<%BANNER%>

Integration of Ground-Penetrating Radar, Geographic Information Systems, and Global Positioning Systems for 3-Dimensiona...


PAGE 1

INTEGRATING GROUND-PENETRAT ING RADAR, GEOGRAPHIC INFORMATION SYSTEMS AND GLOBAL POSITIONING SYSTEMS FOR 3-DIMENSIONAL SOIL MODELING By MICHAEL A. TISCHLER A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2003

PAGE 2

Copyright 2003 by Micheal A. Tischler

PAGE 3

I would like to dedicate this manuscript to my family and friends. My father and stepmother; James and Laura Tischler; mother and stepfather, Patricia and Merrill Davis, and brother; James Tischler, Jr never wavered in their love and support. To them and the rest of my family, I am truly and forever grateful.

PAGE 4

iv ACKNOWLEDGMENTS This research could not have been co mpleted without the input, advice, and support of dozens of individuals. Although th ere is neither the space nor the words to convey my gratitude to everyone who offered support, I would like to express my thanks to a few. Dr. Mary Collins was an exceptional advisor in every aspect of the word. She allowed enough freedom for me to forge my own ideas while always guiding me toward the final goal. I will always be grateful to her for the time and energy that she invested in my education at the University of Florid a. Without the assistance of Dr. Sabine Grunwald, I would never have had the knowledge to perform the complex spatial analysis and validation procedures necessary to comp lete the research. John Schultz and Larry Ellis both offered more assistance and advi ce than I could expect anyone not personally involved to provide. I would also like to thank Dr. Jimmie Richardson and Dr. David Hopkins of North Dakota State University. Th ey saw the scientist within me before I knew there was one myself. Finally, I would like to thank my family and friends in full for their support and kindness. None of them will ever realize how thankful I really am. Come to the edge they said. He said: I am afraid. Come to the edge they said. He came. They pushed him, and he flew. . ." Adapted from Guillaume Apollinaire

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................................................................................. iv LIST OF TABLES............................................................................................................ vii LIST OF FIGURES ......................................................................................................... viii ABSTRACT....................................................................................................................... ..x CHAPTERS 1 INTRODUCTION ...........................................................................................................2 2 GROUND-PENETRATING RADAR (GPR).................................................................3 Background..................................................................................................................... 3 Theory......................................................................................................................... .... 4 Site Suitability............................................................................................................... .. 8 Data Structure ............................................................................................................... 11 3 GEOGRAPHIC INFORMATION SYSTEMS (GIS)....................................................13 History........................................................................................................................ ... 13 Georeferenced Data ...................................................................................................... 15 File Formats .................................................................................................................. 1 8 Analysis....................................................................................................................... .. 18 4 GLOBAL POSITIONING SYSTEMS (GPS)...............................................................20 5 FIELD SITE AT PINE ACRES.....................................................................................23 Soils.......................................................................................................................... ..... 23 Geology........................................................................................................................ 27 Climate........................................................................................................................ .. 30 Topography................................................................................................................... 31 6 MATERIALS AND METHODS...................................................................................33 Data Collection ............................................................................................................. 33 Ground-Penetrating Radar Post-Processing.................................................................. 34

PAGE 6

vi Model Generation ......................................................................................................... 40 Data Preparation............................................................................................................ 41 7 RESULTS AND DISCUSSION....................................................................................46 Variograms and Kriging ............................................................................................... 46 Modeling Parameters .................................................................................................... 52 Model 1........................................................................................................................ 54 Model 2........................................................................................................................ 57 Model 3........................................................................................................................ 59 8 MODEL VALIDATION ...............................................................................................64 Model 1........................................................................................................................ 64 Model 2........................................................................................................................ 67 Model 3........................................................................................................................ 68 Model 4........................................................................................................................ 68 9 SUMMARY AND CONCLUSIONS ............................................................................71 Data Collection ............................................................................................................. 71 Ground-Penetrating Radar Processing.......................................................................... 72 Qualitative Methods...................................................................................................... 73 Model 1........................................................................................................................ 75 Model 2........................................................................................................................ 75 Models 3 and 4.............................................................................................................. 76 APPENDIX SOIL PROFILE DESCRIPTIONS ..............................................................77 BIOGRAPHICAL SKETCH .............................................................................................82

PAGE 7

vii LIST OF TABLES Table page 2-1 Dielectric constant s of various materials........................................................................10 2-2 Cation exchange capaci ties of soil materials..................................................................11 5-1 Corner coordinates of the research plot..........................................................................26 6-1 Example of the tabular data before any statistical manipulation...................................43 7-1 Tabular GPR data ready for 3D modeling.....................................................................56 A-1 Description of representativ e pedons at the research site..............................................77

PAGE 8

viii LIST OF FIGURES Figure page 2-1 Monostatic and bistatic antenna configuration...............................................................5 2-2 Relative depth of penetration and resolution of various antennas..................................6 2-3 Ground-Penetrating Radar wave pr opagation, scattering and reflection........................9 2-4 Ground-Penetrating Radar traces and samples...............................................................12 3-1 Three types of vector da ta: points, lines, and polygons.................................................16 3-2 Raster data consisti ng of equal sized cells......................................................................17 5-1 Location of research s ite in Marion County, Florida......................................................24 5-2 Research plot at Pine Acres............................................................................................25 5-3 Argillic horizon at Pine Acres Research Facility. ..........................................................27 5-4 General Soil Associati ons Marion County, Florida.....................................................28 5-5 Modified Palmer Drought Index ....................................................................................31 5-6 Palmer Z Index of precipitation......................................................................................32 6-1 Simplified interior of a GPR antenna housing unit........................................................36 6-2 Uncorrected depth to surface GPR profile. ....................................................................37 6-3 Corrected depth to surface GPR profile.. .......................................................................37 6-4 Background filter of GPR profile...................................................................................39 6-5 Spectral whitening filter of GPR profile.........................................................................40 6-6 Compression of GPR profile. .........................................................................................41 6-7 Scatterplot of A versus depth for all GPR data............................................................44 6-8 Ground-Penetrating Radar profile image described by Figure 6-7. ...............................45

PAGE 9

ix 7-1 Range, sill, and nugget value for a variogram............................................................47 7-2 Variograms of GPR data at different depths..............................................................48 7-3 Depth versus Range for several variograms...............................................................53 7-4 Mean depth of clustered variables..............................................................................55 7-5 Model 1, using RockWorks 2002 software................................................................58 7-6 3D display of Model 2 as viewed in ArcScene..........................................................60 7-7 Standard error of prediction of Model 2.....................................................................61 7-8 Picked" argillic horizon............................................................................................62 7-9 Image of Model 3.......................................................................................................62 7-10 The standard error of prediction for Model 3...........................................................63 8-1 Validation point locations...........................................................................................65 8-2 Histogram of depth to argill ic values for validation points........................................66 8-3 Validation procedure for Model 1..............................................................................67 8-4 Regression equation fo r validation point, Model 2....................................................69 8-5 Image of Model 4.......................................................................................................70

PAGE 10

x Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science INTEGRATING GROUND-PENETRAT ING RADAR, GEOGRAPHIC INFORMATION SYSTEMS, AND GLOBAL POSITIONI NG SYSTEMS FOR 3-DIMENSIONAL SOIL MODELING By Michael A. Tischler May 2003 Chair: Dr. Mary E. Collins Department: Soil and Water Science Department Ground-Penetrating Radar (GPR) has become a useful and efficient instrument for gathering information about subsurface diagnost ic horizons in Florida soils. Geographic Information Systems (GIS) in are a popular and valuable tool for spa tial data analysis of real world features in a digital environment. Florida has a vast array of freely and readily available GIS data. Ground-Penetrating Ra dar can be linked to GIS by using Global Positioning Systems (GPS). By combining GPR, GPS, and GIS technologies, a more detailed geophysical survey can be comple ted for an area of interest by integrating hydrologic, pedologic, and geologi c data. Thus, the objectives of this research were to identify subsurface soil layers using GPR and their geographic position with a highly accurate GPS; to develop a procedure to import GPR data into a popular software package, such as ArcGIS, and; to create 3D subsurface models based on the imported GPR data. The site for this study was the Pl ant Science Research and Education Center in Marion County, Florida. The soils are characterized by Recent-Pleistocene-age sand over the clayey, marine deposited Plio-Mio cene-age Hawthorn Formation which drapes the Eocene-age Ocala Limestone. Consequently soils in the research area vary from

PAGE 11

xi deep quartz sands (Typic Quartzipsamments) to shallow outcrops of the Hawthorn Formation (Arenic Hapludalfs). A GPR su rvey was performed on a 160 m 320 m grid to gather data for processing. Four subsurface models estimating the depth to argillic horizon were created using a variety of specialized GPR data filters and geostatistical data analyses. The models were compared with ground-truth point s that measured the depth to argillic horizon to validate each mode l and calculate error. These models may assist research station pers onnel to determine best mana gement practices (including experimental plot placement, irrigation mana gement, fertilizer treatment, and pesticide applications).

PAGE 12

1 CHAPTER 1 INTRODUCTION Ground-penetrating radar (GPR) has proven to be a useful and efficient remote sensing geophysical instrument for gathering information about near-surface pedologic and geologic materials. Geographic Inform ation Systems (GIS) provide a means of storing, manipulating, analyzing, and displaying spatially distributed data in a two-dimensional (2D) or three-dimensional (3D) view. Combining the efficiency and practicality of GPR data with the visual appeal, analysis, a nd interpretive pow er of GIS is the next logical step in the evolution of both technologies, and can be accomplished using Global Positioning Systems (GPS). Recently, georeferenced datasets providi ng information about the earth and its environment have become widely available. The accumulation of this information has created a demand for 3D geophysic al data that can be analy zed in conjunction with these easily accessible datasets. However, no met hods exist for combining georeferenced GPR data with GIS datasets, (which would reduce time and costs while increasing the interpretive quality of the information). Developing such a method is the most important goal of this research. By transforming graphic GPR data to nume ric values and spatial coordinates, the data can be manipulated, transformed, converted, a nd integrated into a variety of software packages capable of accomplishing specific port ions of the overall objectives. Ground-Penetrating Radar analysis software can process data to allow better visualization of the subsurface features. Statistical soft ware can be used to make mathematical

PAGE 13

2 manipulations of the data to ease interpreta tion. Modeling software can interpolate areas where little or no data exists to create a solid 3D diagram. Finally, GIS software can be used to display the modeled information in combination with already available information about the earth, environment, and na tural resources. This research intends to bridge the gap between GPR and GIS to allow powerful analysis in detailed studies and projects. The specific objectives of this research project are to: • Conduct a 3D GPR survey to obtain information about soil stratigraphy • Create a 3D model from the GPR data • Develop methodology to integrate the 3D GPR data into a GIS The 3D models created need to meet 3 stringent criteria. They must • Be easily understood Conventional 3D diagrams can be difficult for users to comprehend and interpret. • Represent the soil features to an acceptable degree of accuracy However, we were more interested in developing a method than in refining the accuracy and precision of the model. • Have the ability to be combined with GIS datasets The most imperative objective of this proj ect is the integration of GPR data to a GIS platform. We used GPR information to create both vector and raster models viewable in GIS. The processing and modeling methodol ogy were developed specifically for this research and for the site-specific conditions at the data collection si te. The methods used were intended to produce an accurate model of subsurface soil horizons while maintaining efficiency in their production.

