Point-source inversion neglecting a nearby free surface: Simulation of the underground Research Laboratory, Canada

Abstract

The aim of this study was to investigate the effects of the inconsistency between the model used and actual structure on the inversion of point-source parameters. Specifically, we studied the effects of ignoring a reflecting interface in the model. Such a situation frequently occurs when events with sources close to mine excavations are recorded and subsequently processed using a simplified model describing the interface only approximately or ignoting completely. In this case, it is important to ask how reliable the retrieved maichanism is. An algorithm with indirect parametrization of the source (INPAR) was applied. A numerical study was performed, in which the mechanosm and source time function of a point source were sought ina model inconsistent with the actual structure. The ray synthetic wavfield generated by a source situated close to a free surface of a homo geneous half-pace was treated as the observed data. The free surface was then ignored when constructing the Green's function necessary for the point-source inversion. Thus, the incosistent response of the medium including only direct phases was used to invert the observed seismograms, which contain both direct and reflected phases. The configuration of the Underground Research Laboratory (URL) of the Atomic Energy Canada Ltd. was simulated in the synthetic experiments. 16 triaxial were situated around the underground tunnel, the face of which was taken to represent the free surface near the hypocentre. High-frequency seismograms were synthetized with a frequency of around 10 kHz. This is close to the prevailing frequency of the actual URL records. It was found that gross features of the source such as orientation of the double couple and the general features of the source time function can be retrieved satisfactorily when the hypocentre is localized. Formal error analysis, however, yields rather large error estimates due to the omission of the free surface, providing us with acceptably constrained solutions at about 70 per cent confidence level only. Mislocation of the hypocentre of the order of seven to 14 wavelenghts and/or contamination of the data by noise with an amplitude amounting to 20 per cent of the data amplitude both distort retrieved source parameters and make rather uncertain. The retrieved orientation deviates by more than 20° and its 70 per cent confidence region extends several tens of degrees. As a consequence of the mismodelling of the medium represented by the neglect of the free surface, spurious non-double (DC) components appear in the mechanism. Two URL events from 1991 September 25 separated by 4 s were processed, and a large majority of the compensated linear-vector dipole (CLVD) was found there. The CLVD along the P-axis at the earlier event was replaced by the CLVD along the T-axis at the following one, which suggests an over-relaxation of the stress the first event and its partial restoration by the subsequent event.