Robert Bainer, Environmental Restoration Division, Lawrence Livermore National Laboratory, Livermore, CA (November 10th, 1999) Abstract

Transcription

1 3-D velocity imaging in the shallow subsurface using multi-well, multi-offset, VSP data: a case study from the Lawrence Livermore National Laboratory site. Paul A. Milligan and James W. Rector III, Engineering Geoscience, Dept. Material Science & Mineral Engineering, University of California, Berkeley, CA Robert Bainer, Environmental Restoration Division, Lawrence Livermore National Laboratory, Livermore, CA (November 10th, 1999) Abstract A vertical seismic profiling (VSP) method was developed to produce compressional (p-wave) velocity sections in a 3-D survey volume, down to a maximum depth of 57 m. The VSP method consisted of a multilevel hydrophone tool deployed down several wells. Multiple offset surface shotpoints were recorded at multiple azimuths around each well head, using an impact source. The VSP data were processed and inverted to produce p-wave interval velocities. These velocity sections were then projected into a model of the survey volume, together with borehole log information and other geologic information, to produce a 3-D visualization which aided the location and interpretation of aquifer boundaries below the water table. The final image displayed several low velocity zones, and these were attributed to partially saturated pore spaces, possibly both natural, or pump induced by air suction. Low velocity layers appeared above several actively pumped aquifers, and it was concluded that the multi-offset, multi-azimuth, VSP method may be suitable for mapping both aquifer channels, and zones of pumping influence. 1

2 Introduction Environmental characterization and cleanup has been progressing at the Lawrence Livermore National Laboratory (LLNL) Livermore Site, Livermore, California since 1983 (Fig. 1). In 1987, LLNL was added to the U.S. Environmental Protection Agency National Priorities (Superfund) List, and serious evaluation and cleanup activity began. Initial releases of hazardous materials to the environment occurred in the mid to late 1940s when the site was used by the U.S. Navy as an air training base. Most of these releases are believed to have been cleaning solvents. Smaller releases of gasoline, diesel and other compounds are also known to have occurred. From 1950 to 1954, California Research and Development Company, a subsidiary of Standard Oil, occupied the southeastern portion of the site. This marked the beginning of testing with radioactive materials at the site and probably the first releases of radioactive materials to the environment. LLNL has occupied the site since 1952, and additional releases of volatile organic compounds (VOCs), polychlorinated biphenyls (PCBs), metals, radionuclides (primarily tritium), gasoline, and pesticides have also occurred. Since 1983, over 800 boreholes have been drilled and cased to create both monitoring, and extraction wells for the contaminated ground water and vadose zones. Fig. 1 includes a contour map of ground water VOC (volatile organic compounds) concentrations in parts per billion (ppb) estimated for the LLNL Livermore Site in A concerted effort began in 1987 to extract and treat contaminated ground water, resulting in about a ten times reduction in concentrations over the first five years, reducing the contaminant mass to less than 100 ppb over most of the Livermore Site. Several smaller areas of high concentration remain. With the shrinking of the contaminant plumes, more accurate and higher resolution imaging of aquifer structures are required for both contaminant monitoring and the accurate location of new wells. Prior site characterization and hydrogeology The LLNL Livermore Site and surrounding vicinity are underlain by over 100 m of heterogeneous, unconsolidated alluvial sediments, consisting of clays, sands, gravels, and silts. Depth to ground water is over 33 m (110 ft) in the southeast corner of LLNL and about 10 m (33 ft) in the 2