PAGE 14

3 CHAPTER 2 GROUND-PENETRATING RADAR (GPR) Background The history of GPR in research spans ove r 70 years. The first practical use of GPR was by Sterns in 1929 to determine the de pth of a glacier in Au stria (Olhoeft 1996). Advances and uses lay virtually dormant until the United States Air Force became interested in developi ng better radar systems in the late 1950s. Several pilots crashed while attempting to land on icy runways in Greenland because the Air Force radar was seeing through ice and reading inaccurate altitudes. In 1972, GPR was sent to the moon on the Apollo 17 mission, using an apparatus sim ilar to that used in Sterns’ glacier survey in 1929. Before 1972, GPR was not commerc ially available. However, in 1972, Geophysical Survey Systems Inc. (GSSI) was founded and began to sell the first commercially available GPR units. Today, n early 300 patents are associated with GPR technology, and many universities and privat e companies use GPR technology (Olhoeft 2000). Used in such varied disciplines as so il science, geology, archeology, engineering, construction, mining, military, agriculture, and environmenta l science, GPR has proven to be a valuable and diverse instrument. Soil scientists and geologists are able to discern subsurface anomalies and formations (Collin s 1990); archeologists are able to locate buried artifacts (Basile et al 2000); engineers and the cons truction industry are able to quickly locate utility pipes and buried cables (Grandjean et al. 2000); miners are able to more efficiently retrieve valuable minerals and ore (Sigurdsson and Overgaard 1996); the

PAGE 15

4 military saves lives by locating buried land mines and unexploded munitions (Hyde 1997); environmental scientists are able to track contaminant leaks and hazard waste movement in the soil (Atekwana et al. 1998); and hydrologists use 3D georeferenced GPR transects to identify subsurface flow path ways (Gish et al. 2002). In addition to the myriad of present uses, an international gr oup of scientists plans to conduct GPR surveys of Mars in the future (Marsal et al. 2000) Ground-Penetrating Radar has proven to be a useful, accurate, and efficient means of da ta collection in countless instances, and continues to be at the forefr ont of research and technology. Theory The function of GPR is similar to reflection seismology, a technique traditionally used by exploration geophysicists. By se nding electromagnetic energy into the earth, information about the lithology of a site can be determined. Ground-penetrating radar uses a transmitting and receiving antenna in bistatic or monostatic mode to send a variable frequency radar signal into the ground. The signal is reflected and sent back to the antenna. In bistatic mode the transmitter and receiver antennas are held apart at a constant distance by a rigid frame. Frequent ly, antennas in bistatic mode are at zerooffset, meaning that the point of transmission is also the point of reception, or that both antennas are held at the same point. In monostatic mode, the transmitting and receiving antennas are independent of each other and can be used in a variety of arrays to gather information. Figure 2-1 illustrates the differe nces between bistatic and monostatic data collection modes. Depending on the desired output and soil conditions, severa l antenna options exist. Generally, two antenna properties determine the use of the equipment at a particular site: depth of penetration and re solution. The depth of penetration is the

PAGE 16

5 Figure 2-1. Monostatic and bi static antenna configurati on. In monostatic collection configuration, the transmitting and receiving antenna are located in the same position. In bistatic collection config uration, the transmitting and receiving antennas are separated by an offset distance determined by the user. deepest depth at which a GPR signal will be re flected back to the surface. Resolution is the ability to differentiate features of diffe rent sizes from the surrounding media. Smaller features can be identified by an antenna providing high resolution than by an antenna providing low resolution. While depth of pe netration and resolution are very strongly correlated with antenna frequenc y, their relationships are diffe rent. As antenna frequency increases, depth of penetration decrease s and resolution increa ses (Figure 2-2). Conversely, as frequency decreases, depth of penetration increases and resolution decreases. As the objectives of a project change, so might the choice of radar equipment for that objective.

PAGE 17

6 Ranked General Antenna Properties Scale of 1-101 2 3 4 5 6 7 8 9 10 10 9 8 7 6 5 4 3 2 1 Depth of Penetration Power of Resolution 100 MHz 200 MHz 300 MHz 400 MHz 500 MHz 600 MHz 700 MHz 800 MHz 900 MHz 1000 MHz Figure 2-2. Relative depth of penetration and resolution of various antennas, based on a scale of 1 to 10. Generally, antenna fre quency is directly proportional to resolving capability and inversely pr oportional to depth of penetration. Gathering as much information about the site a priori allows for an informed equipment selection. Specificall y, knowing the depth of possibl e features of interest and existing soil conditions that may inhibit radar penetration will aid in the selection of proper equipment. Ground-penetrating radar antennas collect information from the subsurface using electromagnetic waves. A voltage pulse is sent into the ground by the antenna, beginning the propagation of a radar wave that travels downward through the soil. As changes in electrical properties of the soil occur, part of the electroma gnetic energy is reflected back to the surface, where the r eceiving antenna amplifies the signal and stores the data (Radzevicius et al. 2000). The output is a measure of the strength of the

PAGE 18

7 electromagnetic wave, or amplitude. Ground-Pe netrating Radar is ab le to measure the time between wave propagation and reception, wh ich can be converted to a depth below the surface. In this fashion, amplitude valu es can be assigned a specific depth below the transmitting antenna. If the data collection process is repeated over a short distance, an image of the subsurface can be constructe d from the amplitude values and their associated depths stored and displayed by the GPR equipment (Knoll 1996). There are two main parameters to be entered by the user for GPR equipment: dielectric constant; and two-way travel time, or range. The depth of penetration for an antenna can be greatly augmented by proper determination and entry of these user parameters. Materials that are “dielectric” have a low dielectric constant and will permit electromagnetic energy to pass thro ugh without dissipation. The more electromagnetically conductive a material, the less dielectric it is. For maximum depth of penetration, material should be highly di electric with low el ectrical conductivity (Conyers and Goodman 1997). The range of the GPR is the amount of time the system will record received reflected signals after propagation (Geophys ical Survey Systems Inc. 1999). For example, a wave propagated at time 0 is pa rtially backscattered at 5 nanoseconds (ns) after transmission, and reaches the receiving antenna at 10 ns. The unscattered portion continues downward, is again partially backsc attered at 10 ns, and reaches the receiving antenna at 20 ns. If the range of the GPR equipment is set at 15 ns, only backscattering that reaches the receiving antenna in less than 15 ns will be recorded. As the range of the GPR is increased, more time is allowed for the wave to penetrate and reflect, thus increasing the total depth of penetration. Howe ver, a point occurs at which increasing the range no longer provides any usable da ta, because the entir e signal has been

PAGE 19

8 backscattered or lost. A gene ric example of GPR scattering is illustrated in Figure 2-3. The dielectric constant of a material is a physical property of the soil and the environmental conditions at the time the surv ey is conducted. It must be properly set during data collection to achie ve accurate results. Dielec tric constants of various materials are shown in Table 2-1. The diel ectric constant, dependant mostly on soil moisture, is a specific value that must be de termined or approximated at each site. A common method used to determine the dielec tric constant in the field is to bury a target at a given depth and position the ra dar antenna directly over the target. Then, adjust the dielectric constant of the GPR until the target appears at the correct depth in the GPR profile. In the absence of a target, contrasting soil horizons such as argillic horizons, can be used if their exact depth is known. Site Suitability The suitability of a site for GPR study is quite variable. While climate, geology, relief, and general feasibility pl ay a large role, the most important aspect of site suitability is the soil. Soils that have a high electri cal conductivity dissipat e the radar signal and do not provide a good medium for GPR studies. In general, the soil properties that affect the electrical conductivity of a soil are the cation exchange capacity (CEC), amount and type of clay, amount of dissolved salts, and moisture content (Dool ittle and Collins 1995) The CEC of a soil is the sum of the total exchangeable cations that can be held by a soil (Thomas et al. 1985). Since most ca tion reactions take pl ace on the surfaces of colloidal particles and organic matter, soils th at are high in clay or organic matter tend to have a higher CEC. Sandy soils typically ha ve a low CEC because coarse-textured soils are necessarily low in clay and organic ma tter. As the CEC of a soil increases, the suitability of the site for GPR drastically decrea ses, due to a decrease in depth of less than

PAGE 20

9 TransmitterReceiver Scattered Wave Energy Wave Energy Reflected to Receiving Antenna Fine Sand Sandy Loam Clay Figure 2-3. Ground-Penetrating Radar wave propagation, sca ttering and reflection. As a GPR wave encounters horizons of differi ng electrical propertie s, the wave is partially scattered. Parts of the b ackscattered energy are lost to the environment, but portions are also reflected back to the receiving GPR antenna. As more energy is scattered, less energy is available to penetrate into the ground. is the dielectric constant.

