You are on page 1of 15

Detecting Cavities with Seismic Refraction Tomography: Can it be done?

Jacob R. Sheehan, William E. Doll, David B. Watson Oak Ridge National Laboratory, Oak Ridge, TN Wayne Mandel, Army Environmental Center, Aberdeen, MD

Abstract
Conventional seismic refraction analyses (e.g. delay-time or generalized reciprocal method) generally perform poorly for imaging many karst-related features because first-arriving seismic waves can generally circumvent the low-velocity target of interest without a major impact on travel time, particularly where the target is three-dimensional. We are using synthetic models and field data to evaluate the performance of seismic refraction tomography codes for karst detection. Two-dimensional karst models have been created and used to generate synthetic travel time data for tomography analysis. The results of this analysis suggest that the presence of a feature can sometimes be determined, but the velocity is not accurately represented. In no synthetic case that we have tested has the velocity of the cavity been an accurate approximation to the true low velocity. In contrast to the inconclusive synthetic results, we have refraction tomography data that clearly indicate karst features. A refraction tomography data set collected on the Oak Ridge Reservation shows a well-defined low velocity feature that crosses three parallel lines along an axis that extends more than 200 meters. The feature that was detected has velocities of about 1500-2000 m/s in a matrix of 3000-4000 m/s, reasonable velocities for a mud filled void in saprolite at these depths, rather than having a velocity slightly lower than the matrix velocity, as would be expected from our synthetic study. This suggests that the synthetic models are incomplete or inaccurate. In one case, matrix smoothing of the synthetic model before generation of synthetic traveltimes generates synthetic results that better match field results.

Introduction
Oak Ridge National Laboratory has been working with the United States Army Environmental Center to assess the performance of seismic refraction tomography (SRT) for karst terrains (Sheehan et. al, in review, Sheehan et. al, 2004, Sheehan et. al, 2003). These terrains frequently contain sinkholes, irregular and gradational bedrock interfaces, remnants of high velocity bedrock above these interfaces, deeply weathered fractures, and voids that may be air-, water-, or mud-filled. The seismic velocity of karst features will usually differ significantly from the background. Sinkholes are usually filled with mud or other unconsolidated material that has a low velocity compared to the carbonate in which sinkholes form. The transition from soil to bedrock is as distinct as that of sinkholes. In this paper, we are concerned with detection of karst voids, and will not be concerned with depressions, pinnacles, grikes, or other karst morphologic features (Carpenter et al, 1998). Voids or cavities that are air-filled will have a high velocity contrast with the surrounding bedrock, as will waterand mud-filled cavities to a lesser extent. This paper focuses on the detection of cavities. Two groups of data are examined here. The first is a set of four synthetic results that are representative of all of the synthetic models that we have examined. These models and the synthetic traveltime curves generated from them are created using the refraction tomography code GeoCT-II (version 2.3) (GeoTomo, LLC). The synthetic models allow us to have a true model with which to

compare the results generated by SRT using another refraction tomography code (Rayfract (version 2.51) from Intelligent Resources Inc). However, no matter how much care is taken when designing a synthetic model, it will never be a completely accurate depiction of the real subsurface. No matter how carefully constructed and applied, numerical analysis is based upon simplified and digitized representations of physical laws and models. In addition, most commercially available numerical modeling packages are based on 2-D models. 3-D numerical analysis is in development, but is currently too computationally intensive to be practical for most practical applications. Therefore field testing of SRT is necessary in addition to synthetic modeling. A set of four refraction tomography profiles has been collected in support of the NABIR Field Research Site (FRC) that supplement data collected at the site three years earlier (Doll et al). The field results have caused us to reconsider the accuracy or completeness of the synthetic models.