3 northwestern corner. Ground water generally flows to the west, but locally to the south and southwest. Substantial vertical hydraulic gradients exist in some parts of the site. Lateral heterogeneities on the order of 6 m or less have been inferred by some inter-well pumping tests, while other tests have indicated hydraulic communication over hundreds of meters. Vertical hydraulic communication between aquifers can also be complex, usually with very little conductivity between units, but sometimes with very localized connections. This is evidence of the complex fluvial structures consisting of inter bedded paleostream channels at this site. Besides pumping tests, the main tools for site characterization and permeable structure mapping have been borehole lithology logs, and geophysical logs (resistivity, SP, gamma). Due to the extremely complex heterogeneous nature of the permeable units at this site, the Environmental Restoration Division at LLNL simplified the geological interpretation by grouping the permeable layers into common hydro stratigraphic units (HSUs) based on their hydraulic interconnectivity. Fig. 2 is a cross sectional view of the interpreted HSU groups, showing how they all tend to dip west, and how some units intersect the water table and penetrate into the vadose zone. Inter-well spacings of typically 30 m, or more, and hydrogeologic correlation lengths of 6 m or less, implies undersampling of the site, with possibly ambiguous interpretations. Consequently, inter-well applied geophysics was necessary to improve sub-surface resolution and structural interpretation. Surface seismic reflection imaging In 1994 the authors conducted a 2-D surface seismic reflection imaging test at the Livermore Site with 1 ft (0.3 m) geophone spacings and 2 ft (0.6 m) shot point spacings. Short station spacings were chosen to prevent spatial aliasing of coherent surface wave noise. The seismic source was a Betsy gun firing blank 8-gauge shot gun shells in water filled, auger drilled holes. The dominant frequency of p-wave reflections was about 90 Hz. The water table depth at the location was about 10 m. After FK filter removal of the ground roll and air wave, and CMP stack processing, there was no confidence in interpreting any of the apparent reflection horizons above 31 m depth, because the first 30 to 40 ms in the near offset traces were dominated by guided 3

4 surface waves and refracted waves. This meant that a surface seismic technique was unsuitable for imaging the upper half of the first 60 m of subsurface, which was the zone of interest. Based on these D test survey results, giving an indication of required station spacings and expected resolution, quotes were requested for a shallow 3-D surface reflection survey over a 250 m square patch at the LLNL Site. Quotes between $150,000 and $500,000 were returned. These costs were based on station spacings between 1 m to 3 m, and data processing turn-around times of over a month. Insufficient funding was available at that time to carry out a 3-D survey. This demonstrated the relatively high cost of shallow surface reflection imaging in a highly heterogeneous environment. The inability to image reflection horizons above a minimum depth is a common problem for the surface CMP method. This problem is principally due to coherent shot generated surface wave noise at early record times. Surface waves in the form of direct waves, air waves, ground roll, and reverberating refractions usually have far higher amplitudes than reflected waves. One must be very careful when using the common mid point (CMP) stacking technique that these surface waves do not stack-in and masquerade as reflection horizons (Steeples, et al., 1997). A partial solution to this problem is to shorten receiver spacings (as the authors did in the 1994 test survey at LLNL) to prevent spatial aliasing of the coherent surface wave noise, and use velocity filtering to attenuate direct waves, air waves and ground roll. However, this still leaves refracted waves because their moveout characteristics are too similar to those of reflection waves at near offsets. Seismic resolution Another issue of importance, to be considered when using any seismic wavefield imaging method at the LLNL site, is the potential resolution. The desired resolution was based on lateral correlation lengths as low as 6 m, with layer thicknesses of 1 m, or less. Consequently, using the 1/4 wavelength resolution criterion, seismic wavelengths of less than 4 m would be needed to image both the top and bottom of layers. In 1994, the authors conducted a VSP survey at a well near the north-west corner of the LLNL site, where the water table was at about 10 m. Fig. 3(b) is a spectral plot of the direct p- 4

5 waves received by downhole hydrophones in this well (W-452). A useful (less than 25 db down from peak amplitude) bandwidth of nearly 500 Hz was available, with a dominant frequency of about 240 Hz for the direct p-waves. An 8-gauge Betsy Gun source was used. Using a measured sub-water-table velocity of 1850 m/s gave a dominant wavelength of about 6.6 m. In 1996, the authors conducted a VSP survey at a well within the same area as the VSP survey reported here, and found the dominant frequency to be about 105 Hz for p-wave signals below the water table (Fig. 3(c)). The calculated dominant wavelength was about 17 m. The water table depth at that time was 21.5 m. Two years later, the VSP data for this study was collected in the same area, but the water table had dropped to a depth of 25 m. The dominant frequency had dropped to about 60 Hz (Fig. 3(d)), giving an average wavelength of about 31 m. The same type of weight drop (Bison E.W.G.) source was used in both surveys. For comparison, in 1994 the authors conducted a VSP survey at another site (Milligan et al, 1997) where the water table depth was only about 3 m, and obtained a dominant frequency of about 550 Hz (Fig. 3(a)), using a hammer-on-plate source. Table 1 summarizes these results, and Fig. 4 plots these values and a best-fit exponential curve. water table. depth (m): center frequency (Hz): useful bandwidth (Hz): Table 1. Water table depths vs. direct wave frequencies for down-hole hydrophone receivers. Although different sources were used for these VSP surveys, the downhole receivers were hydrophones in all cases. It was clear that the useful bandwidth, and hence resolution, decreased non-linearly with increasing vadose layer thickness. The water table depth of 25 m, prevalent for the VSP survey reported here, resulted in a best possible wavefield imaging resolution of about 8 m, which was not considered good enough for reflection imaging. Instead, the direct wave travel-time picks were inverted to generate interval p-wave velocities. The vertical 5