PAGE 21

10 Table 2-1. Dielectric consta nts of various materials MaterialDielectric constantMaterialDielectric constant Air1.0Wet Sandstone6.0 Snow1.5Wet Granite6.5 Dry Loamy/Clayey Soil2.5Travertine8.0 Dry Clay4.0Wet Limestone8.0 Dry Sands4.0Wet Basalt8.5 Ice4.0Tills11.0 Coal4.5Wet Concrete12.5 Asphalt5.0Volcanic Ash13.0 Dry Granite5.0Wet Sands15.0 Frozen Sand & Gravel5.0Wet Sandy Soils23.5 Dry Concrete5.5Dry Bauxite25.0 Dry Limestone5.5Saturated Sands25.0 Dry Sand & Gravel5.5Wet Clay27.0 Potash Ore5.5Peats61.5 Dry Mineral/Sandy Soil6.0Organic Soils64.0 Dry Salt6.0Sea Water81.0 Frozen Soil or Permafrost6.0Water81.0 less than 1 meter of penetrati on is possible (Olhoeft 1999). This is also the case with soils high in other smectites. Different clay minerals have vastly different CEC values. Typically, smectites have a high CEC and severely retard the dept h of GPR penetration. Smectites, which are 2:1 expanding phyllosilicates, exhibit a high CEC due to isomorphic substitution of Mg+2 for Al+3 in the dioctrahedral layer. This creates a negative charge that must be satisfied by cations. Smectites also have an inherently high internal surface area, allowing for more cationic exchanges. In c ontrast, soils with kao linite clay do not pose as much of a problem for GPR survey. Ka olonite is a 1:1 none xpanding phyllosilicate, does not undergo isomorphic substitution, and has a small surface area (Sumner 2000). Clay types and their CEC values are listed in Table 2-2. The amount of salts dissolved in the soil so lution is a major c ontributing factor to the electrical conductivity of a soil. Disso lved salts such as calcium carbonate and sodium chloride dramatically increase the electrical conductivity of soils (Doolittle and

PAGE 22

11 Table 2-2. Cation exchange capa cities of soil materials. Soil MaterialNet Negative Charge (cmolc/kg) Smectite80-120 Vermiculite100-180 Fine Mica 15-40 Chlorite15-40 Kaolinite2-5 Humus100-550 Collins 1995). The correlation is so significant that electrical conductivity is used as a measure of salt content for site assessment and the determination of saline and sodic soils (United States Department of Agriculture – Natural Resources Conservation Service 2001). Data Structure It is important to understand the fashion in which GPR data are organized in order to proceed with processing and quantitative anal ysis steps. A GPR profile is divided into “traces” and “samples”. A “trace” of data is made at a specific X and Y location each time the GPR generates an electromagnetic pulse A “sample” is a data point along the Z (depth) axis of each trace at an interval of depth. The GPR can either record 512 or 1024 samples per trace. Depth, X, Y, and amplit ude are all attributes associated with each “sample”. When filters are applied to GPR da ta, they can be applied to a group of traces or a selection of samples. A graphic re presentation of “traces” and “samples” is illustrated in Figure 2-4.

PAGE 23

12 Figure 2-4. Ground-Penetrating Radar traces and samples. A GPR profile (left) can be divided into traces and sa mples (right). In the el ectromagnetic waves (right) each black line represents a trace, and each red square represents a sample. A trace exists at the X, Y location at which the GPR sends an electromagnetic pulse. A defined nu mber of samples are taken for each trace, either 512 or 1024. Each sample contains X, Y, Z, and amplitude information. .

PAGE 24

13 CHAPTER 3 GEOGRAPHIC INFORMATION SYSTEMS (GIS) A GIS is a system of computer software, hardware and data, and personnel to help manipulate, analyze and present information that is tied to a spatial location (Ormsby et al. 2001). Real world features with spatial coordinates are represented by layers and may be displayed in combination with any other spatially referenced data. History Modern GIS began in the 1960s, with the help of emerging computer technology and technicians. However, the history of GIS is unique in that it developed nearly concurrently by separate rese arch teams from different lo cations and backgrounds. One of the earliest accounts of a computerized GIS is the Canada Geographic Information System (CGIS) developed in the early 1960s Meanwhile, the Harvard Laboratory for Computer Graphics and Spatial Analysis (HLCGSA) created their automated mapping application, SYMAP, which serv ed as the training ground for ma ny of the scientists that developed and created the precursors to the pop ular consumer GIS packages used today. At the same time, the Census Use Study (CUS) in New Haven, Connecticut was beginning “computer mapping experi ments” to use in the 1970 U. S. Census. (Mark et al. 1988). The CGIS was the first GIS to be used and began with The Federal Department of Agriculture land inventory describing agriculture, forest ry, wildlife, and recreation capabilities across the lower one-third of Ca nada. The materials used to create the digitized maps were the first of their kind, and designed specifically for the land

PAGE 25

14 inventory. The drum scanner used in th e project to digitize aerial photography, for example, is now in the National Museum of Science and Technology in Ottawa (Tomlinson 1988). The HLGCSA was important for the academic development and experience it offered the students. Though the laboratory was situated at Harvard University, collaborators from nearby institutions such as Yale University and the Massachusetts Institute of Technology contributed in the research and directiv es of the center. Because these institutions did not have an organized geography department, the early focus of the HLGCSA research was primarily directed to ward practical applic ations in landscape architecture, urban and regional plan ning, and resource management. To prepare for the 1970 U.S. Census, the newly created Census Advisory Committee on Small-Area-Data established th e CUS in 1966 to develop methods of computer mapping. The programmers associated with this project developed the method to incorporate topology, the spatial relations hips between connecting or adjacent vector features (Environmental Science Research Institute 2002), into m unicipal data that became the core of current spatial data fo rmats such as USGS Digital Line Graphs, Spatial Data Transfer Standard (SDTS), and th e polygon vector layers us ed in Arc/Info. While the 1960’s served as the decade of developmen t for GIS, the 1970s were years of “lateral diffusion” (Tomlinson 1988). More universities and government agencies became interested in the technology, expanding the user base of GIS. Adding to this user base was the Environmental Syst ems Research Institute (ESRI) in Redlands, California. Founded in 1969 by Jack Dangerm ond, formerly of the HLGCSA, ESRI has become arguably the most widely used GIS pa ckage in use. While ESRI was primarily a consulting group that happened to develop so ftware in the 1970s, computers became

PAGE 26

15 cheaper, smaller, and more powerful. By the time the mid-1980s arrived, users were “pounding at ESRI’s door” requesting their so ftware. In response, ESRI released ARC/INFO, the GIS package upon which ESRI’s current models are based. (Dangermond and Smith 1988). The success of ARC/INFO turned ESRI into the internationally known GIS developer and manufacturer that it is today. Georeferenced Data Raw data must be digitized and georefer enced to be incorporated into a GIS platform. Digitizing is the process of encodi ng geographic features in digital form as x, y, and z coordinates (Environmental Science Research Institute 2002). Digital formats may be either vector or raster Vector data represents information about the earth in the forms of points, lines, or polygons. For exam ple, a sinkhole location may be represented as a “point”, interstate hi ghways as “line”, and a lake as a “polygon”. Figure 3-1 illustrates the three types of vector data. Raster data is information about the earth represented by equal-sized cells. For exam ple, population density, elevation, and land use are frequently represented as raster data. Satellite and aerial photography is also considered raster data. Figure 3-2 is an exam ple of elevation repres ented as raster data. Georeferencing is the process of refere ncing real world da ta to geographic coordinates, such as longitude and lat itude. These known coordinates will place the feature at the correct location on the earths’ surface. Frequently, Global Positioning Systems (GPS) are used to correctly geor eference the datasets. Once correctly georeferenced, a dataset can be imported to a GIS to be part of any spatial analysis procedures. Furthermore, spatially referenced data can be analyzed for interpolation by using kriging and other geosta tistical methods (Ovalles 1988). Depending on the objective of a project, se veral GIS layers may be used in

PAGE 27

16 ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( Legend Interstate! (Cities Conservation Land 0225450 112.5Kilometers Figure 3-1. Three types of vector data: points, lines, and polygons. In this image, cities are represented as points, inters tates as lines, and conservation land as polygons.

PAGE 28

17 Elevation (m)Value High : 94.5 Low : -5.3 00.40.8 0.2KilometersElevation data from National Elevation Dataset Central Florida Grid Figure 3-2. Raster data consis ting of equal sized cells. Each cell contains a single value representing real world data. In th is case each cell contains the average elevation for the cell area. The elevat ion values are then color coded to enhance the visual display. The elevation data are a subset of the entire state.

PAGE 29

18 combination to create a visual output to help in the interpretation of the site. The state of Florida is unique in the amount and availability of free, digita l, georeferenced data easily importable to a GIS. For this reason, data which can be imported to a GIS becomes much more useful because it can be combined with other data from a wide array of scientific disciplines. For example, soil data can be instantly combined with elevation, land use, geology, vegetation, civic, and political in formation for an area of interest. File Formats Vector and raster data are stored in a vari ety of file formats. The two main types of vector files are shapefiles and coverage s. While coverages necessarily contain geographic projection informa tion, shapefiles do not. Covera ges also contain topology, i.e. the spatial relationships between connect ing or adjacent features. Shapefiles contain no topologic information. Raster data can be st ored in several formats, e.g. discrete of continuous rasters. Imagery su ch as satellite or aerial photography is usually stored in completely different formats than raster data representing elevation or landuse. JPEG, TIFF, and Mr. Sid are all commonly used file formats for imagery. Analysis Multiple analysis procedures can be performed on both raster and vector data entered into a GIS. Selection of spatial features by querying specific characteristics of the data, called attributes, is a widely used pr ocedure. For example, all well-drained soils can be selected and removed from a datase t containing soil mapping units for a county. Spatial analysis procedures commonly used include proximity analysis and geoprocessing steps. Spatial analysis pro cedures act upon the geogr aphy of a feature rather than its specific attributes. Proximity analysis, or buffering, is a frequently used spatial analysis procedure that creates a zone of specified distance around a vector feature

PAGE 30

19 of interest (Environmental Science Research Institute 2002). For example, proximity analysis can determine the features of a data set that are within one mile of a selected interstate. Geoprocessing step s use two or more themes to create a new theme. For example, a shapefile of soils for the state of Florida can be clipped by a Marion County boundary shapefile to create a new dataset containing soil data for only Marion County.

