Three-dimensional seismic-reflection imaging of a

0 downloads 0 Views 4MB Size Report
filled with a mix of sands, clays, and gravels of different com- paction and ... Consequently, even small amounts of DNAPLs can present a long-term source of ..... To maintain a more uniform CMP fold and offset distribution, flex binning was .... cube of stacked traces needed for migration, dead traces were filled in, padding ...
Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

GEOPHYSICS, VOL. 73, NO. 5 共SEPTEMBER-OCTOBER 2008兲; P. B85–B98, 15 FIGS., 2 TABLES. 10.1190/1.2969775

Case History Three-dimensional seismic-reflection imaging of a shallow buried paleochannel

Gian Luigi Fradelizio1, Alan Levander2, and Colin A. Zelt2

Upon release, DNAPLs sink until their neutral buoyancy levels are attained and capillary trapping prevents further movement or until an impermeable layer is reached, which can lead to pooling and lateral spreading. DNAPLs typically are characterized by low solubility and are resistant to biodegradation and natural attenuation 共Pankow and Cherry, 1996兲. They are, however, soluble enough to cause high levels of contamination over time. The U. S. Environmental Protection Agency’s 共EPA兲 maximum contaminant levels for common DNAPLs are measured in parts per billion. Consequently, even small amounts of DNAPLs can present a long-term source of groundwater pollution. In natural heterogeneous groundwater systems, delineation and detection of contaminants can be extremely difficult. As a result, several technologies for characterizing contaminated sites have been developed. Boreholes, trial pitting, and cone penetration tests are common investigation techniques at contamination sites. However, they have inherent lateral limitations and the potential to spread contaminants by opening migration pathways through low-permeability zones. In addition to these intrusive techniques, seismic methods now are beginning to be applied to groundwater-contamination problems because they are noninvasive and can provide spatially continuous information regarding shallow subsurface structures 共Steeples, 2001兲. Although generally it is not possible to detect pollutants directly in the subsurface, seismic techniques can be used to characterize and differentiate geologic units that can harbor a contaminant or control its movement. Seismic methods have applicability in a wide range of shallow environments, particularly in areas where the target depth, soil, or hydrologic conditions preclude effective use of ground-penetrating radar 共GPR兲 共Young and Sun, 1996兲. In 2000, we acquired several P-wave seismic data sets at Operable Unit 2 共OU-2兲, Hill Air Force Base 共AFB兲, near Ogden, Utah. The purpose of the seismic surveys was to image the shallow subsurface at the site and to develop high-resolution seismic methods for identifying structural or stratigraphic traps for contaminants 共Gao et al.,

ABSTRACT A suite of reflection seismology investigations of the shallow subsurface was conducted at Hill Air Force Base, Ogden, Utah, at a groundwater contamination site with very large subsurface mechanical heterogeneity. The investigations were designed to expand and improve understanding of the subsurface environment developed from well data and previous seismic investigations. The goal of the investigations was to image the sides and bottom of a paleochannel eroded in a clay layer 10–15 m below the surface. The paleochannel is filled with a mix of sands, clays, and gravels of different compaction and water saturation. The paleochannel acts as a contaminant trap for dense nonaqueous-phase liquids 共DNAPLs兲. Therefore, a detailed map of its geometry, lateral boundaries, and depth is crucial to remediation efforts. The results of processing a 3D reflection data set over the channel showed that it is possible to image and delineate the geometry and depth of a very shallow structure. Seismic results agreed well with the available well data in the deepest part of the buried paleochannel. The map of the channel derived from seismic data replaces the need for many wells.

INTRODUCTION Accurate characterization of groundwater-contamination sites is crucial for the successful design of effective, cost-efficient remediation strategies, especially when the contaminants are dense nonaqueous-phase liquids 共DNAPLs兲. DNAPL groundwater contamination is a serious environmental problem. Widely used for many decades as industrial solvents, DNAPLs have been introduced into the ground by spills and leaking storage and disposal facilities.

Manuscript received by the Editor 25 March 2007; revised manuscript received 15 October 2007; published online 12 September 2008. 1 Formerly Rice University, Department of Earth Science, Houston, Texas, U.S.A.; presently Chevron Energy Technology Company, Geophysical Specialty Services Unit, Houston, Texas, U.S.A. E-mail: [email protected]. 2 Rice University, Department of Earth Science, Houston, Texas, U.S.A. E-mail: [email protected]; [email protected]. © 2008 Society of Exploration Geophysicists. All rights reserved.

B85

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B86

Fradelizio et al.

2006, 2007; Zelt et al., 2006兲. In this paper, we present the results of processing and depth migration of the 3D compressional-wave seismic-reflection data specifically acquired to image a very shallow buried paleochannel to aid groundwater remediation efforts at the site. From the seismic point of view, what makes this site simultaneously interesting and challenging is the extreme shallowness of the targets 共⬍16 m兲 and the extreme heterogeneity of the shallow sediments. As previous seismic studies at this site have shown, the variation in compressional velocity is about one order of magnitude vertically and more than a factor of two laterally at any depth 共Gao et al., 2006, 2007; Zelt et al., 2006兲. This variation produces a complicated seismic response. Compressional seismic velocity seems to be controlled by the water-saturation profile rather than by a sharp boundary as defined by the phreatic surface, and the surficial sediments have different compaction and water-saturation characteristics. In addition, the very top of the alluvium is desert caliche, a hardened deposit of calcium carbonate that cements materials together, including gravel, sand, clay, and silt. This extra-hard ground attenuates seismic waves. In addition, it makes it difficult to couple source and receivers with the earth. The availability of hundreds of borehole logs and cone penetrometer tests 共CPTs兲 offers the unique opportunity to evaluate the results of the seismic processing, compare and corroborate the validity of the approach, and highlight the limitations of the method. Although no sonic log is available for this site, DNAPL is found in wells on the bottom of the channel. However, no enhanced reflectivity is expected from the contaminant. Additionally, we can compare the reflection result with results from 3D traveltime tomography and 2D waveform inversion performed at the site.

