Absolute versus relative event location calibrated by induced blasts for real-time microseismic monitoring of post-mining ground failures

Accuracy of source locations of microseismic spatial swarms monitored in the vicinity of underground cavities by local seismic stations is a critical issue for the geotechnical expert in charge of assessing the local failure mechanism and its potential significance on the short-term overall stability. In order to quantify the gain in hypocenter resolution that earthquake relocation algorithms offer, multiplets were simulated using artificial sources performed in abandoned mines of the Lorraine iron-basin, region of north-eastern France. Sources consisted in small mining blasts shot in underground mine pillars at 200 meters deep, accurately controlled in terms of coordinates, orientation and energy. These "multiplet blasts" were configured to reproduce multiplet characteristics that may be defined as a group of spatially close events with similar waveforms, source mechanisms, and different origin times and magnitudes. With such a recorded dataset, first, the true orientation of all 3D sensors was counterchecked and corrected if necessary. Then a multilayered P wave velocity model was computed by taking into account all available input data, i.e. geological structure, positions of the blasts, direct P-wave arrival times and dips angles measured at the 3D probes on each experimented site. Those two steps were carried out to obtain a reliable 3D absolute location of the blasts from global inversion. On average, the misfit distance between measured and computed locations is found to be 70 meters, while for blasts correctly placed inside the microseismic network, this misfit comes down to 40 meters. This uncertainty appearing too high for clear distinction inside the several superimposed room-and-pillar mined layouts. These events were then reprocessed, using a relative location technique, considering all event pair combinations based on cross-correlated P- wave travel time delays computed in the spectral domain. Results show that relocation indeed improves the resolution, especially in the horizontal plane compared to the absolute location. If theoretical uncertainties are expected to be of the order of cm-dm, they are found to be of the order of tens of meter. Nevertheless, the true 2D geometry of the blast distribution is reliably and robustly estimated, reducing considerably error of interpretation in the risk analysis