scalar values representing the location confidence distances for a given method. In turn, this can be used to easily compare the location capabilities of different sensor layouts.