Synthetic modeling
A number of synthetic models have been created and used to investigate the effectiveness of SRT on karst. These models have been designed to test various properties, limitations and capabilities of SRT. These include: What factors affect ray coverage? How much and what type of ray coverage is needed to detect a feature? What is the resolution of SRT? How accurate are the velocities found using SRT? The depth of penetration is determined by a combination of the survey geometry and the velocity structure. As the effect of survey geometry is well known, we focus on the latter factor. In order to determine the factors of the velocity structure that control the ray coverage in the vicinity of a cavity several variations of a cavity model were constructed. One model has a sharp velocity contrast and no velocity gradient within the bedrock (Figure 1a). For this model no ray coverage penetrates deeper than the interface between the top layer that represents overburden and the layer representing bedrock that contains the modeled cavity. Therefore the cavity cannot be detected. Another has a small (10 m/s2) gradient within the bedrock (Figure 1b). This model generates ray coverage below and above the cavity, such that the cavity could affect the traveltimes and the resulting tomograms. The last two models have a change in bedrock velocity below the cavity. One of these has no internal gradient within the bedrock containing the cavity (Figure 1c) and one has a gradient of 10 m/s2 (Figure 1d). In both cases there is ray coverage in vicinity of the cavity, making it possible that the cavity will be detected. These models suggest one basic requisite: that the rays that penetrate to depths sufficient to image the cavity must have some path for being refracted back to the surface. This can happen if there is a change in bedrock velocity under the cavity or if the medium that contains the cavity has even a small vertical velocity gradient. It is normal for velocities to increase slightly with depth in sedimentary rocks, so in a karst investigation this requisite should be met. Therefore all subsequent modeling includes a slight gradient in the bedrock.

No gradient

Gradient

No gradient

Gradient

D
Figure 1: Variations velocity models (in m/s) to show ray coverage. Model A has a sharp boundary with no gradient in the bottom later. Model B has a less sharp transition and a small velocity gradient within the bottom later. Model C has no gradients, but does have a second boundary below the cavity. Model D is the same as Model C but does have velocity gradients in the top two layers.

Ray coverage alone is not enough to insure that the cavity can be detected. Figure 2 shows the traveltimes for the models shown in Figure 1b with and without the cavity. Although there is some difference in the traveltimes at large offset, they do not suggest a cavity. This is a good demonstration of a fundamental difficulty in seismic analysis, namely non-uniqueness. Although the model has significant ray coverage around the cavity, there is little sign of it in the traveltimes.

0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0.00 0 50

With Cavity Without Cavity

Time (s)

100
Position (m)

150

200

Figure 2: Overlay of traveltimes with and without ray coverage in the vicinity of the cavity.

Four additional models presented here show the full range of results that are obtained with SRT (Figure 3-Figure 6). For each model three images are shown. In each, the top image shows the model used to generate the synthetic data. The middle image is the model once again, with the ray coverage overlain. The model is best represented by Surfer, while the ray paths are best shown in the modeling software (GeoTomo). The bottom image shows the synthetic data generated from the model. Synthetic modeling results and conclusions The tomograms resulting from the analysis of the synthetic data sets for Models 1-4 are shown in Figure 7. Only in the Model 2 and 3 results are there any indications of the cavity in the synthetic model. In both cases the cavity is located in about the right place, but its velocity is not accurately represented. For Models 1 and 4, there is no sign of a cavity, despite ray coverage in the vicinity of the modeled cavity. From these results we conclude that SRT has the potential to detect cavities, but is not guaranteed to detect them. In addition, when a cavity is detected it does not appear distinctly. Instead of the true model velocity, the inverted velocity is only a little lower than that of the surrounding volume. In addition, there is a feature in the Model 1 results located between 60 and 70 meters that appears to be a cavity where none was modeled. This demonstrates that SRT can create false positives as well as false negatives. These false negative and positives are showing up with synthetic data, which does not include the inevitable picking errors and inaccuracies. The inclusion of such errors is likely to increase the occurrence of both false negatives and positives.

Elevation (m)

100 90 80 70 0 10 20 30 40 50 60 70 80 90 Position (m)

500 1100 1700 2300 2900 3500

Figure 3: Velocity (m/s) Model 1 (top), ray coverage (middle) and travel time curves (bottom).

Elevation (m)

100 50 0 50 100 150 200 Position (m) 250 300

200 1000 1800 2600 3400 4200

Figure 4: Velocity (m/s) Model 2(top), ray coverage (middle) and travel time curves (bottom).

Elevation (m)

100 50 0

50

100 150 200 Position (m)

250

300

250 1250 2250 3250 4250

Figure 5: Velocity (m/s) Model 3(top), ray coverage (middle) and travel time curves (bottom).

Elevation (m)

100 50 0 50 100 150 200 Position (m) 1800 2600 250 300

200

1000

Figure 6: Velocity (m/s) Model 4(top), ray coverage (middle) and travel time curves (bottom).

