Testing seismic velocity models with quarry blast data

Seismic event locations are usually performed by means of iterative, linearized arrival time inversion considering 1D velocity models with fixed data errors. Both the use of inaccurate velocity structure or data error estimates may however affect the quality of event location and uncertainty evaluation. Here, we test a 3D P- and S-wave velocity model built in a previous work for Metropolitan France (the part of France located in Europe) and we compare it with the “Auvergne” 1D velocity model used by BCSF-RéNaSS (Bureau Central Sismologique Français - Réseau National de Surveillance Sismique) using quarry blast data. The reason for using quarry blast data is that, to some extent, their epicentral location and depth are known, which is not the case for earthquakes. We first identify potential active quarries over the territory of Metropolitan France by comparing catalog quarry blast locations with those from quarries visible from satellite images. Relocation is achieved by means of a Hierarchical Bayesian inversion procedure in which not only the hypocentral parameters (longitude, latitude, depth, origin time) are inverted for, but also P- and S-wave arrival time errors. The area of interest is a 1° by 1° zone located between 4°E and 5°E in longitude and 44°N and 45°N in latitude, the region where the Le Teil earthquake occurred (Mw 4.9, 2019/11/11). We first demonstrate the ability of the algorithm to properly determine hypocentral parameters and data noise using two simple synthetic experiments. Then we apply it to real data and relocate 147 quarry blasts that occurred in the region between 1980 and 2020 and that were located wit the “Auvergne” 1D velocity model. Relocations obtained with the 1D and 3D model are rather similar. Estimated data errors are larger, in both cases, than the amplitude of picking uncertainties, meaning that both models could be improved, by seismic arrival time tomography for instance. They are larger in the 3D case, suggesting that, from that point of view, the 1D model is in better agreement with the data. Distances between relocated hypocenters and the closest known quarry are comparable but relocations with the 3D model are characterized by shallower hypocenters than those obtained with the 1D model, so they appear more consistent with the fact that events are quarry blasts. In both cases, some events are quite far away from the closest quarry, suggesting that some of them might be natural events.