PAGE 31

20 CHAPTER 4 GLOBAL POSITIONING SYSTEMS (GPS) Global Positioning Systems (GPS) provides a link between GPR data and GIS. The Global Positioning System is composed of a constellation of 24 satellites that provide accurate spatial coordinates to us ers worldwide. Satellites are convenient because the user does not need to have line of sight to the satellite, only a clear view of the sky. The GPS system is owned and opera ted by the United States Department of Defense (DoD) and can be accessed and used by civilians for free (Trimble Navigation Limited 1998). To access the system, a GPR receiver must be used to communicate with the satellites. Initially, the DoD scrambled the GPS satellite signals, creating a large error. Only military GPS units were able to decode the e rror, called selective availability, and receive the decoded and more accurate measurement. The error associated with selective availability is approximately 100 meters. Duri ng the Gulf War, the United States military used GPS units to track troop and equipmen t movements. However, not enough military GPS units were available to satisfy demand, so civilian GPS units were employed. In order for these commercial units to achieve similar accuracy to the military units, selective availability was turned off. Afte r the Gulf War was finished, President Clinton ordered selective availability to be turned off permanently. This action allowed civilians to use the military satellites to receive accurate measurements anywhere in the world. Today, many users use handheld GPS receive rs to access the sate llites and receive spatial coordinates. While handheld receiver s are inexpensive and portable, their margin

PAGE 32

21 of error, particularly in the vertical plane, is too large fo r detailed research projects. Typically, the vertical error in a GPS meas urement is 1.5 to 3 times as great as the horizontal error. The Garmin etrex Vista™ handheld GPR receiver, for example, has a horizontal error of approximately 15 meters (GARMIN Corporation 2001). If a more accurate measurement is necessary, a GPS receiver capable of differential correction should be used. Differential correction is the precise measur ement of the relative positions of two receivers tracking the same GPS signal. One of the GPS receivers must be placed over a known coordinate, and is termed the base statio n. The base station determines what error is inherent to the satellite signal and transmits corrections to the other GPS receiver, called the rover. When using Real-Time Di fferential Correction (RTDC), the corrections are sent from a base station to a rover vi a FM radio signal. Ma ny base stations are maintained by government agencies such as the coast guard. Differential correction can also be accomplished by post-pr ocessing the GPS data gathered in the field. Base station corrections can be downloaded and applied to the rover data by matching the exact time of the measurements, which is sent by the sa tellite. By applying differential correction, whether in real-time or postprocessing, the error in measurement can be reduced to less than 1 meter (Trimble Navigation Limited 1998). The error in a GPS measurement can be at tributed to four sources: atmospheric delay, obstructions, mulitpath, and human error. As the satellite signal travels through the ionosphere and trophosphere the signal can be bounced. The time the signal takes to reach earth is altered by this bouncing, and is called atmospheric delay. Obstructions obscure satellite signals by bloc king the line of sight to the sky and can be large trees, buildings, or other solid objects. Multipathing oc curs when a satellite signal is reflected

PAGE 33

22 from an object, such as a building wall, and is still received by the GPS receiver. Human error occurs in equipment configuration, e quipment setup, and equipment use (Trimble Navigation Limited 1998).

PAGE 34

23 CHAPTER 5 FIELD SITE AT PINE ACRES The research site used for this project is located at the University of Florida Institute of Food and Agricultural Science Plant Science Research and Education Unit located near Citra, FL in Marion County, a nd commonly referred to as Pine Acres. The facility contains over 400 hectares of land for university research in plant science, turfgrass science, soil science, agronomy, a nd other related disciplin es. The location of Pine Acres is shown in Figure 5-1, the specifi c plot in 5-2, and the coordinates in Table 5-1. Soils The site used for this research is part of a larger plant scie nce research facility, and detailed information about the soil has been gathered in recent years. Properties of the soil determined by mapping and laboratory an alyses show the soils to have suitable properties for GPR use (Doolittle and Co llins 1995). These properties include • The texture of the non-argillic horiz on soil is very sandy, with very little clay content (Thomas et al. 1979). In several places, a fine sand texture exists for several meters. • The soil morphology is distinguishable by use of GPR. Argillic horizons can be separated from overlying sa ndy horizons when proper conditions exist (Doolittle and Collins 1995). • The water table is quite deep at this location in Marion County, in part due to the drought of recent years. In this condition, the water table cannot be confused with a soil horizon. • The amount of dissolved salts is low in these soils. Fortunately, a large scale soil survey ha d been completed on the Pine Acres

PAGE 35

24 [ Gainesville Research Center 0230460 115Kilometers Figure 5-1. Location of research site in Ma rion County, Florida. Pine Acres is located in northern Marion County, south of Orange Lake.

PAGE 36

25 0.500.51Miles Research Plot N E W S 385500 385500 386000 386000 386500 386500 387000 387000 387500 387500 388000 388000 3252500 3252500 3253000 3253000 3253500 3253500 3254000 3254000 3254500 3254500 Figure 5-2. Research plot at Pine Acres. Located in field 5a, the research plot was approximately 320 meters 160 meters.

PAGE 37

26 Table 5-1. Corner coordinates of the research plot. The projection system used was UTM, NAD 1983, Zone 17 North. CornerXY Southeast3866663253871 Southwest3869883253875 Northeast3866673254032 Northwest3869903254030 property. Alfisols, Ultisols, and Entisols occu r on the property. The argillic horizon, as in Figure 5-3, is quite variable in its depth, texture, and ba se saturation, providing for the presence of both Alfisols and Ultisols (M.E Collins, personal communication, 2001). In areas where no argillic horizon is present, En tisols exist, due to the lack of other diagnostic subsurface horizons. Soil profile descriptions taken from the study area are located in the Appendix. The site used for this research is located in the Arredondo-Sparr-Tavares Association, Figure 5-4, which is characterized by nearly le vel to sloping soils, and sandy to a depth of less than 100 cm. Arredondo is a well-drained loamy, siliceous, semiactive, hyperthermic Grossarenic Paleudult, Spa rr is a somewhat poorly drained loamy, siliceous, subactive, hyperthermic Aquic Aren ic Paleudult, and Tavares is a moderately well drained hyperthermic, uncoated Typic Qu artzipsamment. Depending on drainage and the presence of an argillic horizon, the soils at the fiel d in which the GPR transects were collected are Sparr, Millhopper, or Adamsville. Millhopper soils are loamy, siliceous, semiactive, hyperthermic Grossa renic Paleudults, and Adamsville soils are hyperthermic, uncoated Aquic Quartzipsa mments (Thomas et al. 1979). Geology

PAGE 38

27 Figure 5-3. Argillic horizon at Pine Acres Research Facility. The presence of an argillic horizon in this profil e places this soil in the Ultisols or Alfisols soil order, depending on the base saturation of th e soil. The red line indicates the approximate top of the argillic horizon.

PAGE 39

28 [ Research Site General Soil Association Map Unit Names ARENTS-MATLACHA-HYDRAQUENTS ARREDONDO-SPARR-TAVARES BLICHTON-FLEMINGTON-KANAPAHA CANDLER-ASTATULA-TAVARES CANDLER-TAVARES-ASTATULA FELDA-CHOBEE-KALIGA FLORIDANA-RIVIERA-TERRA HYDRAQUENTS-UDORTHENTS-ARENTS MILLHOPPER-SPARR-LOCHLOOSA POMONA-EAUGALLIE-MALABAR POMONA-MYAKKA-WAUCHULA SAMSULA-HONTOON-EVERGLADES SMYRNA-IMMOKALEE-BASINGER SMYRNA-MYAKKA-IMMOKALEE TAVARES-ZOLFO-SATELLITE TERRA CEIA WABASSO-FELDA-PINEDA WATER 02550 12.5Kilometers Figure 5-4 General Soil Associations – Marion County, Florida. The Pine Acres research site is located on the Arredondo-Sparr-Tavares Association (Thomas et al. 1979).

PAGE 40

29 The geology of Marion County has been well documented and mapped. Sands deposited in the Pleistocene epoch cap th e Hawthorn Group Formation (HGF), a clayrich Plio-Miocene-age deposit. In many ar eas the Ocala Limestone of Eocene age is present directly beneath the HGF. In many cases the argillic horizons present in soils of this area are the upper portion of the HGF, fo rming a lithologic discontinuity with the Pleistocene sands. Frequently, phosphate nodules are present at the interface betw een the Pleistocene sands and the HGF. Ranging in size from less than 1 cm to over 50 cm in diameter, the phosphate nodules are sand grains cemented together by phosphatic minerals, primarily wavellite (Harris 2002). Though large phosphate nodules are frequently identified at Pine Acres, none were encountered at the specific site of research. The HGF consists of mixed lithology. Consisting of interbedded phosphatic clay, sand, dolomite, and limestone, the formation is frequently referred to as undifferentiated in origin. The primary diagnostic feature of the HGF is the presence of phosphatic material in the sediment (Lane and Hoens tine 1991). Frequentl y, the HGF exhibits a blue-gray hue in the field, differentiati ng it from a pedogenic argillic horizon. The nature of the geology and the soils provide an excellent opportunity for GPR research. The clayey sediments contrast th e overlying sands. Clay content increases by as much as 20% or more acro ss the interface. Furthermore, the water holding capacity of the clay is much higher than the coarse-tex tured sand (Thomas et al. 1979). In addition, smectite is a typical component in the clay fraction of the HGF. While higher water content and presence of smectite do not provide great depth of penetration, they provide a significant contrast in electromagnetic prope rties from the overlying coarse-textured pleistocene sand material.

PAGE 41

30 Climate Climatic conditions of the past few year s in Marion County are a departure from normal, especially with regard to hydrologi c issues. The Modified Palmer Drought Index (MPDI) in Figure 5-5 illustrates the climatic pa ttern over the last 4 y ears. The MPDI is a modification of the Palmer Drought Index indica ting the severity of a wet or dry spell. This index is based on the principles of a balance between moisture supply and demand. The index generally ranges from -6 to +6, with negative values denoting dry spells and positive values indicating wet spells. PMDI va lue ranges can be interpreted to mean 0 to -0.5 = normal; -0.5 to -1.0 = incipient drought; -1.0 to -2.0 = mild drought; -2.0 to -3.0 = moderate drought; -3.0 to -4.0 = severe drough t; and greater than 4.0 = extreme drought (National Oceanic and Atmos pheric Association 1994). As illustrated in Figure 5-5, Marion County has been under drought conditions for most of the past 4 years. Though the MPDI does indicate relatively normal cond itions for the past 12 months, the climatic impact on the soils at Pine Acres is still one of drought. The precipitation pattern, represented by the Palmer Z index of precipitation (PZP) in Figure 5-6, of the Pine Acres area for the last 4 years provides more data to support the droughty conditions experienced in the field. The PZP is a monthly generated value that can be expresse d as the "Moisture Anomaly Index." Each monthly Z value is a measure of th e departure from normal of the moisture climate for that month. This index can res pond to a month of above-normal precipitation, even during periods of drought (National Oceanic and Atmosp heric Association 1994). Over the course of the GPR investigation, se veral auger holes were dug to a depth of 2 meters of more. In no instance was a near surface or perched wate r table encountered. This absence of a soil water table in the re search area provided le ss interference for the

PAGE 42

31 GPR signal. -6.000 -5.000 -4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 3.000 4.000 5.000 6.000 7.0001/1998 3/1 9 9 8 5/1998 7/1 9 9 8 9 / 1 9 9 8 1 1 / 19 9 8 1 /1999 3/1999 5/1 9 9 9 7/1999 9/1 9 9 9 1 1 / 1 999 1 / 2 0 0 0 3 /2000 5/2000 7 /2000 9/2000 11/ 2 000 1 / 2 0 0 1 3 / 2 001 5 /2001 7/2001 9 /2001 11/20 0 1 1/2 0 0 2 3 / 2 0 0 2 5 / 2 002Month and YearIndex Value Figure 5-5. Modified Palmer Drought Index. This data represents the MPDI state division number 3 for the stat e of Florida from 1998 to 2002. Topography The amount of relief at the Pine Acre s property is minimal. Agricultural production in past years has produced a fairly flat to slightly rolling landscape. However, topographic changes are present in microrelief, slight variations in the height of a land surface that are too small or intricate to delineate on a topographic or soils map at commonly used map scales (Soil Survey Di vision Staff 1993). The microrelief is particularly pronounced in areas of kars t activity, such as sinkholes.

