Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Interpretation In characterizing a site, existing data often guide the specific methodology of additional data collection and should be integrated with the newly collected information. This integration is part of the modeling process. Modeling also includes the interpretation of data from specific instruments prior to integration efforts. Model output can be visualized to check for consistency as well as for presentation to the client. REVIEW OF EXISTING DATA Efforts to examine and interpret the near-surface portion of the earth usually involve multiple types of data. In addition to basic geographic map data, there are usually some geologic and hydrological data initially available, at least on a regional scale. Perhaps there might be some geophysical data that were collected for a particular project at a nearby location. One or more boreholes also may be available, often including natural gamma radiation and electrical resistivity logs. These various types of auxiliary data may be of unknown and variable qual- ity, and collected with instruments often of unknown calibration. Even nearby "ground-truth" borehole data may not be very useful or reliable. Not all descrip- tions of sample cuttings from a drilling operation are equally useful for ex- ample, some observations may have treated changes in color as the most impor- tant attribute rather than grain-size observations, which are technically more valuable. Hence, before interpretation is begun a critical review must be done of all existing data. This review serves to identify gaps and errors in the existing data, which can be addressed in subsequent field efforts. Data gaps may occur when 97
98 SEEING INTO THE EARTH data are not sampled often enough in space or time to prevent aliasing, as men- tioned briefly in the seismology section of Chapter 4. A common example of aliasing occurs in western movies when the wheels on a forward-moving buggy appear to spin backward because the visual field is not sampled often enough to represent the true picture. Consequently, data review should include consider- ation of the adequacy of the sampling, with respect to the project objectives, for each type of data. The process of assessing data to identify errors and omissions requires close attention to detail and is a laborious effort. A solution is to use well-trained and experienced people who are able to focus upon basics and are sensitive to the fact that errors and omissions can and do occur. Complex statistical methods or so- phisticated computer imaging cannot substitute for invalid or missing data. One of the most common methods of data display in two dimensions is through the use of contouring. Although human interpretive contouring is often difficult to beat in the geologic sense, machine contouring algorithms are now routinely used to prepare displays of geological and geophysical data, especially structural contour maps and potential field data maps. More recently, three- dimensional displays of seismic data have been used to beneficial effect (e.g., Dorn, 1998~. Multiple sources of data must be used to confirm site-specific conditions. When measurements by different methods agree, our interpretations will have a higher level of confidence. By virtue of redundancy, this process also provides a secondary form of quality assurance for individual sets of data, offering a reli- able, defensible means of testing the hypothesis embedded in the conceptual site model. DATA INTEGRATION When a single geophysical method is used to survey a complicated site, it usually is possible to create multiple models of the subsurface that fit the result- ing data. Another method, measuring different phenomena, will produce a differ- ent set of plausible models. In most cases, the intersection of the two sets of possible models is a smaller set that reduces the number of possible interpreta- tions. More surveys that measure even more phenomena will further constrain the interpretation. In an ideal case, enough data will be collected to produce a unique geologic or hydrologic model. In many cases, ground-truth data from boreholes and outcrops can be used to calibrate the geophysical parameters and result in model interpretation with a higher degree of confidence. Most site characterization projects use several types of geophysical measure- ments and sources of data. Data from each of these measurements often are interpreted in isolation; when this occurs, such data are neither integrated (a process sometimes called data fusion) nor interpreted simultaneously with data from other techniques. In those cases where an attempt is made to combine data