6 resolution for this method was about 0.5 m near the well, which was equivalent to the hydrophone level spacing. VSP data acquisition A multi-well and multi-offset VSP survey geometry was used to record the data. A suitable area for the survey was selected in the northeast quadrant of the site (Fig. 1). Six wells were available, within 20 m to 30m of each other, and all about 60 m deep. Surface elevation differences were less than 1.5 m at this site. Fig. 5 is a map of the well locations. This area covered part of a major hydrostratigraphic unit (HSU4) at about 40 m depth. Fig. 5 includes the interpreted isopach contours for HSU4 created from borehole logs and pumping tests in Part of the purpose of this VSP survey was to try and improve resolution of the HSU4 boundaries between the wells. Fig. 5 also shows the multiple radial lines at various azimuths away from each well, along which surface shotpoints were located at 2-m intervals, starting at a minimum offset of 3 m. The nominal maximum offset was 33 m, effectively giving 3-D coverage of the survey volume. The surface source was a Bison Instruments EWG weight drop machine. Multiple shot records at each shotpoint were summed (stacked) to increase signal-to-noise. The number of stacks varied between three for the close offsets, and six for the far offsets. The downhole receivers consisted of a 48 level hydrophone string, with 0.5 m spacing, and total active length of 23.5 m. Because a hydrophone transducer must be immersed in fluid to be acoustically coupled, the string had to be deployed below the water table, which resulted in the most shallow receiver depth being 25m. Note that it would have been possible to record hydrophone VSP data above the water table if a packing device had been used to block off the screened (perforated) section of casing, and if the well had been filled to the top with water. However, a packing device was not available at the time. The 48 level hydrophone string is an inexpensive multi-level borehole tool when compared with a single level downhole geophone clamping tool. Deployment is quick and easy, with less chance of a downhole mishap due to clamping device failure, and 48 channels were recorded 6

7 simultaneously without having to repeat the same shot points. This reduced acquisition time considerably, and typically over 8000 data traces were recorded per day. Tube wave attenuation Tube wave noise is the main problem with downhole hydrophones as compared with downhole geophones. Tube waves are generated in the borehole fluid when incident body waves encounter elastic property changes in the rocks surrounding the borehole. In addition to these tubewave sources, cross-sectional area changes in the borehole fluid column also act as tube wave source locations (Hardage, 1981, 1983). The most notable of these are the free surface at the top of the fluid column, and the stiff borehole base. Other cross-sectional area changes can occur elsewhere along the borehole (like caliper changes, wash-outs, or insertion of a downhole tool). Once body wave seismic energy becomes coupled with the fluid at a tube wave source location, then two guided (Stoneley) waves are created, both upgoing and downgoing (Hardage, 1981, 1983), consisting of circular fluid motion in the radial and axial directions. Hydrophones are most sensitive to the radial component, whereas a clamped vertical geophone is most sensitive to the axial component. Clamping of the geophone to the well casing helps isolate it from the axial component. Two steps were required to attenuate the coherent tube wave noise from the data. The first step was a modification of the hydrophone string, and the second step was in the data processing. To help attenuate and slow down tube wave energy along the hydrophone string, a system of baffles was developed and deployed between each hydrophone element (Milligan et al., 1997). It was observed that the baffles attenuated the radial tube wave component by about a factor of ten, and because these baffles represent a partially saturated effective media, the tube wave velocity slowed down by a factor of about five. Both attenuation and velocity reduction of the tube waves enabled them to be wavefield separated in the data processing stages. The data processing step for tube wave separation consisted of zero-phase bandpass filtering ( Hz trapezoid window), trace amplitude equalization, and FK filtering to 7