Elevation (m)

100

80

10 20 30 40 50 60 70 80 90 1250 2250 3250 4500

250 Elevation (m) 100 75 50 25 0 Elevation (m)

50

100

150

200

250

1250 2250 3250 4250 5250 100 75 50 0 500 100 75 50 0 500 50 1250 100 2000 150 2750 200 250 50 100 150 200 250

1500 2500 3500 4500

Figure 7: Resulting velocity models (m/s) from refraction tomography analysis of synthetic data generated from, top to bottom, Models 1, 2 3 and 4.

Elevation (m)

Field Results
A set of four new refraction tomography profiles (designated by Line A, B, C and D, Figure 12) were acquired in support of research at the Natural and Accelerated Bioremediation Research (NABIR) Field Research Center (FRC). Lines C and D are oriented parallel to an earlier line (Doll et al., 2002), designated Line E for this paper. The FRC site geology (Figure 8) consists of fill of varying depth over silt/clay saprolite. Below the saprolite is a transition zone of weathered bedrock with a gradation into more competent bedrock. The bedrock consists of interbedded shale and limestone.

Figure 8: Approximation of geology at the FRC field site.

Lines A, B, C and E used one-meter receiver spacing and two-meter shot spacing. Line E consisted of three collinear lines and combined for analysis. Line D was collected using 2 meter receiver spacing and 4 meter shot spacing. All data were collected using a 48 channel Geometrics Strataview seismograph. Ten Hz geophones were used for Lines A and C and 40 Hz receivers were used for line B. Lines C, D, and E each show a very well-defined (~ 10m wide) low velocity feature (Figure 9Figure 11). These low velocity features are all similar in size, at the same approximate depth, and fall on a line that is parallel to geologic strike at the field site (Figure 12). There is no such feature in lines A or B, which run roughly parallel to strike and perpendicular to the other three lines. Because of this and the correlation to geologic strike it is reasonable to assume that these low velocity features are not artifacts, but rather indicate a long conduit in the carbonate bedrock. This feature yields seismic velocities of approximately 1500-2000 m/s in a matrix of 3000-4000 m/s, representing a higher velocity contrast than obtained in any of our model results. The apparent cavity is below the water table so it cannot be air-filled, but its velocity is so low that we must surmise that it is water- or mud-filled.

100 Elevation (m) 95

3400

2800

90
2200

85 80 75 0 5 10 15 20 25 30 35 40 45 Position (m)
1600

1000

400

Figure 9: FRC line C velocity results in m/s. Elevation is relative, such that the surface is at 100 meters.
4400

Elevation (m)

300 290 280 30 40 50 60 70 80 90 100 110 120 130 Position (m)

3800 3200 2600 2000 1400 800 200

Figure 10: FRC line E velocity results in m/s. Elevation is actual elevation.

100 Elevation (m) 90 80 70 0 10 20 30 40 50 60 70 80 90 Position (m)

3800

3200

2600

2000

1400

800

200

Figure 11: FRC line D velocity results in m/s. Elevation is relative, such that the surface is at 100 meters.

Primary zone of contaminant transport and direction of strike

Seismic Feature
E

Figure 12: Seismic lines A, B, C, D and E with the low velocity feature location shown in blue. The previously determined primary zone of contaminant transport and the direction as strike is in green. The fact that there is no low-velocity feature found at lines A and B is further evidence that the ones found in lines C, D, and E are not artifacts of the inversion process.

What this means for synthetic modeling The synthetic modeling presented earlier indicates that a feature of this contrast and depth would be undetectable or would appear as a weak velocity low. However, field data show a feature that appears on three lines and follows the geologic strike at the site. These facts argue that the feature is real. We are anticipating that it will be verified by drilling in the near future; in the meantime we suspect that the synthetic modeling must be unsuitable / failing / inappropriate in one way or another. There are a few possible reasons for the apparent shortfalls in modeling. One is that velocity heterogeneity in the real subsurface generates ray coverage in areas where inhomogeneous layers would have no ray paths. One way test this possibility is to add velocity heterogeneity to the synthetic models. We did this by adding random Gaussian velocity variation to Model 4. Two of the velocity

