Retrospective seismology by source-receiver interferometry
Seismology is the study of earthquakes and the Earth’s internal structure using seismic waves. Traditional seismology is constrained by the timing and location of seismic sources, and by the location of seismometers with which energy from the sources are recorded. Improvements in the global seismometer networks have reduced the latter constraint. Furthermore, recent advances into Seismic Interferometry (SI) have enabled detailed information about the Earth’s interior to be obtained using ambient seismic noise, hence even in areas with low natural seismicity. The most common approach to SI is to use the cross-correlation of ambient noise recordings to construct an estimate of the Green’s function between two seismometer locations. The Green’s function estimate is then analysed or inverted for seismic properties of the Earth. This method of noise interferometry is now a popular approach in earthquake seismology as in some situations it renders active seismic sources (earthquakes or synthesised explosions) obsolete, as subsurface information can be obtained even in times of seismic quiescence. This thesis investigates a different method: Source-Receiver Interferometry (SRI). SRI can be used to construct earthquake seismograms on seismometers that were not necessarily deployed when the earthquakes occurred - a form of ‘retrospective seismology’. This might be useful if, for example, we wish to analyse old earthquakes with newly installed seismometers. The application of SRI involves evaluating two interferometric integrals. The first integral is evaluated using ambient noise interferometry: at least 6 months of noise data is cross-correlated to estimate the Green’s functions between pairs of seismometers. These inter-receiver Green’s functions are then used as the “propagators” for SRI. Their role is to project earthquake signals recorded on a backbone array of seismometers to the location of a target sensor at which a new, novel earthquake seismogram is to be constructed - a form of spatial redatuming. To spatially redatum the earthquake data, the second interferometric integral is evaluated using either processes of correlation (resulting in correlation-correlation SRI) or convolution (correlation-convolution SRI). The method used depends on the relative location of the target sensors with respect to both the backbone seismometer array and the earthquake epicentre. The SRI process is completed by integrating (summing) over all projected earthquake signals. To regularise the spatial distribution of the projected earthquake data and to invoke this second interferometric integral more precisely, the backbone seismometers are embedded within 2D spatial Voronoi cells. New seismograms for 87 earthquakes were reconstructed on up to eight target sensors, seven of which were deployed when the earthquakes occurred and are used to test the success of the method by comparing with the SRI results with the directly-recorded seismograms. The seismogram reconstructions on the eighth target sensor are truly novel. The SRI method was developed to operate over two length scales. The first focusses on relatively small length scales in which the inter-station distance between the eight target sensors and the backbone array seismometers is between ~ 210 km and 540 km. Both correlation-correlation SRI and correlation-convolution SRI are used to reconstruct the earthquake seismograms on four of the same target sensors. Applying correlation-convolution SRI is shown to remove spurious signals associated with correlation-correlation SRI. Second, a significantly larger length scale is considered where a second set of target sensors are located up to 2420 km from a second backbone seismometer array. The correlation-correlation and correlation-convolution SRI methods are used in parallel to increase the spatial extent of the study. The quality of the SRI seismograms constructed is shown to depend on the quality of three components: 1) the SRI propagators constructed using ambient noise interferometry, 2) the earthquake signals recorded on the backbone seismometer array, and 3) the correlation (or convolution) functions that are summed in the second interferometric integral to construct the final SRI seismogram. The quality of each component is quantified by its signal-to-noise ratio and root-mean-square value, and criteria are proposed to obtain optimal earthquake seismogram reconstructions using SRI. SRI is most successful when the target sensors are located less than 540 km from the backbone array seismometers. Such SRI seismograms are being used to create a catalogue of new, ‘virtual’ earthquake seismograms that are available to complement real earthquake data for use in future earthquake seismology studies. An alternative approach to noise interferometry is also considered: the recordings from just 15 earthquakes are used to perform multidimensional deconvolution (MDD) to estimate the Green’s functions between pairs of seismometers. This is the first time such data has been used to perform MDD, which is valid in attenuating media and is thus theoretically more valid in earthquake seismology settings than correlational interferometry. The Green’s functions estimated using MDD are compared with those same Green’s functions estimated using ambient noise interferometry and the results are comparable on several occasions, despite using far fewer data for MDD. However, the quality of the results of MDD is significantly affected by the illumination of the receiver array from the earthquake sources. A greater density of earthquakes that sufficiently illuminates all backbone array seismometers is required to obtain accurate Green’s functions by MDD.