8 subtract wavefields with velocities less than 600 m/s. This was followed by another trace equalization to restore amplitudes of events previously dominated by the subtracted tube waves. Fig. 6(a) shows a raw common shot trace gather from well W-1250 at azimuth 304 and offset 15 m. Fig. 6(b) shows the same traces but after tube wave removal. Tube wave attenuation helped considerably in making more reliable and accurate first break (FB) time picks. VSP Data 2-D Velocity Inversion Interval velocity sections were computed up to 20 m away from each well, with both lateral and vertical resolution almost as good as the receiver spacing (0.5 m) close to the well, but with diminishing resolution laterally away from the well. The best obtainable lateral resolution away from the well was limited by the shot point spacing (2.0 m), and the increasing Fresnel zone size with offset and path length. This increasing Fresnel zone size also diminishes the vertical resolution away from the well. It was estimated that vertical resolution dropped to about 2 m, and lateral resolution dropped to about 4 m at the far offsets. Making FB time picks on each trace was a time consuming task, requiring sub millisecond accuracy on the 17 ms period direct waves to avoid wild velocity jumps during the inversion process. Without attenuation of the tubewave energy with the FK filter, FB time picking would have been very inaccurate. The FK filter also smoothed the trace-to-trace direct wavefield, and so some depth resolution was lost. After tube wave subtraction, Fig. 6(b) shows FB time picks along the negative "trough" of the direct wave arrivals. It can be seen from Fig. 6(b) that there is an obvious stair-step like increase in FB arrival times below 37 m depths, indicating a low velocity layer at this depth. Inversion of the FB time picks was achieved by combining 2-D raytracing and layer stripping, as shown in Fig D interval velocity sections were computed corresponding to the shot point (SP) data gathers along each radial arm. Starting with the closest offset SP data gather and the top layer, an interval velocity was computed for each layer, honoring Snell s law for ray bending at each layer boundary, to match the FB time. Once the interval velocity layers for the closest SP had been computed, FB data from the next offset SP gather was used. The 2-D ray bending points were logged, enabling the mapping of the calculated interval velocities into a 8

9 vertical 2-D section. This process was repeated for all SP offsets, resulting in a section of irregularly spaced velocity values. This section of interval velocity points was then interpolated onto a regularly spaced (0.5 m by 0.5 m) grid to produce the final section. Note that lateral coverage away from the well is increasingly limited with depth, as dictated by the direct wave coverage. Two prior pieces of information were required to initiate the velocity inversion: the water table depth (measured) and a best guess value for the first layer velocity. The inversion program would then start the computation by calculating a vadose layer velocity to match the FB time at the topmost receiver. Varying the first layer velocity had little effect on calculated lower layer velocities, however, if it was too different from the calculated second layer velocity, then raytracing would sometimes fail by critical ray refraction. A first layer velocity matching the average background velocity at that level was usually a good choice. Note that allowing the inversion process to calculate the vadose layer velocity removed any small static shifts and/or SP elevation changes relative to the water table datum. Also note that full resolution in the interval velocity image could only be attained below the highest level receiver, i.e., there was no vertical resolution in the vadose zone, or within the first layer above the top receiver. Often several picking and inversion iterations were performed to arrive at a geologically reasonable model, usually due to FB pick inaccuracies. It was estimated that picking errors resulted in velocity inversion inaccuracies of up to ± 20%, which was sufficient for distinguishing between saturated sand (2100 m/s) and mud and clay (1600 m/s) layers. Inversion Results Of the more than 20 p-wave interval velocity sections generated by the inversion method, only a small subset is included here to show some general features. Fig. 8 shows three of these sections. The most notable feature in all velocity sections are one or more low-velocity zones below the water table. Velocities as low as 500 m/s were encountered in some zones below the water table, and partial saturation of pore spaces would be the best way to explain p-wave velocities this low (Mavko and Mukerji, 1995). The velocity sections do not extend to the well casings at zero offset, because of the hydrophone s insensitivity to vertical waves. Hydrophones exhibit a 9