Figure 1. Map of the study area in northern Utah. Chlorinated solvents disposed of in unlined trenches at Hill AFB, OU-2, migrated into a shallow aquifer. A 90-m escarpment leads from Hill AFB to the Weber River Valley, where groundwater contamination was detected.

SITE BACKGROUND Hill AFB is located in northern Utah about 30 miles north of Salt Lake City and west of the Wasatch Front mountain range 共Figure 1兲. Aircraft maintenance and repair activities have generated waste, including chlorinated and nonchlorinated solvents and degreasers, fuels, and other hydrocarbons, historically disposed of in open pits and landfills on the base. The EPA placed Hill AFB on the Superfund National Priorities List in 1987. Presently, there are 11 contaminated geographic areas, known as Operable Units, at Hill AFB. They encompass one or more hazardous-waste sites, and many of them extend off base into the surrounding communities. One of these units, OU-2, contained at least two unlined disposal trenches a few meters wide and approximately 20–30 m long, underlain by an alluvial aquifer. These trenches were used from 1967 to 1975 for the disposal of unknown quantities of chlorinated organic solvents, primarily trichloroethene 共TCE兲, from degreasing operations. The estimated volume of waste solvents disposed in the trenches exceeds 40,000 liters and might exceed 400,000 liters. Their composition is approximately 70% TCE 共EPA, 2003兲. The waste solvents migrated downward through the unsaturated zone above the water table into the shallow aquifer. This unconfined aquifer is composed of sediments of the Provo Alluvium, consisting of silt, sand, and gravel in a cut-and-fill stratigraphy with little lateral continuity. The aquifer is contained in a buried paleochannel eroded into the underlying Alpine Formation, a thick, silty clay unit with sporadic patches of fine sand and silt. Because of the very low hydraulic conductivity 共⬃108 cm/s兲, the Alpine clay acts as an aquiclude.Alarge volume of DNAPL remains in the subsurface, predominantly as an immobile or “residual” phase, retained as ganglia by capillary forces in the aquifer’s pore spaces, and also as a mobile phase pooled in topographic lows on the surface of the clay aquiclude. As part of the site characterization, a set of geotechnical investigations were performed at the contamination site. The primary focus of these activities was to delineate the buried paleochannel. Because of the limited areal extent and saturated thickness of the aquifer and the density of the DNAPL, the location and topography of the buried channel were considered critically important in determining where to locate remediation wells. The morphology and depth of the top of the clay layer determined from soil-boring logs, gamma-ray logs, CPT logs, and monitoring wells shows a buried channel approximately 10–15 m deep, trending north-south 共Figure 2兲. The base of the channel contains deeper ponds that are likely sites of DNAPL accumulation. Despite drilling about 140 wells in an area of about 50⫻ 100 m, the site remains characterized inadequately for proper remediation. Contaminated soils and pooled DNAPL are the sources of shallow groundwater contamination, which extends approximately 500 m downgradient and beyond the Hill AFB boundary. A steep, terraced, northeast-facing escarpment leading from the base to the Weber River Valley below separates the on-base portion of OU-2 from the off-base portion. There is about 90 m of relief between Hill AFB and the valley. Downgradient seeps and springs have also exhibited contamination 共EPA, 2003兲.

3D SEISMIC ACQUISITION When the seismic data acquisition parameters 共Table 1兲 were designed, previous on-site experience gained from a pilot 2D experiment in 1998 共Dana et al., 1999兲, equipment availability, and logistic

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D imaging of a buried channel limitations were taken into account. The seismic survey measures approximately 92⫻ 36 m. Its shape is rectangular, except for the southernmost part, where it becomes a parallelogram. This change was necessary to adjust for obstacles and to center the survey on the location of the paleochannel as determined from well data. Seismic data were acquired along 45 east-west receiver lines. The lines were 36.5 m long and were spaced 2.1 m apart 共Figure 2兲. The field layout is a nonorthogonal brick pattern with source lines at a 45° angle from the receiver lines. During acquisition, six receiver lines were active. Source-line spacing was 2.1 m and repeated every other patch. Receiver and source increments were 0.35 m in the inline direction 共Figure 3a兲. The source-point interval was stretched out because of the oblique source-line orientation, so midpoints fell into the center of the bins, using a natural bin size of half the receiver increment. The slanted acquisition geometry had a maximum source-receiver offset of about 37.0 m and nominal fold of 52 traces in 0.175-m square bins. Because of the limited number of active receiver lines, the range of azimuths was rather limited and focused in

B87

the inline direction 共Figure 3b兲. The narrow-azimuth recording patch led to an overall offset distribution skewed toward short offsets 共Figure 3c兲. Seismic data were acquired using 630 REFTEK 125A singlechannel seismic recorders with 40-Hz geophones. Because of the extra-hard soil, most of the geophones were planted in predrilled holes. The seismic source was a 0.223-caliber rifle fired into predrilled holes 20 cm deep. The recording time was 200 ms, and the sample interval 1.0 ms. Each of the six active receiver lines was assigned 104 channels for a total of 624 channels per shot gather. Aircraft activities and remediation activities, including pumping and well injection, personnel movement, and vehicular traffic, generated coherent and random noise at the site.