INTERPRETATION Tory Geolo:Seism~c Ma~etic - Lan~ll Thickness 99 Ground Electncal Ele=m PenetTating Resistibility magnetic Radar Gravity / / ~ it, r Ir~t~gration | . ~ r ~ I Inten:'retation - - ~ J/~ 1 ~ / \ ~ l~ndf;11 Bu red igighly Landfill Boundaries NSetaLs Conducive Stratigraphy Is - - FIGURE 5.1 Schematic diagram of the concept of integrating geologic information and data from diverse geophysical methods for determining properties of a landfill. (After Roberts et al., 1989.) sets, the data may be integrated and interpreted in only a qualitative fashion. Intrinsic relationships among different types of data often are uninvestigated, setting the stage for conflicting and irreconcilable interpretations. Successful site characterization often combines several different objectives and requires multiple measurements. Combining data from numerous methods might help resolve ambiguities and prevent faulty interpretation of individual measurements. In data interpretation it is important to take advantage of comple- mentary and redundant information in all available data from a site (see Figure 5.1~. However, because data can be combined and manipulated in so many ways, the end user or client (e.g., the site manager) is often confused, with no guide to determine the meaning of the composite results. Data integration should consider all of the data, not just geophysical data, acquired during a site characterization. Multiple sources of data provide the ability to check the quality of individual data sets against each other. Data inte- gration also provides an estimate of the statistics involved in characterizing a site and the uncertainty in the overall solution. Each observation contains an associ- ated error, and each data set is the result of a statistical distribution in space and/ or time. If disparate data sets are mapped into some common equivalent space, they should overlap. If not, a closer examination of the possible measurement or processing errors may be needed. For example, using Poisson's relation, it is possible to transform a magnetic map into a "pseudogravity" map by assuming some value of magnetic susceptibility for rock materials below the earth's sur- face. If the resulting pseudogravity map does not resemble an actual gravity map,
100 SEEING INTO THE EARTH one can assume that the magnetic susceptibility used in the calculation was not correct. Seismic, magnetic, electrical, gravity, and GPR signals arise from different subsurface physical parameters. The data can be inverted to obtain three-dimen- sional estimates of the constitutive properties of the ground. However, because the data sets arise from different physical properties, the various data sets cannot be readily combined before inversion. Data integration, then, is most often per- formed after cross-sectional, areal, or three-dimensional maps of the intrinsic physical properties uncovered by each of the imaging methods have been pre- pared. The process is iterative; each data set is reinterpreted, taking into account interpretation from other data sets until a consistent interpretation is obtained. For example, information obtained from GPR can be compared with that obtained from shallow seismic reflection, both of which are based on wave propa- gation. When different methods provide complementary information at a particu- lar site, combining the data is likely to provide more information than using any one method alone. The use of shallow, high-resolution seismic reflection tech- niques in concert with GPR has the potential to assist in characterizing sites in environmentally sensitive areas. Seismic and GPR techniques measure different physical parameters, but as shown in Figure 5.2, the two techniques can yield consistent results. At other sites, the two techniques might respond to changes in different regions of the subsurface and not yield a consistent interpretation. Seismic reflections arise from changes in acoustic impedance, that is, the product of seismic wave velocity and density must change for a seismic reflection to occur. If seismic velocity increases by the same amount that density decreases at a given interface, no seismic reflection is produced from the interface. An example occurs in salt deposits, which commonly do not yield good seismic reflectors at internal inter- faces. Ground penetrating radar, on the other hand, responds to changes in the constitutive electrical parameters (permittivity and conductivity) of the subsur- face. If either of these electromagnetic parameters changes at the interface used in the example given above, a radar reflection may occur where no seismic reflec- tion would occur. Imagine an opposite example where the constitutive param- eters are constant across an interface at which either bulk density or seismic wave velocity varies. In sum, seismic data and GPR data tell us about different physical parameters of the earth material being surveyed and often can be used to compare each other's results. Important geologic and hydrologic interfaces often represent changes in prop- erties that include density, seismic velocity, electrical conductivity, and dielectric permittivity. Knowledge of these four parameters enhances the possibility of predicting fluid flow paths, particularly in fractured media. For example, seismic velocity usually decreases in fracture zones, and radar wave velocity may in- crease in these same fracture zones, particularly when a large increase in air-filled
