In order to improve our understanding of the large-scale ground failure phenomena caused by old underground mining works, a solution mine was instrumented in 2004 prior to its collapse as part of the mining scheme. A permanent monitoring system was set up, including a high-resolution microseismic network linked to a surface field-displacement measurement system. The large amount of data transmitted for on-line processing provided daily insight into the evolution of the geological system. First, microseismic activity showed upward progressive failure migration throughout 2008 without any significant surface movement. Second, after two days of intensive brine extraction, a high microseismicity and energy release rate marked the failure of a thin and very rigid bed at a depth of 120 m. This failure occurred 24 hours before the final collapse; it was followed by transient brine pressure signals, and by acceleration of the surface subsidence rate, reaching 1 m/h in the final phase.

Ground failures, caving processes and collapses of large natural or man-made underground cavities can produce significant socio-economic damages and represent a serious risk envisaged by the mine managements and municipalities. In order to improve our understanding of the mechanisms governing such a geohazard and to test the potential of geophysical methods to prevent them, the development and collapse of a salt solution mining cavity was monitored in the Lorraine basin in northeastern France. During the experiment, a huge microseismic data set (∼50 000 event files) was recorded by a local microseismic network. 80 per cent of the data comprised unusual swarming sequences with complex clusters of superimposed microseismic events which could not be processed through standard automatic detection and location routines. Here, we present two probabilistic methods which provide a powerful tool to assess the spatio-temporal characteristics of these swarming sequences in an automatic manner. Both methods take advantage of strong attenuation effects and significantly polarized P-wave energies at higher frequencies (>100 Hz). The first location approach uses simple signal amplitude estimates for different frequency bands, and an attenuation model to constrain the hypocentre locations. The second approach was designed to identify significantly polarized P-wave energies and the associated polarization angles which provide very valuable information on the hypocentre location. Both methods are applied to a microseismic data set recorded during an important step of the development of the cavity, that is, before its collapse. From our results, systematic spatio-temporal epicentre migration trends are observed in the order of seconds to minutes and several tens of meters which are partially associated with cyclic behaviours. In addition, from spatio-temporal distribution of epicentre clusters we observed similar epicentre migration in the order of hours and days. All together, we suggest that the recorded microseismicity mainly represents detachment and block breakage processes acting at the cavity's roof, indicating a zone of critical state of stress and where partial fractures cause chain reaction failures as a result of stress redistribution processes.