DATA PROCESSING To optimize the final stacked volume, the processing flow 共Table 2兲 was tailored to the particular problems of the data — specifically, strong ground roll and low signal-to-noise ratio 共S/N兲. Coherent noise not effectively removed from the data adds constructively during the stacking procedure and generates artificial, reflectionlike events on the stacked sections. A conservative approach to processing was adopted to eliminate coherent noise events, even when leading to a final image of poorer quality, to avoid pitfalls arising from stacking coherent noise 共Steeples and Miller, 1998兲.

Trace editing Consistently bad channels or traces that were particularly noisy were zeroed. In addition, direct and refracted waves were muted

Table 1. Acquisition parameters for the 3D seismic survey.

Figure 2. Clay depth surface as defined by kriging on logged value 共soil boring logs, gamma-ray logs, CPT logs兲 from monitoring wells 共white circles兲; the datum is 1431 m above sea level. Overlain is the acquisition layout of the 3D seismic survey performed in 2000 by Rice University. Black and magenta dots represent the actual position of the receivers and sources, respectively. Gaps and shifts in the source/receiver position result from bad shots and obstacles on the ground.

Parameter

Measurement

Receiver-line direction Number of receiver lines Receiver-line interval Number of channels/line Channel spacing Number of receiver lines/shot Source-line direction Source-line interval Inline source spacing Crossline source spacing Actual number of source points Actual number of traces Record length Sample rate Frequency of geophones

East-west 45 2.1 m ⬃104 0.35 m 6 Northeast-southwest 2.1 m 0.35 m 0.35 m 3126 1,816,958 200 ms 1 ms 40 Hz

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B88

Fradelizio et al.

where distinct from reflections. A mute ramp of 5 ms was used to smooth the transition from live to muted data samples and to prevent undesirable side effects in the later application of filters. Traces to be killed and the top mute time gate were selected on equalized, gained, and band-pass-filtered shot gathers as the first step in the processing sequence. Amplitude variation between traces in common-source gathers was reduced by trace balancing, which was computed on a mean basis in a 200-ms window. Because preserving true amplitudes was not a priority in our study, trace amplitudes were corrected for waveform spreading, inelastic attenuation, and transmission losses using automatic gain control 共AGC兲, computed in a sliding 75-ms time gate. The gained shot gathers 共Figure 4a兲 show that the most prominent feature is the high-amplitude ground roll, which in vertical-component P-wave seismic data is composed of Rayleigh waves.

Band-pass filtering Ground roll typically has a higher amplitude and lower dominant frequency than near-surface reflections. Therefore, the power spectrum of the raw data is not a very effective tool for choosing the frequency filter parameters because it is dominated by the surface-wave response 共Figure 5a兲. The gained shot gathers were examined using a wide range of filter panels to determine the frequency content of the signal up to the Nyquist frequency of 500 Hz. A minimum-phase Butterworth band-pass filter from 160 to 400 Hz 共12 dB/octave roll off兲 best attenuated noise and brought out reflections.

Figure 3. 共a兲 The acquisition patch is a 45° slanted pattern, with receiver and source lines at a distance of 2.1 m. Source lines repeat every other patch. Source/receiver spacing is 0.35 m. Midpoints fall into the center of the bins, using natural 0.175-m square bins 共black grid兲. The acquisition geometry leads to 共b兲 a narrow, focused distribution in the inline 共east-west兲 range of source-receiver azimuths and to 共c兲 an even 共although skewed toward short offsets兲 overall distribution of the source-receiver offsets.

Band-pass-filtered shot gathers showed that surface waves had been attenuated by the frequency filter, singling out the main reflection 共Figure 4b兲. However, ground roll still was present in the seismic record after band-pass filtering. In addition, coherent phases not removed by frequency filtering because of overlapping frequency content were apparent on the seismic record, as revealed on both the seismic record and the equivalent frequency-wavenumber 共f-k兲 spectrum of the shot gathers 共Figure 6兲. Additional processing was required to remove or attenuate coherent noise from the seismic record.

Attenuation of coherent noise Several methods to remove or attenuate ground roll were evaluated. Muting, which requires picking a time window on each shot gather 共Baker et al., 1998兲, was detrimental to the coherent reflection signals because of the overlap of the two phases. Spatial filters, such as f-k filtering and trace mixing, increased the coherency of the reflections but also generated strong artifacts and enhanced aliased phases in the data. More sophisticated techniques, such as ␶ -p transformation 共Spitzer et al., 2001兲 and wave-equation datuming 共Larkin and Levander, 1996兲, require considerable manpower for correct application and are very sensitive to aliased phases in the data. Therefore, they were not considered. As an alternative, coherent noise attenuation 共Hornbostel, 1991兲 proved to be effective in attenuating the residual ground roll from common-source gathers without introducing artifacts in the data or deteriorating the reflection signal. Coherent noise attenuation ap-