INTERPRETATION 101 pore space is involved. Conversely, where a fracture and the pore spaces are filled with precipitated minerals, seismic waves may propagate more quickly and radar waves more slowly. Clay tends to attenuate radar energy, whereas seismic energy often is not attenuated rapidly by propagation in clays. At other sites, seismic waves might be attenuated rapidly in dry, quartzitic sand, whereas radar waves propagate well in the same medium. Depending on local geologic and hydrologic conditions certain types of stratigraphic variation may be detectable directly (by the presence of a reflec- tion), indirectly (by the disruption of some other reflector), or not at all. The presence or absence of a reflection can also depend on the seismic or radar parameters used (see Figure 5.3~. The absence of evidence of an seismic or GPR reflection is not necessarily evidence of the absence of a stratigraphic variation. Analysis of both elastic and radar-frequency electromagnetic survey data with densely spaced measurements is essential to the construction of a high- quality subsurface image. Although a variety of field procedures has been used to produce such coverage in individual seismic and GPR surveys, little is known about how the two techniques might practicably be combined in very high reso- lution site characterization surveys. In the same set of data it is possible to show large differences in the resolu- tion and accuracy of features depending on a priori assumptions about what is in the subsurface. Therefore, inappropriate visualization of data and integration of multiple sources of data could be misleading. Developing the process of data integration requires considerable future research. How do we integrate disparate data sets from geophysics, geochemistry, hydrology, and biology, and map the multidimensional data into an integrated solution? What kind of statistics should be used, and what levels of confidence are required? Further investigation of such questions is needed to optimize data integration and interpretation. MODELING Models based on an understanding of physical, chemical, and biological properties and processes (in contrast to those based on empirical correlation) are of great value in the effective use of noninvasive methods in site investigations. Numerical models can provide linkages between the phenomena being measured and the properties and processes occurring in the earth. They provide tools for optimizing survey design, quantifying uncertainties and limitations associated with data acquisition, and validating interpretations. There are many existing numerical models that are potentially useful, but they should be catalogued, documented, and made user friendly and easy to locate to fulfill their potential. Our understanding of, and ability to exploit, a particular characterization phenomenon can be improved by an iterative process involving the following approaches: (1) empirical (observation of an apparent relationship between the phenomenon and a property or process of interest), (2) analytical (experimental
102 (a) . o a . `,,30 ._ In ._ (in 30 In (b) ~ onto ~ ~ ~F20 o (C) SEEING INTO THE EARTH Offset Along Profile (m) 10 15 a_ ~ 10 30 20 25 30 Seismic-Reflection/GPR-Based Geologic Interpretation 0 5 O s ~ 2 OffsetAlong Profile (m) 1 0 1 5 20 25 30 and theoretical research to explain the relationship), and (3) numerical (computer models of cause and effect, which can be useful in a predictive sense). If models are well designed and easy to use, they (1) make analytical expertise available to practitioners, (2) enable conceptual understanding of the relationships between the phenomenon and the properties or processes, and (3) help practitioners and clients understand the capabilities and limitations of the measurements. In addition, rigorous numerical models can be used to improve the quality and reliability of nonintrusive site characterization surveys. During survey de- sign, numerical models can be used to help choose the characterization method, quantify the anticipated signal and noise, and optimize the proposed survey pa- rameters. Processing and interpretation make extensive use of computer models,