PAGE 43

32 -7.000 -6.000 -5.000 -4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 3.000 4.000 5.000 6.000 7.000 8.0001/ 1998 3/ 1998 5/ 1998 7/ 1998 9/1998 11/1998 1 / 1999 3/ 1999 5/ 1999 7/ 1999 9/ 1999 1 1 / 1999 1/2000 3 / 2000 5/ 2000 7/2000 9/ 2000 11/ 2000 1/2001 3/2001 5 / 2001 7/ 2001 9/2001 11/ 2001 1/ 2002 3/2002 5/2002Month and YearZ Index Value Figure 5-6. Palmer Z Index of precipitation. From 1998 to 2002, the majority of Z Index values were below 0, indicating a water deficit for each month. This data represents state division number 3 for the state of Florida from 1998 to 2002. .

PAGE 44

33 CHAPTER 6 MATERIALS AND METHODS Data Collection The GPR transects at Pine Acres were take n in a grid formation, 40 meters apart. The extent of the grid was 160 m 320 m. Ei ght transects were coll ected south to north, and four transects were collected east to west Four additional transects were collected from the corners of the grid to the center of the opposite baseline. The GPR used was a SIR-2000 control unit (Geophysical Survey Systems, Incorporated, North Salem, NH) with a 300 MHz ground-coupled monostatic antenna. To collect the transects, the antenna was pulled behind a field vehicle running at approximately 5 miles per hour. The range wa s set at 60 nanoseconds, with a dielectric constant of 5. The dielectric constant was estimated by calibrating the GPR to a known depth to the argillic horizon. Descriptions of soil profiles located within the study area are located in the Appendix. GPS measurements were taken with a Pathfinder series differerential GPS Receiver (Trimble Navigation, Ltd., Sunnyvale, CA). By using real-time differential correction when collecting GPS measurements the horizontal error of measurement was decreased to approximately 1 meter. Several software packages were used in the manipulation and calculation of GPR, GIS, and GPS data. ReflexW 2.5.x (Sandmeier Scientific Software, Karlsruhe, Germany) was used for all GPR post-processing. Tabular manipulation and calculation was completed in SPSS 11.0 (SPSS, Inc., Chicago). The primary GIS software packages used

PAGE 45

34 were ArcGIS 8.1 and ArcGIS 8.2 (Environmenta l Systems Research Institute, Redlands, CA). All raster interpolations were complete d within ArcGIS with the use of the Spatial Analyst, 3D Analyst, and Geostatistic al Analyst extensions. RockWorks 2002 (Rockware, Inc., Golden, CO) was the seconda ry GIS software package employed for 3D solid modeling. Variogram analysis was comp leted in S-Plus 6.0 (Insightful Corporation, Seattle, WA) using the Spatial module. G PS data was imported and converted to GIS vector data with GPS Pathfinde r Office (Trimbe Navigation, Ltd.) Ground-Penetrating Radar Post-Processing “There is no single processing, visuali zation, and interpretiv e strategy that is applicable to all datasets.” —Dr. Alan Green, 2002 International C onference on Ground-Penetrating Radar The data generated from this research are an exemplary case of the quote above. The desired result was an easily understood 3D soil model representing only two main horizons: E or Bw and argillic. The nome nclature and designation of the sandy nonargillic horizon is subjective, depending upon th e soil scientists describing the site and their theory of soil formation. Whether the horizon is called an E or Bw is irrelevant to GPR data collection. In this study the fo cus was not on soil classification but the electrical and textural propert ies of the underlying soils. The electrical and textural properties of the soil are physical and do not change with the taxonomy or name of the soil or diagnostic horizons. The post-processing steps used to prepare the GPR data for the model generated an easily understood model, but are not reco mmended for studies in which accuracy and precision are integral to the succ ess of the project. At most, the data were processed by six ordered procedures before they we re exported to the modeling software. A linear correction to the GPR profile was the first process completed. The linear

PAGE 46

35 correction step is needed because the GPR an tenna cannot be pulled at a constant speed across a transect. If the antenna could be pulled at a known constant speed, distances could be calculated from the elapsed time from start for a given point along each transect. However, the GPR antenna is pulled at a variab le rate. To correct for this inconsistancy, intermediary “marks” were created at 40 meter intervals along each GPR transect line when collecting data. These marks can be displayed in the GPR software. Knowing the distance between intervals allows the GPR soft ware to stretch or compress the sections between the interval marks to create the correct distance. Next, the GPR profiles were corrected to the proper surface depth. This step was performed early because radar features typically removed or augmented by other processing steps are used in determining the co rrect depth to the soil surface. In the GPR analysis software, time 0 ns is assumed to be equivalent to a depth of 0 meters, interpreted as the soil surface. However, th e actual soil surface is typically at a depth equivalent to a time value greater than 0 ns. This phenomenon occurs because of the architecture of the antenna, as illustrated in Figure 6-1. An example GPR profile before and after the depth correction has been applie d is illustrated in Figures 6-2 and 6-3. The wave is propagated from the transmitting an tenna inside the fibe rglass housing. But, while the transmitting antenna is located inside the fiberglass shell, it is at some distance above the surface of the earth. This allows the GPR wave to begin propagation through the air sealed within the fi berglass housing, yielding a very signature uniform reflection on the GPR output image. The problem beco mes more serious with lower frequency antennas that are physically larger. As the size of the housing increa ses, the antenna is removed further from the soil surface. Since the wave is traveling through a comparatively homogeneous material (air) th e GPR reflection is quite uniform, in

PAGE 47

36 contrast to the undulating and varied reflections seen as GPR waves travel through the soil. The interface between the uniform air reflection and non-uniform soil reflection can be easily identified and corrected in the pos t-processing software. In the GPR profiles associated with the research, the soil surface was at a depth equivalent to 6 ns. Soil Surface Soil Surface GPR Antenna GPR antenna housing Air filled interior GPR wave A B Figure 6-1. Simplified interior of a GPR antenna housing unit. The actual GPR antenna is located at some point abov e the actual soil surface. At point A indicated in the figure, the GPR wave be gins propogation. This point is also designated time 0 by the GPR unit, and becomes the perceived soil surface. However, the actual soil surface is lo cated at point B, a known time later than point A. The radar wave travels through the uniform air interior of the antenna and antenna housing before reaching the actual soil surface. Analysis software was used to correctly identify the actual soil surface. The picture is not drawn to scale. The ac tivity of the GPR wave is not intended to be accurate in wavelength, frequency, or amplitude. After the soil surface is corre ctly determined, a background filter, as in Figure 6-4, is applied to all transects. Typically, the upper part of the GPR profile contains fairly uniform horizontal bands that are of no fu rther interpretive use and are attributed to radar attenuation or noise. By applyi ng the background filter, the bands are removed allowing the anlyst a better view of the featur es within the GPR profile. The filter works

PAGE 48

37 Actual soil surface, depth 0 Figure 6-2. Uncorrected depth to surface GPR profile. This profile has not been corrected to indicate th e correct depth to soil su rface. The actual soil surface is indicated by the arrow on the ri ght hand side of the image. The yellow and orange bands at the top of the GPR profile aid in identifying the correct surface depth. Because the GPR wave is traveling through static air inside the antenna housing before entry to the soil, the GPR will show very uniform horizontal bands while wave is still traveling thr ough air. Once the wave enters the soil, the signal b ecomes much less uniform and more variable. The soil surface exists at th e interface between the uniform orange band and variable yellow band in this picture. Corrected soil surface, depth 0 Figure 6-3. Corrected depth to surface GPR prof ile. This figure illustrates a GPR profile corrected to indicate the actual soil surface. In this case, there was a time difference of 6 nanoseconds between th e perceived soil surface and actual soil surface. The correct soil surface is indicated by the 0.0 depth on the right hand axis.

PAGE 49

38 on each profile individually. The software determines the time and amplitude range of the radar “noise” that appear continuously throughout the pr ofile. That noise is then subtracted from the traces within the GP R profile. In this fashion, horizontally continuous energy patterns and radar attenuati on features are removed (Sandmeier 2002). It is important to note that this does not remove horizontall y continuous soil features such as pedogenic horizons. The background filter is a valuable tool able to visualize soil features that may be initia lly hidden or suppressed by rada r features of no interpretive use. The fourth processing step applied was a spectral whitening filter, illustrated in Figure 6-5. In order to accurately model th e argillic horizon from GPR data, the division between the sand and argillic horizon had to be very dist inct. The spectral whitening filter accentuated and highlighted the argillic horizon, while further defining the sand-argillic interface. Th e spectral whitening filter works by applying a series of complex algorithms to the GPR profile invol ving narrow overlapping band-pass filters, amplitude decay curves, and spectral flattening procedures (Sandmeier 2002). This filter helps correct the scattering of the GPR signal deeper in the profile. Again, while these processing steps were ideal for this research, th ey may not applicable to all datasets. The next processing step was employed primarily to speed calculations and computation of the data. The sheer volume of data was so immense that further compression of the data was absolutely necessary to accomplish modeling in a timely fashion. A compression rate of 10x in both the X and Z directions wa s decided upon after a trial and error period of various compressions rates. The 10x rate retained an acce ptable level of resolution while decreasing the data to a manageable size, as seen in Figure 6-6. After compression, a data observation was recorded every 8 cm in the x-di rection, and 10 cm in the

PAGE 50

39 z-direction. After the data were compresse d, they were ready to be georeferenced. Figure 6-4 – Background filter of GPR profile. This illustration depicts the process of background filtering, as applied in this re search. In the t op profile, several horizontal bands mask the underlying information. By applying a background filter, a clearer picture of the subsurface becomes available in the bottom profile. The starting and ending points of each GPR transect were georeferenced in the field using differential GPS. However, the process of actually geor eferencing the entire transect occurred in the GPR analysis software Real world coordina tes were entered into the file header for each GPR transect as the start and end locations. The relative X and Y coordinates of the file then became absolute x and y coordinates, placeable in the real world. The direction in which the transect was run (x or y) was also indicated in the file header. Since linear correction was already performed, as soon as a real world coordinate was entered into the file header, the entire tr ansect became georeferenced. This approach will not work for coordinate systems in which angles of latitude and longitude are the chosen unit of measurement.