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D imaging of a buried channel plies a spatial filter in the frequency-offset domain to attenuate linear coherent noise in common-source ensembles, given a frequency-velocity pair. After testing, values of 10–400 Hz for frequency and 160–240 m/s for velocity were used. To remove the airwave still present on some of the records after top muting and band-pass filtering, air-blast attenuation was applied. Air-blast attenuation uses a pilot velocity and an approximate energy envelope width to seek out anomalous energy on a trace-by-trace basis. Anomalously strong energy, when found, is attenuated to an amplitude level similar to that of the surrounding data. Values of 331 m/s for the airwave velocity and 10 ms for the energy envelope width were used. Noise resulting from ongoing remediation activities during the survey was linearly coherent and was generated by pumping equipment at the site. These noise trains were well behaved because of the relatively low amplitude, narrow bandwidth, and nondispersive constant velocity. Furthermore, this kind of noise was present only locally and on a limited number of shot gathers. Remediation noise was removed from the seismic record using coherent noise attenuation with parameters adjusted case by case. The successful attenuation of ground roll, remediation noise, and airwave has the dual effect of enhancing the continuity of the main reflection event共s兲 in the shot gathers and removing nonreflection coherent events 共Figure 4c兲. The power spectrum of a sample common-source gather after band-pass filtering and coherent noise attenuation shows a usable bandwidth from 100 to 300 Hz, although the spectrum is unbalanced and has a dominant frequency of about 180 Hz 共Figure 5b兲.

Deconvolution To remove unwanted reverberations from the seismic record, predictive deconvolution was applied to the data. A combination of autocorrelation, deconvolved data, and comparison of the spectra of undeconvolved and deconvolved records were used to determine the deconvolution parameters and test the effectiveness of the procedure. The final parameters were an 8-ms prediction lag, 40-ms filter length, and 0–160-ms time gate for the filter design. Furthermore, a surface-consistent scheme 共Levin, 1989兲 was used because it appeared to lessen the detrimental effect of noise and nonstationarity on the deconvolution. Particular attention was placed on checking the performance of deconvolution across the entire data set and avoiding negative effects on the coherent signal, such as removal of reflection energy or creation of spurious spikes. The deconvolved data appear sharper, and the similarity of the wavelets improved 共Figure 4d兲. In addition, the power spectrum of the deconvolved data is balanced and broadened, especially toward the higher frequencies 共Figure 5c兲. Because of the particular nature of shallow-reflection seismic data, we were concerned about false coherent events created by processing. However, in our case, predictive deconvolution had a minimal effect on the final image, mainly reducing the amplitude of the reverberations associated with the strongest reflection. Moreover, data were evaluated carefully after each processing step to ensure that no coherent event was created during processing.

B89

Static corrections Elevation static corrections were used to compensate for temporal shifts caused by deviations from the datum plane of the source and receivers during the recording process. Topography at the site has a maximum elevation change of 3.0 m, although in most parts of the survey area the differences are much smaller. However, considering the scale of the seismic experiment and the low velocity of the nearsurface layer, such elevation differences are not trivial. The final datum was placed at the highest elevation in the survey 共1431 m above sea level兲 so as not to shorten the arrival time of the already extremely shallow reflections. A variable replacement velocity ranging from 200 to 700 m/s was used. Near-surface velocity was extrapolated from the available waveform tomography velocity model 共Gao et al., 2006兲. The maximum resulting static shift was ⬃5 ms. In an attempt to correct for small trace-to-trace variations in arrival times resulting from very localized velocity variations in the common-midpoint 共CMP兲 gathers related to heterogeneity of the medium, surface-consistent residual static corrections 共autostatics兲 were calculated for the NMO-corrected CMP along time gates centered on the main reflection. The temporal width of such gates always was kept sufficiently large 共⬎20 ms兲. Moreover, the maximum allowed

Table 2. Processing flow for the 3D seismic data set. Process element Trace killing Trace muting Trace equalization AGC Bandpass filter Ground-roll removal Airwave removal Deconvolution Elevation statics

Velocity analysis CMP sorting NMO correction Stack Deconvolution Band pass filter Trace equalization Depth migration

Type/parameter

Measurement

Top mute Time gate Frequency Slope Frequency range Velocity range Air-blast attenuation Operator length Prediction lag Velocity Final datum CVS range Flexi-bin size Stretch mute 3D Operator length Prediction lag Frequency Slope Explicit finite-difference Maximum dip

75 ms 160–400 Hz 12 dB/oct 10–400 Hz 160–240 m/s 40 ms 8 ms Variable 1431 m above sea level 200–1000 m/s 0.20⫻ 0.20 m 10% 36 ms 6 ms 100–350 Hz 12 dB/oct

50°

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B90 Figure 4. Processing sequence applied to the seismic data. 共a兲 After trace killing, top mute, trace equalization, and amplitude recovery, surface waves are dominant in the receiver lines of this shot gather. 共b兲 Ground roll is attenuated and a coherent reflection becomes visible after band-pass filtering. 共c兲 Subsequent attenuation of the coherent noise is necessary to remove any unwanted phase from the seismic record. 共d兲 The final processing flow, including predictive deconvolution, successfully minimizes the noise content and increases the S/N of the data.