10 vertical dipole-like antenna pattern when employed in a vertical borehole because of their maximum sensitivity to horizontal waves (Rector and Lazaratos, 1994). However, vertically orientated geophones in a vertical borehole would exhibit maximum sensitivity in the vertical direction. All the 2-D velocity sections were combined into a 3-D data set. Fig. 9 shows one view of the 3-D data set projected into the survey volume. Viewing of this projection from different angles and at different magnifications helped to correlate between low-velocity zones and the existing hydrogeology interpretations. The first 22 m of vadose zone coverage was removed to facilitate viewing features below the water table, and so the upper low velocity zone (in red) that is common to all velocity sections in Fig. 9 is the lower part of the vadose zone. Note that projection of 2-D velocity sections into 3-D space does not imply that a full 3-D inversion has been computed, i.e., some of the direct wave paths that were used in the inversion may have been from out of the vertical plane into which they were projected. A full 3-D inversion, using 3-D raytracing, would be required to solve for interval velocities at the correct 3-D position. Interpretation There were two predominant low velocity zones (less than 1200 m/s) unevenly distributed throughout the survey volume below the water table (25 m depth). The lower zone appeared as a layer less than a meter thick in most places, and it correlated to a silty-sand unit in the well logs; this lower layer varied in depth between 37 m and 40 m in HSU3b, and usually appeared to be perched on top HSU4, a main aquifer consisting of fully saturated sand-gravel, with velocities greater than 1900 m/s. Pinch-outs of this lower low-velocity layer appeared to correlate with both the existing isopach interpretation of HSU4 (from well logs), and with pinch-outs of the higher velocity aquifer underneath. This was not always the case, as indicated in Fig. 8(b), which shows the apparent merging of two low velocity zones, without any high velocity layer underneath. The upper low velocity layer had a higher variability in thickness, sometimes appearing to almost merge with the vadose zone (see Fig. 8(c)), and it was perched on top of another aquifer unit (in HSU3a). The aquifers in both HSU4 and HSU3a were being pumped for water treatment from several nearby wells, with a combined pump rate of over 100 gallons per minute. 10

11 The perching of these low-velocity layers on top of aquifers from which water was being pumped was a major discovery. An explanation for the decrease in p-wave velocity (down to 500 m/s) is a partial saturation of the pore spaces, i.e.: gas must be present. Small micro bubbles of gas were noticeable in water samples taken from the aquifer in HSU4, i.e., the water samples "fizzed" when reduced to atmospheric pressure. Gas analysis of these water samples indicated that atmospheric air, aged by about 13 years (by the tritium half-life method), was the probable gas source. The typical age of gas samples collected from outside the zone of pumping influence was 24 years. It was also found that the gas was present in super-saturated concentrations that would require at least 2 atmospheres (above 1 atmosphere) of pressure to keep it in solution, i.e., to prevent bubble formation. Super-saturation by gas is not uncommon in aquifers, and sometimes up to 5 atmospheres of pressure (above 1 atmosphere) is required to prevent dissolution. A possible explanation for this super-saturation state would be the rapid sinking of gas-water mixtures at re-charge zones. It is probable that high pumping rates from these two aquifer units (HSU3a and HSU4) have caused significant pressure drops in the hydraulically conductive sand-gravel layers near the pump wells. The natural water table depth in the VSP survey area was about 19 m, before pumping began several years ago. But at the time of this survey the water table depth had dropped to 25 m. Considering the pressure head in the HSU4 aquifer at 40 m depth, these water table measurements gave pressures of about 2.1 atmospheres before pumping, and about 1.5 atmospheres at the time of the survey, i.e., low enough to allow gas dissolution and bubble formation. However (as disclosed earlier), the low velocity zones did not appear in the hydraulically conductive parts of the aquifer, but perched on top in silty-sand units. There are two possible hydrological models to explain this. The first would be the formation of micro-bubbles in the hydraulically conductive parts of aquifers, by water that has entered the depressed water table zone, and these bubbles have floated to the top, passed into the silty-sand layer above, and have accumulated there in high enough concentrations to become seismicly detectable. The second would be depressurization of water in the silty-sand layer above, and accumulation of bubbles due to the relatively low hydraulic conductivity of silty-sand compared to the sand-gravel aquifer 11