INTERPRETATION 103 FIGURE 5.2 An example of qualitatively merged geophysical imaging of the shal- low subsurface. (a) An uninterpreted and interpreted seismic reflection profile along a 30-m transect in the Arkansas River alluvial valley ~1 km southeast of Great Bend, Kansas. Geophone spacing was 10 cm and the seismic source was a 22-caliber rifle with subsonic short ammunition fired 10 cm downhole. (b) A common-offset GPR profile using a 225 MHz antenna coincident with the seismic profile. The seismic interpretation is overlain on the GPR section. (c) Geological interpretation of the 30 m transect created by merging the individual interpretations of the seismic and GPR data and adjusting coincident reflectors. The three main layers, from top to bottom, represent the Platte series soil profile, an unstratified medium sand (bound on the top by an erosional unconformity), and a cross-stratified medium sand to medium gravel with various bounding surfaces (identified as individual lines on the interpretation). The interpretation was field constrained by a nearby ~2-m-deep hand- dug hole. At-2.1 m is the top of the saturated zone, constrained by a nearby monitoring well. The water table is easily identified on the seismic section but absent on the GPR section (possibly related to the diffuse nature of the boundary relative to the GPR wavelength). Although not quantitatively "fused" by some in- version technique, the coincident profiling using seismic and GPR methods im- proved the detail and confidence of the interpretation. Figure courtesy of Gregory S. Baker, 1999, University of Kansas). especially for inversion techniques. For critical (e.g., hazardous) sites, the most important use of models is to validate interpretations and do quantitative sensitiv- ity analyses. Models, whether physical, chemical, geological, or hydrological, must be mathematically validated by use of easily verifiable cases. One way to do this is to compare results for cases that have known analytical solutions. Another way to analyze models is to use sensitivity analysis (e.g., McElwee and Yukler, 1978) in which the response of a model is examined in terms of the mathematical deriva- tives of the constitutive equations. VISUALIZATION An important advance of recent years is in the visualization of geophysical data. Previously, data were usually presented as measured field values, corrected for drift with other simple corrections. For example, with electrical and electro- magnetic data, a common form of data processing is still to normalize to apparent resistivity or apparent conductivity, which simply matches the data to a homoge- neous earth model. Plotting the data in "pseudosection" form using simple guide- lines provides depth interpretation. The impact of this presentation is limited; an experienced interpreter usually is needed to convert these plots to a geologic model of the earth. Interpretation with such displays usually consists of locating
104 SEEING INTO THE EARTH a) b) on 5.e 1 0.o _` ~ say POSITION (~) 15.D G) 20.O 25.D A' ~',-,,'i,~ it' ~ - iV !' t ~ i, ,.. F. 4~- ~ " ~ i ~ ' ~ {~: ~ ,~ ~ · i1!!, li~i!ili !i':'li i ~I.,',i,' ." i!.jl.t ·2l~1, ~ Id, ~ ·, ~ !I1`:~..~ ......... !. bilk ~ 'it! '\' Oral ; ! i It '4i l ' ~t ~7 ~ ~ ~ [t , ,! j j~l. j, ;I!t `rl,!l,jl. ~j I i I i; ii, i ||li Wapiti i:i l 'i i; 7 i ~:: ~ ~ i ~) O a,....,':, i., .~ ~ i.' .~.~,,.,i~j~,j!~e i ~ i FIGURE 5.3 Visibility of GPR reflections decreases as the signal-to-noise (S/N) ratio increases; (a) is for S/N of 15.0, (b) for S/N of 7.5, and (c) for S/N of 1.5. T is the reflection from the top of a buried tank, B from the bottom of the tank, and F from a fluid (air/gasoline) interface inside the tank. Figure is from Zeng and McMechan (1997). anomalies ("bump hunting") and ascribing geologic significance. These presen- tations can be misleading because pseudodepth may not be true depth and data artifacts might appear as geologic features. Advances in modeling, inversion, and visualization now make it possible to present data in a geologically and visually meaningful way. Shaded relief maps, for instance, have revolutionized the presentation of potential field data (e.g., Plate 3~. The shaded surface can reveal features that may be invisible when displayed using contouring or simple pseudocolor. Color presentations also con- vey a great deal of information to users and often enable easier recognition of significant features. However, with the introduction of displays that are pleasing to the eye comes the danger of misleading viewers a change of color scale or rendering can often completely alter the significance an interpreter places on a feature. It is incumbent on the individuals presenting the data that their presenta- tion conveys as accurately as possible the actual geology or geologic process. Error estimates should be displayed with each presentation, along with alterna- tive displays. In addition to visualization of the subsurface, the data can be integrated with many other types of data about the site's location (e.g., transportation networks, population, ecosystems, topography, resources, land-use, and other locationally referenced themes) using geographic information systems (GIS). When inte- grated within a GIS context, models or "what-if" scenarios can be tested for use in a broader decision-making process.