Fradelizio et al.

a)

b)

c)

d)

Figure 5. 共a兲 Low-frequency, high-amplitude ground roll dominates the power spectrum of the raw shot gathers. 共b兲 After band-pass filtering and attenuation of the coherent noise, the frequency range of the signal is highlighted, with a dominant frequency of about 170 Hz. 共c兲 Deconvolution restores a balanced spectrum, broadened in its frequency range.

time shift at each station was constrained to be ⱕ Ⳳ.2 ms. In spite of the conservative approach, after testing different parameters, surface-consistent residual statics were not applied to the CMPs to avoid artificial enhancement of data coherency. In fact, autostatics determination is based solely on trace crosscorrelations. Therefore, a coherent event easily can be created out of random noise. This is actually a very powerful aspect of the processing: coherent events were obtained without any residual static or coherency filter. The nonrobustness of the residual static solution was corroborated by the low quality factor, a measure of confidence in the static solutions for each shot and receiver based on the correlation coefficients of the input traces against the pilot trace.

CMP sorting Because of source and receiver gaps and shifts resulting from bad shots and obstacles such as propane tanks, pipelines, cement well pads, and trailers, the actual CMP fold obtained using a natural bin size of half the receiver spacing varied significantly. This variation

a) 121 0

141

161

Channel number 181 201

221

231

Time (ms)

30

60

90

120

150

b)

1.4 450 400 350

Frequency (Hz)

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D imaging of a buried channel

300 250 200 150 100 50 0

1.0

0.5

Wavenumber (1/m) 0 –0.5

–1.0

–1.4

B91

degraded velocity analysis, stacking, and migration. Moreover, in some areas 共e.g., inlines 1110-1140兲, the fold — and consequently the S/N — drops drastically 共Figure 7a兲. Likewise, the distribution of the offsets in the CMP gathers is arranged well, considering the entire data set 共Figure 3c兲, but presents significant gaps when single gathers are evaluated. To maintain a more uniform CMP fold and offset distribution, flex binning was applied to the data set 共Spitzer et al., 2003兲. Bin size was enlarged from 0.175 to 0.20 m, but bin centers were maintained. Thus, the bins overlap, and some traces are assigned to more than one CMP gather. Flex binning generally improves the lateral continuity of prestack events. In our case, because of the minimal increment in bin size 共0.025 m兲, decreased lateral resolution was minimal, although the benefit in terms of uniform and increased fold was substantial. The new trace binning resulted in a much more uniformly distributed fold. Most importantly, the entire area of interest gained significant fold 共Figure 7b兲. As a result, the coherent signal in the CMP gathers is enhanced effectively by more uniform and consistent offsets 共Figure 8兲.

Figure 6. 共a兲 Single receiver line of band-pass-filtered shot gather and 共b兲 relative f-k spectrum. Both the time series and the spectrum highlight different kinds of coherent noise, specifically remediation noise 共N兲, airwave 共A兲, and ground roll 共G兲. In the f-k domain, they occur in different positions based on their velocity and frequency. These noise trains overcast in part the main reflection 共R兲.

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B92 Figure 7. 共a兲 Sorting the midpoints using the natural bin size of one-half the receiver spacing 共0.175 m兲 results in a nonuniform fold of the CMP gathers. 共b兲 A higher, more uniform fold is obtained by increasing the bin size to 0.2 m while maintaining the position of the bin centers. The dashed white line represents the 12-m depth contour of the channel as determined by well logs.

Figure 8. 共a兲 Three adjacent CMPs sorted using natural bins are compared to 共b兲 same-location CMPs sorted using enlarged, overlapping bins. Common traces are highlighted. The higher, more uniform offset content in the flex-binning gathers produces a more evident coherent signal 共arrows兲 that is very difficult to discriminate in the original CMPs. The data have been gained and band-pass filtered.

Fradelizio et al.

a)

b)

3D imaging of a buried channel

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Velocity analysis Although the quality of the data at this phase was improved considerably with respect to the original data, areas remained where determining the optimal stacking velocity of the reflections using typical tools, such as velocity spectrum plots and interactive velocity analysis, was difficult. Likewise, sequentially picking hyperbolas on CMP gathers was problematic because of the large lateral velocity variations and the limited offset ranges in the gathers. Therefore, velocity analysis was performed via constant-velocity-stack 共CVS兲 sections, perhaps the easiest and most robust method to use with shallow reflection data. Making CVS sections is also quite efficient, considering the relatively small volume of data. For the final analysis, CVS sections in the interval of 200–1000 m/s were displayed and studied carefully to determine the velocity field. A velocity step size of 40 m/s was considered adequate to get the best stack result. Finally, the stacking velocity field was smoothed over 2.1 and 4.2 m in the inline and crossline directions, respectively, to reduce large lateral velocity gradients in the stacking velocity model, which can distort NMO corrections 共Miller and Xia, 1998兲. Following NMO correction and stacking, a quick, convenient way to evaluate changes in the amplitude in the stacked data volume is to examine time slices. In our case, the structure of the paleochannel was well delineated in the horizontal slice at 50 ms 共Figure 9兲. The approximate depth was 12–13 m. The seismic response weakened in the southernmost part of the survey, where the fold was reduced 共see Figure 7b兲. The 3D seismic time slice also revealed a linear spatial pattern in the data, aligned east-west, an acquisition footprint resulting from the particular distribution of source-receiver azimuths.