PAGE 51

40 Befor e After Figure 6-5. Spectral whitening filter of GP R profile. The Spectral whitening filter was applied to accentuate and highlight th e argillic horizon in many profiles. The filtered output made differentiation of the horizons much easier and more efficient to model. A scaling factor of approximately 10 was used when applying this filter.analysis de manded compression of the data. A raw data file recorded a data observati on every 8 mm in the x-direction and every 1 cm in the z-direction. A sing le profile of 320 m contained about 2 million records. All data were exported as an ASCII text file, with four columns. Each row represented a single observation, with X,Y, Z, and amplitude values, respectively. The data were exported in this format to ease tr ansfer to statistical a nd 3D modeling software packages. Each transect was exported separa tely, creating 18 different tables of data. Once the data were correctly analyzed and exported, preparation for 3D model generation could begin. Model Generation Three specific goals were to be obtaine d from the generation of a 3D model. First, the model must be easily interprete d. A person with average knowledge of GPR has difficulty comprehending and interpreting 3D GPR models. Second, the model must represent the soil features to an acceptab le degree of accuracy. Third, and most

PAGE 52

41 Before After Figure 6-6. Compression of GP R profile. A compression procedure was necessary to minimize the file size of each transect A compression rate of 10x in the X and Z directions, as in this figure, pr ovided a compromise between file size and resolution. important, the data must be transferable to a GIS software package for display. These objectives helped to define the modeling pr ocedures and methods for the post-processed GPR data. Four different models were created us ing GPR data. Model 1 consists of categorical data that predict the presence of sand or an argillic horizon based on GPR amplitude readings. Model 2 is a numerical raster model that predicted GPR amplitude values continuously throughout the field. Mode l 3 and Model 4 are bot h numerical raster models that predict the depth to argillic hor izon, but differ in their methods of generation. The GPR processing steps taken, specific obj ectives, and the quantitative analysis performed on the data, differentiate all the models. Data Preparation The four column ASCII data exported fr om each GPR profile were first imported to SPSS 11.0 Statistical Software for Windows us ing the proprietary import procedure.

PAGE 53

42 Each transect contained approximately 15,000 poi nts of data. A fifth column was added to indicate the file of origin. However, in or der to create the 3D model, all the GPR data needed to be part of the same table, as in Table 6-1. Using SPSS, the 18 separate GPR profile tables were combined into one la rge table containing the data from all GPR profiles. This table consisted of approxi mately 260,000 points of data. Once the data were combined to a common table, a dept h conversion was performed on the time column (Zns) of the table. This step was n ecessary to create a depth scale (meters) rather than a less intuitive time scale (ns). An exampl e of the data at this point is displayed in Table 6-1. The values in the Zns column were conve rted from time to depth using equation 6-1. This conversion assumes an average ve locity through the soil medium. The average velocity was calculated from the dielectric cons tant used while collecting the data in the field. Equation 6-2 was used to determine the velocity of the medium for use in the depth conversion. By using both equations, the z dime nsion is no longer a time scale (ns) but a depth scale (meters). D = tp (0.15 (1/ ) (6-1) Where D is the maximum depth of GPR signal penetrations, tp is the two-way travel time, in nanoseconds, is the dielectric constant (Collins 1990) v = c / (6-2) Where v is the velocity of the wave in meters /s, c is the speed of light in a vacuum, 3 x 108 m/s, and is the dielectric c onstant (Olhoeft 2000) In addition to creating a column for de pth, a column containing the relative amplitude difference, or change in amplitude was created. For each trace, the amplitude change, A, is the absolute difference in amp litude between neighboring samples. The

PAGE 54

43 Table 6-1. Example of the tabular data befo re any statistical mani pulation. X and Y are the horizontal real world coordinates fo r each point of data, Zns is the timedepth of the data point in nanoseconds, and Amplitude is the amplitude value as exported from ReflexW. FileXYZnsAmplitude 2689386786.8443253873.750.8361899 2.2031981 3.57-7260 4.938-9854 6.305-7353 7.672-4281 9.039-6703 10.406-14475 11.7732005 13.141-5364 14.5084375 15.875-21584 17.242-9853 18.609-32760 19.977-32760 21.344-32760 22.711-32760 24.078-32760 calculation of amplitude difference intended to capture the rapidity of GPR signal change when the argillic horizon was encountere d. Figures 6-7 shows the distribution of A with depth and the relativity to the GPR image in Figure 6-8. It is important to note at what depth large values of A begin to appear. Only at de pths deeper than the expected depths to argillic, approximately 1.5 m, do A values greater than 30,000 appear.Based on the assumption that changes in soil morphology occur concurrent with large A values, A became the modeling variable. However, A is not a value with which average GPR users are familiar.

PAGE 55

44 0200004000060000delta_a 0 20 40 60d e p t h n s Figure 6-7. Scatterplot of A versus depth for all GPR data. An example illustrating the distribution of A with depth for a selected GPR profile. Each point represents one sample of processed GPR data. The Y axis is the time below the surface in nanoseconds, and the X axis is A. Notice that large values of A (> 30,000) do not occur at less than 20 nanoseconds below the surface. The expected interface betw een the sand and argillic horizon is between 20 and 40 nanoseconds below the surface. A was used for model generation because it captures the inte rface between the sand and argillic horizon. The two spikes in A above and below 0 nanoseconds are residual noise left from background filtering and depth conversions, and are seen in Figure 6-8.

PAGE 56

45 Figure 6-8. Ground-Penetrating Radar profile image described by Figure 6-7. In this profile the alternating lavender and dark blue lines represent the argillic horizon from a depth of approxi mately 1.5 meters to greater than 4 meters. The interface is located be tween 1.5 and 2.75 meters. The depth distribution of large A values in Figure 6-7 qua ntifies this relationship. .

PAGE 57

46 CHAPTER 7 RESULTS AND DISCUSSION Variograms and Kriging To create a model of the argi llic horizon at Pine Acres, de pth to argillic values for areas between data points coll ected by GPR needed to be estimated. The procedure used to fill in these values, or interpolate, was ordinary kriging. Kriging is a family of interpolation procedures that use both the distance and degree of variation between known data points to estimate new values. Krigin g also provides an estimation of error at each interpolated point, creating a measure of confidence in the model. Ordinary kriging, the most popular kriging method, assumes the mean of the data is unknown. The ordinary kriging procedure made use of variograms that were calculated for each subset used in Model 2. A variogr am summarizes the relationship between differences in pairs of measurements and the distance of the co rresponding points from each other. A variogram is constructed by plotting the semivariance, or squared difference in value, between every pair of poi nts in a dataset against the distance the two points are apart from each other in space. (Wackernagel 1995). For GPR data, data points physically lo cated close to each other should have A values that are numerically close. As co mparative points become spaced further apart, their differences in value will be greater. At some point, however, the increase in distance no longer causes an increase in th e semivariance, and the variogram graph plateaus. The spatial distance at which this occu rs is called the range. It is assumed that beyond the range, the data does not have a ny autocorrelation. Au tocorrelation is the

PAGE 58

47 degree to which a variable is correlated with itself. The variogram value (semivariance) at the range distance is called the sill. Theoretically, at a distance of zero the semivariance is zero. And as distances approach zero, the semivariance should also approach zero. This is rarely the case. Sa mpling error and short-scale variability cause dissimilarities in values that are separated by ve ry small distances. This causes a vertical jump at the origin of a variogram, and is called the nugget-effect (Is aaks and Srivastava 1990). Knowing the sill, range, and nugget al lows for a more accurate interpolation procedure in 3D modeling. Th e range, sill, and nugget are illustrated in Figure 7-1. range: 1.450633e+001 sill: 1.120933e+008 nugget: 6.227408e+006Variogram at 250 cm, 37.31 ns 050100150 distance 04000000080000000120000000 gamma objective = 4.292475e+015 Sill Range Nugget Figure 7-1 Range, sill, and nugget value for a variogram. The sill value is the semivariance, gamma, at the point at which the variogram graph plateaus. The range value is the distance at wh ich the variogram plateaus. The nugget is the Y-intercept value at extremely small distances. To produce variograms for this datase t, the data was stratified by depth. variogram graphs of the GPR data were cons tructed at approximately 10 cm intervals for

PAGE 59

48 the research site. A values were plotted against the dist ance in y for x,z pairs. Figure 7-2 show the variogram graphs for sele cted depths of the survey area. Figure 7-2. Variograms of GPR da ta at different depths. Re d dots are indicative of low A, and blue dots are indicative of high A. N is the nugget value, S is the sill value, and R is the range value. The x-axis is lag distance, and the yaxis is semivariance. A) Variogram of GPR data at 30 cm. B) Variogram of GPR data at 50 cm. Variograms of GP R data at different depths. C) Variogram of GPR data at 73 cm. D) Variogram of GPR data at 115 cm. E) Variogram of GPR data at 145 cm. F) Variogram of GPR data at 177 cm. G) Variogram of GPR data at 250 cm. H) Variogram of GPR data at 302 cm. A B N=70253232 S=65180241 R=17.13 N=53093394 S=17317353 R=16.65