INTERPRETATION RECOMMENDATION Scientists and engineers must improve their ability to integrate multidisciplinary data for modeling, visualizing, and understanding the subsurface. 105 The interpretation of characterization data has both creative and quantitative components. The creative component consists of conceiving all of the possible geologic models likely to explain the data; the quantitative component involves generating synthetic data for every possible model to demonstrate whether a particular model is consistent with the field data. Generating multiple synthetic data sets from a single geologic model, although not done routinely, is technically feasible. It requires only the development of modeling codes based on an ac- cepted set of programming standards. Conceiving a geologic model that will fit multiple data sets is much more difficult. Advances in this area will include both technical (e.g., simultaneous inversion) and human (e.g., studies of team dynam- ics) elements. Interpretations of any set of multiple measurements will be strengthened, and ambiguity reduced, if they are the result of early integration and simultaneous inversion of diverse data types. The following areas of research are needed to improve the efficacy and rigor of data fusion and integrated interpretation: . Develop a better understanding of the coupling and interactions among the physical, chemical, and biological properties and processes that affect the measurements done in characterization surveys. . Develop integrated models that allow simultaneous modeling of simu- lated data sets from several multiple surveys over a single geologic model. · Based on the understanding of physical, biological, and chemical proper- ties and processes, develop mathematical tools and computer programs that are able to perform simultaneous, quantitative inversion of multiparameter data sets. · Create three-dimensional scientific visualization tools and techniques that will allow human interpreters to monitor the inversion process, assess the result- ing geologic models, and improve the quality of the interpretation. In the area of modeling, a variety of needs can be identified: · Many numerical modeling programs currently exist in universities and government laboratories. However, some are obscure, are difficult to use, and require computing facilities not readily available to many practitioners. Major benefits can be achieved in a short time by adapting existing codes to "standard"
106 computing platforms and by adding age practitioner to find and use. SEEING INTO THE EARTH r ;~ Interfaces that make them easy for the aver · Numencal models become, in effect, "expert advisers" to the practitio- ners who use them. In some cases they have great influence because of the human tendency to believe what comes out of a computer. Therefore, it is important that the expertise embedded in numerical models be up-to-date and correct. Appropn- ate regulatory agencies or professional societies should establish a program of certification of numerical models to be used in site characterization surveys. · Some phenomena have not yet been modeled; others have been modeled with so many simplifying assumptions that the models often are not realistic. Universities and government laboratories should be encouraged and supported to identify deficiencies and develop rigorous computer models that provide realistic descriptions of subsurface properties and processes. · Given the need for data fusion and integrated interpretation, universities and government laboratories also should be encouraged to develop and validate integrated modeling software explicitly designed for site charactenzation. · To facilitate the broader use of computer modeling there should be a clearinghouse or repository to (1) facilitate discovery of available modeling soft- ware; (2) provide standard data sets against which codes and models can be tested; and (3) assist the pnvate, academic, and government sectors in developing training curricula in the use of computer modeling. REFERENCES Baker, G. S., 1999. Seismic Imaging Shallower than Three Meters, Ph.D. dissertation, Department of Geology, The University of Kansas, Lawrence, 320 pp. Dorn, G. A., 1998. Modern 3-D seismic interpretation, The Leading Edge 17, 1262-1272. McElwee, C. D., and M. A. Yukler, 1978. Sensitivity of groundwater flow models with respect to variations in transmissivity and storage, Water Resources Research 14, 451. Roberts, R. L., W. J. Hinze, and D. I. Leap, 1989. A Multi-techniques geophysical approach to landfill investigations, in Proceedings of the Third National Outdoor Action Conference on Aquifer Restoration, Ground Water Monitoring, and Geophysical Methods, National Water Well Association, Dublin, Ohio, pp. 797-811. Zeng, X., and G. A. McMechan, 1997. GPR characterization of buried tanks and pipes, Geophysics 62(3), 797-806.