B93

small 共0.175 m兲. At the Hill AFB site, the structure was sufficiently complicated that migration was necessary to produce a geometrically accurate image of the buried paleochannel. Poststack migration can produce poor results if the data have a low S/N and if the velocity structure is poorly determined or very complicated 共Yilmaz, 2001兲. Therefore, in addition to a processing sequence designed to maximize the signal response and minimize the noise, an accurate velocity model is needed. The interval velocity in depth used for the depth migration was derived directly from waveform tomography inversion of the seismic data along the 45 receiver lines reported by Gao et al. 共2007兲. The 2D velocity profiles were combined and interpolated in a 3D

Poststack processing Poststack processing is quite critical, and it strongly influences the final image quality. Because the desired final product of processing is the depth-migrated image of the paleochannel, a more aggressive predictive deconvolution with a 6-ms prediction distance and a 36 -ms filter length was applied to the traces after stacking to further sharpen the poststack image. In addition, the attenuated ground roll and lower frequency content resulting from stacking allowed for a 100–350 Hz band-pass filter. Finally, trace balancing and AGC were reapplied to the stacked data, given that lateral variations in amplitude adversely affect the migration process. No spatial filter was applied at this point to avoid introducing coherency in the stacked data.

Depth migration The need to migrate very shallow, high-resolution seismic reflection data is a function of depth and geometry of the target, near-surface velocity, and spatial and temporal sampling interval 共Black et al., 1994兲. If dips are small, conventional stacked sections provide a normal incidence representation of the subsurface. Moreover, for shallow reflectors involving very low seismic velocities, the maximum distance and depth moved by migration can be only a few CMP traces and time samples, respectively. In such a situation, migration makes no detectable change in the stacked data. However, in this specific case, the slopes of the paleochannel were steep, velocity of the near-surface sediments increased rapidly because of water saturation 共⬃1000 m/s for clay layer兲, and spatial sampling was very

Figure 9. Horizontal time slice through the final stacked volume at 50 ms 共⬃12–13-m depth兲, showing the channel’s deepest part. A simple median 2D spatial filter with 3 ⫻ 3 samples has been applied horizontally, and three samples 共3 ms兲 have been stacked vertically. The horizontal lineation in the time slice is an acquisition footprint related to the azimuthal content of the data. A weak response in the south part of the survey corresponds to a lower-than-average nonuniform fold of the CMPs 共see Figure 8b兲.

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B94

Fradelizio et al.

cube. They were smoothed vertically along 1 m and horizontally along one receiver line 共2.1 m兲 in the inline direction and two receiver lines 共4.2 m兲 in the crossline direction, respectively. A 3D poststack depth migration was applied to the stacked volume. A finitedifference algorithm 共Claerbout, 1985兲, implemented using an explicit scheme with 50° dip accuracy, was preferred because of its ability to handle large velocity variations. To produce the full 3D cube of stacked traces needed for migration, dead traces were filled in, padding each empty CMP position with a zero trace and creating a volume matching the velocity cube.

RESULTS AND INTERPRETATION The depth-migrated volume of seismic data represents a 3D image of the subsurface that can be interpreted directly. A quantitative evaluation of the result at Hill AFB is possible through comparison with available borehole logs. Sections cutting east-west across the paleochannel show that the topography of the top of the clay layer from the seismic reflection image is similar to the one measured by borehole logs 共Figure 10兲. The top of the clay layer is visible, and it is in good agreement with borehole data in the bottom part of the channel, although the shape appears smoothed and the steep edges more gentle than in the well-based image. The mismatch between seismic and well data becomes significant at shallower depths, where well control is very limited. The overall good agreement in both geometry and depth of the channel between seismic and well data is emphasized in a northsouth section cut along the buried-channel axis 共Figure 11兲. Agreement with well logs is satisfactory in geometry and depth along the bottom where the channel dips gently. However, some significant differences are evident where the degree of dip is higher. In particular, the southernmost portion of the seismic image shows a strong reflection deeper than the top of the clay layer inferred from borehole logs. Furthermore, the main reflection in the northernmost part of the seismic image, although not continuous, appears prolonged toward the north. Well control in these areas is difficult, in part because of a lack of data points and in part because of large variation between closely spaced measurements. The reliability of the migrated image is reiterated by slices at different depths in the migrated volume. At the bottom of the paleochannel, seismic data show small features not appearing in well data, although they are resolved only partially 共Figure 12兲. We used the minimum interpretation required when picking any seismic horizon, and we picked the reflection assumed to be the top of the clay layer on each of 527 lines. It was then possible to reconstruct the topography of the buried paleochannel as imaged by the reflection seismic survey 共Figure 13兲. Preliminary results of a 2005 shear-wave experiment performed along 2D test lines at the same site 共manuscript in preparation兲 confirm that the main reflection is indeed the alluvium-clay boundary and not the water table. Elevations of the top of the clay layer obtained with seismic data and with borehole logs can be compared. The absolute and signed differences between the two data sets confirm that the accuracy of the seismic image in the central, deepest part of the paleochannel is quite good 共Figure 14a兲. The largest differences between the well data and seismic data are on the western flank of the buried channel, between inlines 1300 and 1400, and on both sides of the channel. The number of well logs is particularly low in these areas.

