Technical Report NTB 97-05
Grimsel Test Site:Further Development of Seismic Tomography
The project entitled "Further Development of Seismic Tomography" has two components: evaluating and testing underground seismic sources for use over large measuring distances and improvement of analysis (inversion) methods in terms of stability, quality and resolution.
Various high-frequency seismic sources have been tested at the Grimsel Test Site (GTS) (Bühnemann, 1996; Bühnemann & Holliger, 1998). The tests were designed to facilitate future tomographic studies of potential radioactive waste disposal sites. A key objective was to identify borehole and tunnel seismic sources capable of generating and sustaining high-frequency signals over distances of up to 1000 m. Seismic sources were located in both water-filled boreholes (sparker, two piezo-electric sources, explosives) and at the tunnel wall (accelerated weight drop, minivibrator, bolt gun, buffalo gun, explosives). In order to evaluate and compare the source characteristics, the direct P-wave generated by the various sources was investigated for decay of its signal-to-noise ratio and dominant frequency with offset and for the maximum distance at which pickable first arrivals could be recognised. Of the seismic sources tested, small explosive charges (5-100 g) were found to have the most favourable energy and frequency characteristics. At the GTS, the target distance of around 1000 m was reached with explosive charges of around 50 g or more. None of the sources tested was capable of sustaining frequencies of 1000 Hz over distances in excess of 100 to 200 m.
The seismic waves are strongly attenuated in the region of the GTS. Q values estimated using the spectral ratio and rise time methods range from about 20 to 60, with a median value of 35 and a standard error of 20% or less (Holliger & Bühnemann, 1996). This explains the observed rapid depletion of high frequencies with increasing offset. In the investigation areas Zürcher Weinland (Opalinus Clay) and Wellenberg (marl) a similar attenuation is expected (Schön, 1996; Tonn, 1989). In the crystalline basement of Northern Switzerland, the attenuation should be lower and the propagation characteristics of the seismic waves should be correspondingly better.
Even when small amounts of explosives (5 g) are used as seismic source, it is to be expected that there will be damage to the borehole wall. This was confirmed by televiewer measurements performed after blasting in borehole BOGS 84.041 a.
In the component of the project dealing with improvement (and development) of analysis techniques in terms of stability, quality and resolution, 3 inversion techniques were tested and developed using the dataset US85 (Gelbke, 1988). Two travel time inversions – anisotropic velocity tomography – AVT (Pratt & Chapman, 1992) and coupled inversion – Cl (Maurer, 1996; Maurer & Green, 1997) – and a wave field inversion (WFI – Song et al., 1995) were used.
Several problems occurred in the first inversion of the US85 dataset using the Simultaneous Iterative Reconstruction Technique (SIRT); these were due to the velocity anisotropy of the rock, the triggering inaccuracy of the shots and uncertainties regarding the source/receiver locations in the boreholes. In the AVT, the velocity anisotropy of the rock is taken into account as a free parameter. In addition to an "isotropic" velocity image, this involves producing tomograms of anisotropy. Taking into account the anisotropy of the rock allows the artefacts of the SIRT inversion to be explained and the travel time inversion to be stabilised.
A fundamental assumption of tomographic inversion techniques currently in use is that the coordinates of the boreholes and tunnels containing the seismic sources and receivers are accurately known. By inverting both synthetic and observed travel time data, it can be demonstrated that relatively minor coordinate errors (1-2%) in the deeper sections of long boreholes (> 100 m) may produce artefacts in the tomographic images that are comparable in extent and amplitude to true velocity anomalies. To address this problem, the coupled inversion method (Maurer, 1996) commonly used in earthquake studies is introduced as a means for simultaneously determining borehole coordinate adjustments and an estimate of the tomographic image. Coupled inversions of two independent subsets of traveltime data (Field US1 and Field US2) that involve a common central borehole (BOUS85.002), together with a coupled inversion of the entire dataset (Field US1, Field US2 and Field US3), yield consistent coordinate adjustments for all boreholes. This also allows the artefacts of the SIRT inversion to be explained and the travel time inversion to be stabilised.
Further investigations using synthetic datasets showed that, for the US85 data, it is impossible to draw a distinction between a weak anisotropy and borehole uncertainties based on travel times only. However, given the inversion results, it is likely that both effects are present.
The WFI was fully tested on synthetic data before processing the real data. Two synthetic datasets were used: the first was generated using the same (acoustic) wave equation software used in the wave field inversion process. The second was generated by Professor M. Korn's group at the University of Leipzig. The data from Professor Korn comprised a full, 2D elastic wave simulation.
The synthetic data were fully processed with the sequence used for the field data. This sequence includes:
i) Travel time tomographic imaging
ii) Data pre-processing (projection of multi-component data, time windowing and amplitude normalisation)
iii) Frequency domain wave field inversion
The results obtained using the synthetic data confirmed that a dramatic increase in resolution relative to travel time tomography can potentially be obtained when wave field inversion is used. The use of synthetic data allowed an optimum pre-processing sequence to be designed with confidence and also confirmed that amplitude normalisation, which is necessary for the real data, does not adversely affect the final velocity images – on the contrary it can improve the convergence rate.
The real data differ from the synthetic data in two major respects. First, the field data contain a higher noise level. The noise consists of random background acoustic noise, systematic problems with source static time shifts and spurious trace-to-trace amplitude variations. The second major effect in the real data, which is not present in the synthetic data, is a small but consistent level of P-wave anisotropy. These problems led to the following modifications being incorporated into the wave field inversion algorithm:
i) A constraint on the images that enforces a degree of "smoothness"
ii) Including the source function in the inverse problem. The data were divided into a number of "groups", each having an approximately consistent source static.
iii) Accounting for a background level of anisotropy by applying a "stretch" to the geometry that simulates anisotropy. In order to account for the apparent angle of the principal anisotropy axis, a coordinate transformation, stretch and retransformation procedure was successfully implemented. All three were critical in obtaining reliable and interpretable images. The final images show the expected improvement in resolution and enhance the detailed characterisation of the test site.
The resulting tomograms were then validated in three steps. The first step involved an internal consistency check which compared the velocity images for the separately inverted Fields US1 and US2 at the field boundary (borehole BOUS85.002). A good agreement of the velocities along borehole BOUS85.002 was achieved for all the in-version techniques.
In the second step, the agreement between the tomograms and the results of seismic velocity measurements performed in boreholes (sonic logs) was assessed. Sonic logs were available for 9 boreholes in the investigation area (BOUS85.001, BOUS85.002, BOUS85.003, BOBK85.004/BOBK86.001, BOBK85.008, BOBK86.003, FEX95.001 and FEX95.002). In the third and final step, the tomograms were compared with a geological model based on all the geological and hydrogeological data from the US, BK and FEBEX zones. Compared with the SIRT inversion, both the new travel time inversions (AVT and Cl) provide a more stable image with less artefacts which coincides better with the measured sonic velocities and geological structures. The difference between the AVT and Cl inversions is minimal. The WFI has the best resolution and the best agreement with the sonic results and geological evidence but it also produces the image with the most artefacts. The reliability of the interpretation is reduced both by strong oscillations mainly at the field boundaries and the extent to which it depends on the selected global anisotropy corrections. More certainty in interpretation can be achieved only by further developing this algorithm for the elastic wave equation, taking into account anisotropy.