12 below. But it was beyond the scope of this study to investigate the validity of either model, as both would probably require comprehensive finite-difference or finite-element hydrological modelling. Conclusions The multi-well, multi-offset VSP method yielded interval velocity data between wells in areas that log information could not cover. Ground truth connections could be made between well logs and the mapped velocity images, increasing confidence for interpreting the surrounding sedimentary structures. The resolution obtained in the velocity images was better than the estimated average correlation length (6 m) of the heterogeneous sedimentary structures, and this was mainly due to the close (0.5 m) downhole hydrophone spacing, and close (2 m) surface shot point spacing. The pre-existence of many wells at the LLNL Livermore Site made the VSP method a suitable choice for geophysical site characterization. The VSP method also had the additional advantage of being able to image structures up to more shallow depths than a surface seismic reflection method could. The VSP method used was cost effective in both acquisition time and the amount of equipment used. Data processing was minimal, and turn around time for velocity inversion was short when compared to the time required for surface seismic reflection data processing. It appeared that several low velocity layers (less than 1200 m/s) could be directly linked to aquifer pump extraction, and this opened up a number of possible uses for mapping these partially saturated layers: (1) as tracers to map out the aquifer boundaries; (2) to give the areal extent of which parts of an aquifer are being influenced by pumping activity; (3) monitoring this area over time to see if it is expanding or contracting; (4) saving money by not having to drill another well and install pumping equipment, if an area containing an aquifer already shows a large low velocity layer above it to indicate adequate pump flow rates into existing pump wells. 12

13 Acknowledgments The authors are grateful for use of equipment, support, data and maps from the Environmental Restoration Division, LLNL. The authors wish to thank Landmark Graphics for providing ProMAX seismic data processing software under a strategic partnership agreement. The authors also wish to thank the Geometry Center, at The University of Minnesota ( for developing and writing the open software source code for Geomview, the 3-D graphics display engine used to project the combined data attributes. This work was supported by the DOE EMSP program and was initiated using Lawrence Livermore LDRD funding. References Hardage, Bob, A., 1981, An examination of tube wave noise in VSP data: Geophysics, 46, p Hardage, Bob, A., 1983, Vertical Seismic Profiling, part A, principles: Geophysical Press, 14A, p Mavko, G., and Mukerji, T., 1995, Seismic pore space compressibility and Gassmann s relation: Geophysics, 60, No. 6, p Milligan, P. A., Rector, J. W., and Bainer, R., 1997, Hydrophone VSP imaging at a shallow site: Geophysics, 62, No. 3, p Rector, J. W., and Lazaratos, S. K., 1994, Field characterization of the seismic radiation pattern of a fluid-coupled piezoelectric source and hydrophone receivers, Presented at the 64th Ann. Mtg. of SEG, Los Angeles, Expanded Abstracts, p Steeples, D. W., Green, A. G., McEvilly, T. V., Miller, R. D., Doll, W. E., Rector, J. W., 1997, A workshop examination of shallow seismic reflection surveying: The Leading Edge, 16, No. 11, SEG, p

14 Figure Captions Figure 1. Location of the VSP survey site at the Lawrence Livermore National Laboratory (LLNL) site, east of San Francisco Bay, in Northern California. The inset view of the LLNL site shows contoured VOC plume distributions in The VSP survey area is shown in the north east. Natural ground water flow is generally westward. Figure 2. HSU cross-section, from west to east, along the southern edge of the LLNL Livermore site (LLNL Environmental Restoration Div., 1995). The approximate projection of the VSP study area is shown. A major aquifer contained in HSU 4 was a primary target for the VSP survey, and it was at a depth of about 40 m (131 ft) at the survey location, about 960 m north of the displayed cross section. Figure 3. Direct wave spectra received by down-hole hydrophones at various VSP survey locations with different vadose layer thicknesses: (a) 3.3 m. (b) 10.2 m. (c) 21.5 m. (d) 25.0 m. Surface sources were: (a) hammer-on-plate. (b) Betsy Gun. (c) weight-drop machine (EWG). (d) weight-drop machine (EWG). Figure 4. Graph of downhole hydrophone direct wave dominant frequencies vs. water table depth (vadose layer thickness). Hydrophones located below the water table. An empirical best-fit exponential function was made. Figure 5. Map view of the sub-site area chosen for this VSP survey. VSP acquisition wells were W-1250 through W-1255 inclusive, while W-1306 and W-1307 were active pumping extraction wells. Radial arms surrounding well heads indicate surface shotpoint location lines. An isopach overlay shows a pre-existing (1995) interpreted thickness (in meters) of a major hydrostratigraphic unit (HSU4) at about 40-m depth. All wells were 4.5 inch fiberglass cased, with screened openings to HSU4, except for W-1306 which was pumping from HSU2 at a depth of 27 m. 14