We have identified spots that appear to be deeper than predicted by well data, potentially containing residual DNAPLs 共Figure 14b and c兲. In particular, three depressions are imaged in the northern half of the channel. The contour map of the deepest part of the paleochannel is of great importance for any remediation effort. Although the resolution is not high enough for the seismic image to be a stand-alone tool, it represents a valuable source of information for designing remediation strategies. The smoothed geometry and depth inaccuracy on the sides of the paleochannel possibly are related to the inherent limitation of poststack migration to resolve steep dips. In addition, monitoring wells are concentrated in the central part of the channel, often clustered in a few spots, whereas the rest of the area has been sampled at only a few points. Because the contaminant DNAPL is assumed to be ponding on the seismically well-resolved bottom of the buried channel, the image of the shallow part of the channel is less critical. Prestack depth migration was tested but not considered for the final image because of the poor result. This is not surprising, considering the low S/N of the shallow reflection data. Finally, we present a comparison between the results of the 3D reflection experiment, the result of both the waveform inversion of the 2D reflection data along each east-west line 共Gao et al., 2007兲, and 3D refraction traveltime tomography at the site 共Zelt et al., 2006兲. The waveform result is presented in the form of depth to the 1000m/s velocity contour, interpreted to be the clay top. A velocity of 1000 m/s was considered a good proxy for the P-wave velocity of the clay, based on the values of the velocity for partly saturated clays at the site and on the comparison between the velocity model and well logs. The depth contour of a velocity between 1000 and 1100 m/s in the tomographic model does, in fact, best match the well data. This proxy was used solely to compare results of different methods, not to drive the 3D seismic data processing. The refraction tomography result is a slice at 10 m depth from the preferred final velocity perturbation 共Figure 15兲. The paleochannel is clearly defined from 8 - to 11- m depth. In the north, the channel curves to the east in all three models and widens. The traveltime model exhibits the strongest anomaly in this area. The waveform result shows two areas with a relatively shallow clay top at about 10 and 40 m northing, which is not recognizable on the reflection image. South of about 20 m northing, the models exhibit significant variation, and the low-velocity anomaly disappears in the refraction image. The results of the 3D reflection experiment, although coming at the price of careful and intense processing, which is often difficult to achieve in the presence of considerable noise, are accurate and provide an image of the shallow subsurface directly usable to identify potential contaminant traps. In comparison, waveform tomography provides higher resolution than traveltime tomography in seismic imaging but requires an initial velocity model close enough to the true model to achieve meaningful convergence. In addition, the high computational cost limits its application to 2D cases. The traveltime inversion is robust, relatively simple, and inexpensive, but the refraction result has lower resolution than the reflection and waveform results. The images made using the different seismic methods are in good accordance with each other and the well data, allowing a crossvalidation of the final results.

B95 Figure 10. A set of sections cutting east-west across the paleochannel 共inlines兲 extracted from the depth-migrated cube show the shape of the channel as imaged by seismic data. The red lines represent the depth of the clay layer as inferred from the borehole logs. Although the depth is comparable, details of the bottom of the channel imaged by the seismic data can be quite different from predictions from the kriged well data. The image from the southernmost cross section suffers from low fold 共see Figure 8b兲.

Figure 11. North-south section out of the poststack depth-migrated cube 共crossline 80兲. The red line is the depth of the clay layer as inferred from the well logs. The yellow lines are well logs in a radius of 1 m around the crossline. Depth differences from well logs are shown as yellow boxes. The match between seismic and well image is notable in terms of geometry and depth along the bottom of the paleochannel. Agreement decreases across the edges of the channel and where well control is poor 共arrows兲.

0

150 0

num

Cro

sslin

50

e nu

150

mbe

r

251

0 110

0

120

0

130

ne

Inli

140

ber

1 100

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D imaging of a buried channel

Dep

th =

8m

Dep

th =

10m

Dep

th =

12m

N

Dep

th =

14m

Figure 12. Depth slices of the poststack-migrated cube. The depth of the clay from the borehole-log contour is overlain on the seismic image for comparison. No spatial filter has been applied. Two samples 共0.2 m兲 have been stacked vertically. Seismic data show deep spots not identified on the well-log reconstruction of the paleochannel 共arrow兲.

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B96 Figure 13. 共a兲 Contour map of the depth of the clay layer as reconstructed after picking the reflector on each of the 527 inlines in the migrated volume and 共b兲 by kriging the available well logs. Maps are plotted using a 0.5-m grid. Although the geometry of the paleochannel appears consistent with the one imaged using borehole logs, the steep slopes of the buried channel appear smoothed in the seismic image. The bias present in the seismic data 共a兲 is related to receivers geometry.

Fradelizio et al.

a)

b)

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D imaging of a buried channel

Figure 14. 共a兲 Map of the absolute difference between depth of the clay layer from borehole logs and depth of the clay from the seismic image confirms that the accuracy of the seismic image is maximized in the central part of the paleochannel. 共b兲 The deepest part of the channel as reconstructed using the well logs is compared to 共c兲, the one reconstructed from the seismic data, presenting spots that appear to be deeper than predicted by well data. The 12-m depth of the clay contour is shown as a white dashed line. The map is plotted using a 0.5⫻ 0.5-m grid.

a)

b)

c)

Figure 15. 共a兲 Map of the buried channel inferred from 3D poststack depth-migrated reflection data. 共b兲 Depth to the 1000-m/s isovelocity surface from 2D full-waveform inversion of 45 east-west 2D reflection lines extracted from the 3D data set 共Gao et al., 2007兲. 共c兲 Horizontal slice of the velocity model perturbations at 10- m depth from 3D refraction traveltime tomography 共Zelt et al., 2006兲. The reference velocity model is 1150 m/s. The 8- and 11–m depth-to-clay contours from the well data are labeled. White edges are unsampled regions.

