Source inversion at teleseismic distances using a neighbourhood algorithm

K. Marson-Pidgeon, B.L.N. Kennett, M. Sambridge

The neighbourhood algorithm (NA) is a new direct search method for inversion (see previous annual reports), and has recently been applied to the problem of source inversion at teleseismic distances. The algorithm preferentially samples those regions of a multidimensional parameter space which have low data misfit. Any suitable definition of misfit can be employed, and the algorithm is based on the rank of the misfit function.

The first application of the NA approach is to event location using arrival time, slowness, and azimuth information (see the accompanying report ). The resulting hypocentre information is then linked to a waveform inversion of short period or broadband records, which uses the NA approach to invert for source depth and mechanism. The depth estimate obtained from the location inversion is used to initiate depth bounds for the waveform inversion. The source time function is specified by a trapezoid, and we invert for the rise time parameter. The source mechanism is represented via the superposition of a double couple and an isotropic component (giving a six dimensional parameter space), or alternatively, in terms of a general moment tensor (giving an eight dimensional parameter space). Synthetic tests with sources at crustal depths, using few stations, have recovered good estimates of the source parameters, in particular the source depth. With good azimuthal coverage ambiguities in source mechanism can be resolved. For smaller events, which will not be recorded at many stations, it is important to exploit all the available information, and make use of both SV and SH wave data in addition to P wave data. The waveform inversion has been applied to an event which occurred off the east coast of Honshu, Japan. This event has an estimated source depth of 19 kilometres, obtained from the centroid moment tensor (CMT) solution. Initial inversions resulted in poor fits to the observed data, though the depth estimate of 22 kilometres is close to the CMT depth. By allowing for a water layer above the source, the seismogram fit is much improved (Figure), and a depth estimate of 18 kilometres is obtained. Thus it seems that the source depth is fairly insensitive to other factors, such as the velocity model, which allows us to achieve good depth resolution.

Figure 1. Source radiation pattern and seismogram comparison for Honshu event. (left) CMT solution. (middle) Resulting source radiation pattern from inversion. (right) Comparison between the 'observed' seismograms (blue traces) and 'predicted' seismograms (light blue traces), for the vertical component of P.

Back to the beginning

Questions about the topics in this Report to Brian Kennett:
Brian.Kennett@anu.edu.au

Comments on the maintenance of these frames to Brian Kennett:
brian@rses.anu.edu.au