15 Figure 6. Example of a raw VSP shot record (a), and same record after tubewave removal (b). Note four obvious tubewaves in (a): downgoing from the top of the hydrophone tool (TW1), up and downgoing from the screened interval in the well casing between depths of 39.6 m and 41.1 m (TW2 & TW3 respectively), and upgoing from the tail of the hydrophone tool (TW4). Many other tubewaves exist, which destroy the trace-to-trace coherency of the direct wave, but are too close in phase with the direct wave for discrete identification. The baffles have slowed down tube wave velocity from 950 m/s to 210 m/s, making their removal by velocity filtering relatively easy. After tubewave removal (b) the dominant waveforms are now the first break (FB) direct wave arrivals, facilitating FB time picking. FB picks are shown as circles following the first major negative peak of the direct wave. Figure 7. Raypath schematic showing the concept of layer-stripping and 2D raytracing, and how these results can be combined to produce a 2D interval velocity section. For clarity, only 2 shot points and the first 5 layers are shown. The actual process used in this study typically consisted of up to 16 SPs, for a total offset of 33 m, and 49 layers. Layer 0 (top) is the vadose (unsaturated) zone (about 25 m thick). Layer 1 is the gap between the base of water table and the first hydrophone (sometimes zero), and all other layers are at 0.5 m spacings (hydrophone spacing). Only a 12-level hydrophone tool is used in this example. The tool we used had 48 receiver levels. Figure 8. Contoured p-wave velocity sections from the 2D inversion process. There were two predominant low velocity zones unevenly distributed throughout the survey volume below the water table at 25 m depth; the lower one has a fairly constant thickness of about 1 m, and correlates to a silty-sand unit in the well logs. This lower low-velocity layer varies in depth between 38 m and 40 m, and appears to pinch out in certain places. The upper low-velocity zone has variable thickness, and appears to join with the lower one in (b), or merges with the vadose zone (c). The top hydrophone level for plots (a) and (b) was at 30 m depth, while for plot (c) it was at 25 m depth. 15

16 Figure 9. A 3D graphic projection of the VSP velocity sections together with wells, HSU layer intersections at each well (color coded starting at the top with: brown, green, blue, purple, and maroon, for HSUs 2, 3a, 3b, 4, 5, and 6, respectively), and a pre-existing interpreted isopach map of HSU4 (Environmental Restoration Div., LLNL, 1995). The isopach data are shown as variable thickness (0.1 m to 3.3 m) blocks. Screened well casing openings are shown as dark grey collars. NB: The topmost low velocity layer is the base of the vadose zone. Total depth of each well is about 63 m (207 ft.). 16

17 VSP survey area Figure 1. 17

18 VSP water table Figure 2. 18

19 frequency (Hz) (a) frequency (Hz) (b) amplitude (db) amplitude (db) frequency (Hz) frequency (Hz) (c) (d) amplitude (db) amplitude (db) Figure 3. 19

20 Figure 4. 20

21 pump well pump well W1306 Figure 5. 21

22 hydrophone depth (m) hydrophone depth (m) oo oooo ooooo oooo o o ooooooooooooooooo oooo o oo oo ooooo TW1 time (ms) TW2 TW3 TW4 time (ms) (a) (b) Figure 6. 22

23 Figure 7. 23

24 (a) (b) (c) Figure 8. 24

25 grid apacing: 5m x 5m velocities (m/s): Figure 9. 25