heterogeneity grids that we used are shown in Figure 13. The synthetic data generated using the perturbed models did not lead to any significant change in the tomographic results. Although these two simple tests cases cannot rule out the possibility that random velocity perturbations present in the real world are responsible for the discrepancy, they imply that this explanation is not widely applicable. As discussed above, sharp boundaries tend to shield underlying areas from ray coverage. The synthetic models examined in this study were all created by adding layers and objects with distinct seismic velocities and velocity gradients. Care was given to avoid unrealistically sharp boundaries but this could not always be avoided. For example, the GeoTomo modeling code does not allow for a gradient transition into a cavity, as would be more realistic. To test this possibility, we have generated and analyzed matrix-smoothed versions of the models. The smoothed Model 4 and the model resulting from the analysis are shown in Figure 14. This result, unlike the one showed in Figure 7d, indicates successful inversion for the cavity. Not only is the cavity detected, but it is clearer than any of the other synthetic results. This suggests that the sharpness of the synthetic models affects the final results. One likely explanation is that the sharp boundary of the cavity was keeping what little ray coverage would pass through the cavity from doing so. However, applying the same smoothing to the other three models does not have the same effect. Therefore one of two things must be true: 1) The models represent situations where SRT would fail or 2) These models are still inaccurate in some way that is not resolved by smoothing.

Elevation (m)

100 50 0 100 50 0 50 100 150 200 Position (m) 250 300 50 100 150 200 Position (m) 250 300

100 80 60 40 20 0 -20 -40 -60 -80 -100

Figure 13: Two sets of random velocity (m/s) noise that were tested.

Elevation (m)

Elevation (m)

100
3000 2750

50 0
100 50 0 50 100 150 200 Position (m)
Conclusions

2500 2250

50

100 150 200 Position (m)

250

300

2000 1750 1500 1250 1000 750 500

Elevation (m)

250

250

Figure 14: Velocity (m/s) Model 4 after applying smoothing (top) and the resulting tomographic result (bottom).

Research with synthetic models for determining the capabilities and limitations of seismic refraction for cavity detection has had mixed results. It is often assumed that synthetic models represent the perfect test for analysis techniques. This is not always the case. In our research, we have found that refraction tomography may be more effective with field data than our model results might indicate. The natural conclusion from this is that the models are incomplete or flawed. For one of the four models (Model 4) smoothing the model before generating the synthetic data allowed the cavity to be detected whereas the unsmoothed model was undetectable. However, smoothing the other three models did not change the outcome. This demonstrates the complexity of synthetic modeling and analysis. We are continuing to evaluate the meaning of this result. In order to evaluate the effectiveness of SRT and to improve synthetic modeling procedures, we hope to construct a physical scale model. This will allow controlled acquisition of a three-dimensional model while avoiding many of the limitations of computer models. The physical model will allow us to avoid much of the oversimplification associated with digitized representations of physical laws and models. To the extent that a model is an accurate representation of the problem of interest, data collected using a physical model will more reliably replicate the physical response without errors associated with discretizing the properties of a model. In addition, a physical model, as long as it is large enough, will include 3-D effects.

References
Carpenter, P. J., W. E. Doll, and R. D. Kaufmann Geophysical character of buried sinkholes on the Oak Ridge Reservation, Tennessee, Jour. Environmental and Engineering Geophysics, v. 3, p. 133-146, 1998 Doll, W. E., T. J. Gamey, D. B. Watson, and P. M. Jardine. Geophysical profiling in support of a nitrate and uranium groundwater remediation study, Expanded abstract, SAGEEP 2002 Sheehan, J.R., W. E. Doll, and W. Mandell, 2003. Evaluation of refraction tomography codes for nearsurface applications. Extended abstract, presented at the 2003 Annual Meeting of the Society of Exploration Geophysicists, Dallas TX, October 26-31, 2003, 4. Sheehan, J.R., W. E. Doll and W. Mandell, 2004. Comparison of MASW and Refraction Tomography. Extended abstract, presented at the 2004 Annual Meeting of the Symposium on the Application of Geophysics to Engineering and Environmental Problems, Colorado Springs, CO, February 22-26 2004. Sheehan, J.R., W. E. Doll and M. Mandell, in review. An Evaluation of Methods and Available Software for Seismic Refraction Tomography Analysis, Accepted for publication in Journal of Environmental and Engineering Geophysics, 2005.

Acknowledgments
Oak Ridge National Laboratory is managed by UT-Battelle, LLC for the U. S. Department of Energy under contract DE-AC05-00OR22725. The submitted manuscript has been authored by a contractor of the U. S. Government. Accordingly, the U. S. Government retains a nonexclusive, royalty-free license to publish or reproduce the published form of this contribution, or allow others to do so, for U. S. Government purposes.

You might also like