PAGE 60

49 Figure 7-2. Continued C D N=55680609 S=15164752 R=30.56 N=124243756 S=69140121 R=43.74

PAGE 61

50 Figure 7-2. Continued E F N=273107206 S=78817138 R=57.26 N=566319393 S=58282864 R=49.62

PAGE 62

51 Figure 7-2. Continued G H N=666772977 S=192783837 R=2.642 N=791339195 S=68515194 R = 3.843

PAGE 63

52 The variograms of the GPR data have pa rticular features in common. First, a definite pattern in fluctuation of covariance exists at distances greater than the range. This cyclicity, or hole effect, may be due to heterogeneity of the soil material, and is particularly evident in the va riograms with depth less than 100 cm. Second, the ranges of the variograms show a steady increase range with depth followed by a sharp decline in range, as illustrated in Figure 7-3. The range of the variogram steadily increased from the surface down to approximately 150 cm, at which depth the range drastically decreased. The depth of 150 cm correlates with the presence of a shallow argillic horizon, suggesting that autocorrelation is sm all within the argill ic horizon. More interestingly, the distance for which autoco rrelation exists increases from the surface downward to the top of the argillic horiz on, suggesting an increased homogeneity with depth, until the argillic horizon is reached. The factor of anisotropy must also be considered when interpolating between points using ordinary kriging. Anisotropy is a directional influence on the data. For example, if a pollutant is dumped into a st ream, the change in concentration of the pollutant would be greater from streambank to streambank than upstream to downstream, due to the flow of the water. The geostatistic al analyst has the ability to account for such anisotropy when performing ordinary kriging calculations. Anisot ropy is statistically manifested within a directional variogram, a variogram that looks in a specific direction, rather than globally. If the variogram ch anges rapidly when di rection is changed, anisotropy exists. If no dramatic changes in the variogram exist, the data is isotropic, and presents no directional influence. Modeling Parameters When modeling A or depth to argillic, specific parameters where entered to

PAGE 64

53 Depth -vsRange 0 10 20 30 40 50 60 70 05010015020025030035 0 Depth (cm)Range (cm) Figure 7-3. Depth versus Range for several va riograms. When the range of a variogram is plotted against the depth at which the data points were collected, an interesting pattern emerges. The rang e of the variogram steadily increased to approximately 150 cm. At deeper depths, the range of the variogram decreases sharply. The depth of 150 cm corresponds with shallow occurances of the argillic horizon. The pattern of the range values may be due to increased homogeneity of the soil material between the surface and the top of the argillic horiz on, and heterogeneity within the argillic horizon. accurately predict or estimate values, or to achie ve a desired result. Model 1 used a K-means clustering technique to group the data A K-means cluster analysis designates a value for each cluster center based on the spread and distribution of th e dataset. The user specifies the exact number of clusters to gr oup the data. For a two cluster example, each data point is placed into one of two groups ba sed on the numerical distance of the specific value from the cluster center. As a result, each data point is given an additional attribute indicating its membership in a cluster. M odel 1 used the closest point interpolation procedure for solid modeling available in Ro ckWorks 2002 for its ability to retain the

PAGE 65

54 integer classification values. Models 2, 3, and 4 used ordinary kriging to interpolate between data points. The geostatistical extension of ArcGIS was used to calculate kriged raster surfaces. A logarithmic transformation was performed on th e data to normalize the distribution of the data before modeling. A consta nt order trend was removed from the data to allow a more global interpolation scheme. When evident, anisotropy was accounted for in the modeling procedure. Model 1 To continue with the goal that the model be easily interpretable, a classification procedure needed to be employed to separa te the data. This method resulted in the production of Model 1, a categoric al model predicting the pres ence of an argillic horizon. Rather than having amplitude or A with their wide ranges in value as the primary variable to be modeled, a number indicating a data point as belonging to either a sand horizon or argillic horizon was desirable. A K-mean s clustering technique was performed on the A values to classify each data point into two groups: sand and argillic horizon. Figure 7-4 shows that points with a me mbership in cluster 1 are located closer to the surface than points with a membership in cluster 2. An example of data that has gone through all the tabular manipulations a nd is ready for the modeling procedure is shown in Table 5. The construction of the 3D model bega n after the data ha d been analyzed, augmented, and clustered. RockWare 2002 (Roc kware, Inc., Golden, CO) was used to create a solid 3D model fr om the GPR table containing over 260,000 data points. The table was saved in SPSS in *.dbf format and imported to RockWare 2002 in order to comply with the import procedure. At this point, no more tabular manipulations were

PAGE 66

55 Figure 7-4. Mean depth of cl ustered variables. The mean depth values for the two cluster classes illustrate their relationshi p to the soils in the field. Cluster 1, representing sand, has a mean depth value much nearer to the surface than Cluster 2, representing the argillic horizon. performed. The data were rea dy to become a usable model. RockWare 2002 creates 3D solid models fr om X, Y, Z, and attribute data by creating a 3D grid of equal-sized 3D cells cal led voxels. The dimensions of the voxels can be changed by the user for optimal reso lution. The voxel centers, or nodes, all have specific coordinates as well as attribute values. The closest point interpolat ion method was used to create the 3D solid model from the cluster membership values. This c hoice was very important to the output of the model. If a different model was used, the cl uster values would not be retained as an Mean depth cluster 1 1.71 meters Mean depth cluster 2 2.61 meters

PAGE 67

56 Table 7-1. Tabular GPR data ready for 3D modeling. FILE designates the original GPR file, X, and Y are the absolute GPR coordinates, Z_NS is the depth in nanoseconds for each point, AMP is the amplitude value for each point, AMPDIFF is A, Z_M is the depth in meters and CLUSTER is the cluster membership value. To create a dummy variable for multivariate analysis, subtract 1 from the cluster value. integer. For example, a K-means cluster va lue of 0 suggests the data point belonged to the sand cluster, and a value of 1 suggest s the data point bel onged to the argillic \horizon cluster. With this numbering c onvention, a decimal value such as 0.65 would have no connection to the real world; the va lue 0.65 is neither a member of the sand cluster nor the argillic horiz on cluster. It was impera tive that an interpolation procedure be used that retained the integer nature of the cluster value. The only method to provide this function was the closest point algorithm. The closest point algorithm assigns a value to each voxel that is equal to the nearest data point. Since every data point used in the model was an integer, the on ly possible voxel values would be integers. FILE X YZ_NS A MP A MPDIFFZ_MCLUSTER 2680386989.63253991.00-4.63-660.311 -3.279150.22 -1.9-5140.13 -0.539140.04 0.842112-0.06 2.2522501-0.15 3.57796274-0.24 4.9433763-0.33 6.324372404-0.42 7.67-3682805-0.51 9.0417332101-0.61 10.415201213-0.7 11.771188668-0.79 13.14-6191807-0.88 14.516591278-0.97 15.88-79288587-1.07 17.24-40763852-1.16 18.61-66442568-1.25 19.98632512969-1.34 21.34-23048629-1.43 22.717173021-1.52 24.08-3208832805-1.62 25.45531037398-1.71 26.813276027450-1.8

PAGE 68

57 The end result was a solid model with a value of 0 or 1 for each voxel, representing sand or argillic horizon, respectively. The closest point method of 3D solid modeling created a model capable of interactive viewing in RockWorks, Figure 7-5. However, the model still needed to be integrated to a more popular GIS package, such as ArcGIS. The entire model was imported to ArcGIS as a large array of points, and will be discussed in detail in later sections. Model 2 Model 2 was constructed as an alternativ e to the categorical nature and display problems associated with Model 1. The K-m eans hierarchical clustering technique and closest point interpola tion method were not used in the development of Model 2 in order to provide a numerically based model. However, A remained the predictive variable. The output of Model 2 is a series of raster surfaces rather than a 3D array of point values,as in Model 1. However, the values calculated and displayed in model 2 are less intuitive than the categorical Model 1. Rather than predicting either sand or an argillic horizon for each data point, Model 2 predicts a A value for a designated depth. Though this provides for a much more dynamic and aes thetically pleasing model, the values are not easy for a person of average GPR knowledge to understand. To create Model 2, the master table of data values was divided into subsets based on depth. A new subset was crea ted at 8 cm intervals for the depth of the GPR survey, resulting in 37 subsets of data. Next, each s ubset table was added to ArcGIS to begin the interpolation procedure. Then, ordinary kriging was performed on each subset to calculate a predicted surface of A values, creating a cell based raster model for each selected depth.

PAGE 69

58 Figure 7-5. Model 1, using RockWorks 2002 so ftware. The red areas in Model 1 are voxels with a value of 1, indicating the argillic hori zon. The purple areas in Model 1 are voxels with a va lue of 0, indicating sand. Ordinary kriging is the most used krigi ng method and attempts to estimate a value at a point of a region for which a variogram is known, using data in the neighborhood of the estimation location. (Wackernagel 1995). Th e interpolated surfac es give Model 2 a visually pleasing appearance, Figure 7-6, and do not use cate gorical data, such as the clusters in Model 1. Howeve r, a model that predicts A is not that well understood. The standard error of prediction associated with the kriging procedure of Model 2 is in Figure 7-7.

PAGE 70

59 Model 3 Model 3 was constructed to combine aspects of Model 1 and Model 2. Model 1 predicted an easily understood variable, presen ce of an argillic horizon, but did not have the visual appeal of Model 2. Model 3 was designed to not only predict the depth to argillic horizon, rather than A, but also provide a model that was easy to interpret. Model 3 accomplished these goals by reproc essing the GPR profiles. Rather than exporting a massive amount of da ta and quantitatively decidi ng where the argillic horizon begins, the horizon interface was picked in Reflex. By picking, a horizon or interface can be traced with the mouse, as in Figure 7-8, and have those traced values stored. Each stored data point can contain X, Y, Z, and ampl itude attribute data. In this case, the argillic horizon of each GPR prof ile was picked, and the stored values exported to a table. Since th e original GPR profile had al ready been georeferenced, the picked data points were ge oreferenced by exporting the X, Y, and Z coordinates. Each picked profile contained a pproximately 3,000 points, depending on the length of the transect. The points from each profile were combined to form a master table, upon which spatial analys is and interpolation could be conducted. But first, the points were imported to ArcGIS and converted to a shapefile. The Z value associated with each point was the picked depth to ar gillic at that location. By interpolating between picked points, a surface could be ca lculated that predicted depth to argillic across the field. This interpolation proce dure was conducted using the kriging function available in the spatial analyst extension. The result of Model 3, as shown in Figure 7-9, is a single raster model predic ting depth to argillic for the research field. Figure 7-10 is the standard error of prediction associated with the kriging pr ocedure for Model 3.