CONCLUSIONS Our study at Hill AFB shows that, although challenging, it is possible to image and delineate the geometry of a buried paleochannel with a high degree of resolution using 3D seismic-reflection data. Moreover, the depth of the imaged clay layer is in good agreement

B97 with that inferred from borehole logs in the central, deepest part of the buried channel, although we find differences in depth of up to 2 m between seismic and well data. This is the most critical element for the remediation effort because the contaminant is assumed to be concentrated on the bottom of the paleochannel. The depth uncertainty on the sides of the channel might be a result of the intrinsic limitation of poststack migration to resolve steep dips or might be a result of unresolved lateral heterogeneity in the migration velocity model. In addition, the very limited number of well logs outside the deepest part of the channel does not allow a confident reconstruction of the depth of the clay layer. The variety of seismic investigations undertaken at this site has shown that OU-2 is an exceptionally complex site. The extremely shallow target 共10–15 m from topography兲 and the highly heterogeneous velocity field resulted in a complex data set. The seismic velocity field is complicated by heterogeneity in lithology, compaction, and water saturation. Poststack depth migration was necessary to correctly image the geometry of the paleochannel, requiring careful prestack and poststack processing to boost the S/N as much as possible, as well as an independent mean to determine a migration velocity model. Thus, the processing flow had to be designed carefully to remove the coherent noise trains from the seismic record, with deconvolution applied at several stages to sharpen the reflection events. The available waveform tomography velocity field provided the migration velocity model. The final result is superior to the one obtained using a less detailed traveltime velocity model. Furthermore, the study took advantage of borehole data available for a cross-validation of the final image and for depth control. In addition, well data must be considered important assets to corroborate seismic results. On the other hand, the map of the paleochannel obtained from seismic data can be used directly, eliminating the need for many wells. Finally, the improved efficiency and diminished cost of shallow seismic acquisition and processing make this method a valid alternative, or a complementary tool, in areas where the target depth or the soil or hydrologic conditions preclude an effective use of more cost-efficient geophysical methods, such as GPR.

ACKNOWLEDGMENTS

This study was supported by DOE grant DEFG07-97ER14827. The authors wish to thank all og the people involved in the field work, the personnel at Hill AFB site, and personnel from the IRIS PASSCAL instrument center who provided us with instrumentation. Special thanks go to Steve Danbom for useful discussions and suggestions.

B98

Fradelizio et al.

Downloaded 06/30/15 to 128.42.161.40. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

REFERENCES Baker, G. S., D. W. Steeples, and M. Drake, 1998, Muting the noise cone in near-surface reflection data: An example from southeastern Kansas: Geophysics, 63, 1332–1338. Black, R. A., D. W. Steeples, and R. D. Miller, 1994, Migration of shallow seismic reflection data: Geophysics, 59, 402–410. Claerbout, J. F., 1985, Imaging the earth’s interior: Blackwell Scientific Publishing Ltd. Dana, D., C. Zelt, and A. Levander, 1999, High-resolution seismic survey over a near-surface contamination site: 69th Annual International Meeting, SEG, Expanded Abstracts, 579–581. EPA — Environmental Protection Agency, 2003, Second five-year review report for Hill Air Force Base, Utah: http://www.epa.gov/superfund/sites/ fiveyear/f03-08003.pdf, accessed 23 March 2007. Gao, F., A. Levander, R. G. Pratt, C. A. Zelt, and G. L. Fradelizio, 2006, Waveform tomography at a groundwater contamination site: VSP-surface data set: Geophysics, 71, no. 1, H1–H11. ——–, 2007, Waveform tomography at a groundwater contamination site: Surface reflection data: Geophysics, 72, no. 5, G45–G55. Hornbostel, S., 1991, Spatial prediction filtering in the t-x and f-x domains: Geophysics, 56, 2019–2026. Larkin, S. P., and A. Levander, 1996, Wave equation datuming for improving deep crustal seismic images: Tectonophysics, 264, 371–379.

Levin, S. A., 1989, Surface-consistent deconvolution: Geophysics, 54, 1123–1133. Miller, R. D., and J. Xia, 1998, Large near-surface velocity gradients on shallow seismic reflection data: Geophysics, 63, 1348–1356. Pankow, J. F., and J. A. Cherry, 1996, Dense chlorinated solvents and other DNAPLs in groundwater: Waterloo Press. Spitzer, R., F. O. Nitsche, and A. G. Green, 2001, Reducing source-generated noise in shallow seismic data using linear and hyperbolic ␶ -p transformations: Geophysics, 66, 1612–1621. Spitzer, R., F. O. Nitsche, A. G. Green, and H. Horstmeyer, 2003, Efficient acquisition, processing, and interpretation strategy for shallow 3D seismic surveying: A case study: Geophysics, 68, 1792–1806. Steeples, D. W., 2001, Engineering and environmental geophysics at the millennium: Geophysics, 66, 31–35. Steeples, D. W., and R. D. Miller, 1998, Avoiding pitfalls in shallow seismic reflection surveys: Geophysics, 63, 1213–1224. Yilmaz, O., 2001, Seismic data analysis: Processing, inversion, and interpretation of seismic data: SEG. Young, R. A., and J. Sun, 1996, 3D ground penetrating radar imaging of a shallow aquifer at Hill Air Force Base, Utah: Journal of Environmental & Engineering Geophysics, 1, 97–108. Zelt, C. A., A. Azaria, and A. Levander, 2006, 3D seismic refraction traveltime tomography at a groundwater contamination site: Geophysics, 71, no. 5, H67–H78.