Linear versus non-linear earthquake location and seismogenic fault detection in the southern Tyrrhenian Sea, Italy

Abstract

We compare the performances of linear and non-linear hypocentre location methods working in 3-D velocity structures, a not-fully explored subject of main interest in the regions where the location problem is ill-conditioned. Comparisons are made between the linear location method known as SIMUL and the non-linear probabilistic algorithm named BAYLOC, using the data sets of the two main seismic sequences which occurred in the last decade in the southern Tyrrhenian Sea. We find that in the suboptimal network conditions of these sequences the SIMUL and BAYLOC algorithms furnish hypocentre coordinates of comparable accuracy leading to similar hypocentre spatial trends, while the location error estimates from SIMUL are, in general, less accurate than BAYLOC's. These findings are further supported by locations of synthetic events performed in the same network-model conditions of the real sequences. We conclude that linearized methods produce lower quality location error estimates but no overall bias in the hypocentral coordinates compared to non-linear methods. Therefore, we extend to 3-D location a conclusion drawn by previous investigators for 1-D location. Because location error estimates may be crucial to establish whether the hypocentre trend of a sequence does really mark the seismogenic structure or simply reflects ill-conditioning of the location process, we based on the BAYLOC probabilistic algorithm our approach to hypocentre trend evaluation for seismogenic fault detection. This procedure, named ISO-TEST, works through isotropic generation of synthetic hypocentres inside the sequence volume (simulations) and comparison by misfit variables of the location probability function of the sequence with probability functions from simulations. The application of ISO-TEST showed that while the NE–SW trend of one of the study sequences can only in minor part be ascribed to ill-conditioning of the location process, and then it may reasonably be proposed as the signature of the source, the NW–SE trend of the other is contaminated in a greater percentage by the location process, and we are led to conclude that source detection is doubtful in this case.