PAGE 71

60 Depth 30 cm 80 cm 122 cm 156 cm 223 cm 306 cm 348 cm Figure 7-6. 3D display of Model 2 as viewed in ArcScene. Each layer is a kriged interpolation of A values from the corresponding depth. A total of 36 layers from 20 cm to 357 cm were cal culated. 7 layers are shown in this figure.

PAGE 72

61 Figure 7-7. Standard error of prediction of Model 2. The sta ndard error of prediction of Model 2 for the kriging procedure is largest in the areas between GPR transects, and lowest in the area s closest to the transects.

PAGE 73

62 Figure 7-8. Picked argillic horizon. A) The argillic horizon has been picked in post-processing software. Picking is selecting a series of data points (B) by tracing a horizon or feature using a mouse. Depth to Argillic (cm)Value High : 330 Low : 68 060120 30 Meters Figure 7-9. Image of Model 3. Model 3 is a continuous surface of argillic horizon depths interpolated from picked argillic horizon interface depths from GPR profile data.

PAGE 74

63 Figure 7-10. The standard error of predic tion for Model 3. As for Model 2, the areas closest to the GPR transect lines sh owed a smaller standard error than those farther away from the GPR transect lines.

PAGE 75

64 CHAPTER 8 MODEL VALIDATION All Models created were statistically validated using a validation dataset from Pine Acres. Test holes (T otal 36) were evaluated thr oughout the research field to ascertain the true depth to argillic and are i llustrated in Figure 81. At the validation locations, the depth to argillic ranged from 103.0 cm to 400.0 cm, with a mean value of 210.6 cm and standard deviation of 64.7 cm. Th is histogram of the values is shown in Figure 8-2. Model 1 In Model 1, each data point is either a 0 or 1. A validation method to correlate these categorical data with numerical depth to argillic measurements had to be developed. By “reading” the nearest values to a validation point, a qualitative measurement of argillic depth could be determined. This value could then be compared with the validation dataset. In ArcScene, the “column” of points closest to each validation point was selected using a spatial query to be interpre ted. Each “column” of points represented the soil at that location in 0’s (sa nd) and 1’s (argillic). The dept h to argillic was determined by evaluating each series of integers to lo cate the depth at which 1’s were uniformly presentThe validation procedure for Model 1 is shown in Figure 8-3. The predicted argillic depths had a mean value of 192.1 cm and a standard deviat ion of 64.7 cm. The Pearson correlation coefficient between the obs erved and predicted ar gillic values was 0.663 and significant at = 0.01

PAGE 76

65 Figure 8-1. Validation point locations. The valid ation points were inte nded to be placed at locations not cove red by GPR transects.

PAGE 77

66 Figure 8-2. Histogram of depth to argillic va lues of validation points. The histogram has a skewness of 0.798 and kurtosis of 0.768. The Shapiro-Wilk statistic is 0.950.

PAGE 78

67 0 1 0 0 0 0 0 1 0 0 1 1 1 1 Clay 225 cm Column of Data Model 1 0 1 0 0 0 0 0 1 0 0 1 1 1 1 Clay 225 cm Column of Data Model 1 Figure 8-3. Validation procedure for Model 1. The column of data closest to each validation data point was evaluated to determine the depth at which integer values representing argillic (1) becam e uniform. The depth at which this occurred was extracted from the attr ibute table of Model 1 and correlated with the depth to argillic fo r the respective validation point. Model 2 Model 2 consisted of a series of raster models that predicted A at depth intervals from 20 cm to 357 cm. Validating the mode l required developing a method to compare A values with depth to argillic values. Th rough the use of raster analysis techniques and regression statistics, a comparison was calculated. Each validation point overlay a single rast er cell for each raster layer of varying depths. The raster cells contained the interpolated A value for that depth. First, each raster cell in each raster layer beneath each validation point was extr acted and organized. By doing this, a table of data was created that listed the predicted A values at depth intervals for each validation point. Then, A was plotted against depth, and a regression curve was calculated that fit the data distribution. The majo rity of the validation point

PAGE 79

68 data were fit best by a 3rd order polynomial regression function, with a mean coefficient of determination of 0.9103. The regression cu rve was used to calculate the predicted A for Model 2, termed Amod2, at the depth to argillic determined for each validation point. This process is il lustrated in Figure 8-4. The Amod2 from each validation point was then plotted against the depths at which the Amod2 values were predicted. A 2n d order polynomial curve with an adjusted r2 = 0.792 and standard error of 5856 wa s calculated to fit th e data distribution. The interpretation of this equation is some what complicated. Th e regression equation will return a predicted A only for the depth at which clay first appears. Likewise, a depth can be calculated only from a A value at the depth at wh ich clay first appears. The data do not support predicted A values at any depth in th e soil other than that at which clay first appears. Model 3 In Model 3, a singe raster layer was interp olated from a ”picked” depth to argillic measurements for each GPR profile. A compar ison of the predicted depth to argillic measurements for each validation point was eval uated with the observed depth to argillic measurement at each location in the raster layer, Dmod3, to determine the degree of correlation. The Dmod3 value was extracted fr om the raster cell associated with each validation point. The Pearson correlati on coefficient for the values was 0.630 and significant at = 0.01. Model 4 The validation procedure for Model 4 differed from the previous three methods. This is because the model was generated fr om a regression equation developed from the validation procedures for other models (Equa tion 8-1). While validating Models 2 and 3,

PAGE 80

69 a statistical relationship was discovered that predicted the depth to argillic at the validation points with a higher degree of accu racy than either model independently. The relationship was linear regressi on Equation 8-1 that used the Amod2 values from Model 2 and Dmod3 from Model 3. y = -0.0127x3 + 6.8125x2 815.41x + 28254 r2 = 0.9507 0 10000 20000 30000 40000 50000 60000 70000 050100150200250300350400Depth (cm)Delta A Argillic at 204 cm Amod2 = ~38500 Figure 8-4. Regression equation for validation point, Model 2. For each validation point, the corresponding rast er cells values ( A) from layers at depth intervals were extracted and plotted against thei r respective depths. An equation was calculated that fit the distribut ion of the data. A predicted A for the sandargillic interface was calculated using the regression equation. For this point, the sand-argillic interface was determined through ground verification to be at 235 cm. The corresponding A value at this depth calculated using the regression equation is 48610. This procedure was performed for each validation point. Depth to argillic= 143.210 + (-.484 Dmod3 ) + (8-1) (.00002224 (Dmod3 Amod2 )) For this model, adjusted r2 = 0.840 and th e standard error of the estimate = 25.85.

PAGE 81

70 To perform the raster calculation, an interpolated raster layer of Amod2 values was created using kriging and multiplied by the M odel 3 raster layer. Equation (8-1) was applied to Model 3 and the newly created Dmod3 Amod2 layer using the raster calculator function in the Spatial Analyst extension to create Model 4, illustrated in Figure 8-5. Depth to Argillic (cm)Value High : 505 Low : 190 060120 30Meters Figure 8-5. Image of Model 4. Model 4 was created by applying a regression equation to Model 3. Model 4 accomplishes all the goals desired; the model is understandable, GIS compatible, accurate to an acceptable degree, and visually appealing.Every MS Word document uses styles to format information. To help prevent the unnecessary copying of different styles into your dissertation, follow these rules of thumb when copying information: .

PAGE 82

71 CHAPTER 9 SUMMARY AND CONCLUSIONS Data Collection If a more accurate model is desired, the fi rst change should be in data collection. The GPR transects for this project were gather ed 40 meters apart. Several GPR analysts have been successful in 3D modeling by us ing a much tighter spacing. Grasmueck and Weger (2002) created 3D data from transects collected 10 cm apart. However, the area covered by these denser grids is much smalle r in size than the research field at Pine Acres. The transect spacing distance is a func tion of several factors. First, what is the size of the feature you are attempting to m odel? Second, how much time are you willing or able to spend in data collection and processing? Third, do you have the financial resources to justify the adde d expenses associated with more data collection and processing? This research intended to model the ar gillic horizon a continuous feature of infinite size. In addition, th e resolution of the model could remain somewhat large. The final cell size of Models 3 and 4 was 2 m. This may initially seem to be a coarse resolution. However, it is assumed that the argillic horizon has a uniform depth over a 2 m area. If more detailed data are required, the amount of time invested in the project increases accordingly. A previ ously unstated goal of this project was to determine an efficient methodology for GPR data collecti on, processing, and modeling. The goal of efficiency may have been overshadowed by mo re quantitative aspect s of the project, but

PAGE 83

72 an efficient plan was always a goal. To colle ct a larger, more detailed amount of data, the project no longer becomes efficient. For exam ple, if the transect spacing were to be decreased to 2 m, a total of 240 transects would need to be collected, not including any transects to be collected at angles through the collection gri d. This equates to 4 1010 points of data for a 320 m 160 m field. Wh en conducting a GPR survey, it is important to collect all the data for a grid at the same time in the field, keeping the soil conditions uniform for all transects. The soil conditions would undoubtedly change in the amount of time it takes to collect 240 transects of data. Furthermore, the processing time has been increased immeasurably. While conducti ng detailed and time-consuming processing steps on 18 transects is feasible and quite pos sible efficient, performing these same tasks on the amount of data proposed by smaller tran sect spacings would take an unpredictable amount of time. Furthermore, it is freque ntly necessary to revise the collection methodology and recollect data for the entire fi eld. A total of 4 complete surveys were completed for this project over a year and a half. Fortunately, each complete survey could be completed in less than one day. Collecting 240 transects multiple times would not be an efficient or practic al use of time and resources. While resolution, time, and money are defini te variables to be considered before developing a methodology, increased sampling is most likely the best way to generate a more accurate model. However, the methods us ed in data collection for this project were satisfactory to all involved, and met the initial goals of the research. Ground-Penetrating Radar Processing Unfortunately, error is an inescapable variable of GPR collection and post processing. A considerable source of error is the way in which dept h is calculated from time. To calculate a depth from time, the ve locity of the wave mu st be known. The GPR

PAGE 84

73 allows only a single dielectric constant to be assigned for the entire GPR transect, no matter how many different soil types or horiz ons occurred within the transect. The dielectric constant is directly related to the velocity of the radar wave through the soil. By allowing only a single di electric constant to be assi gned to a transect, you are assuming only a single velocity for the radar wave through a variety of different soils. Though no study determining the amount of e rror this situation incurs has been completed, it is certainly a factor wh en analyzing GPR data by depth. It is important to understand the context in which this research was completed, and the overall goals. The research was intende d to create a model that predicted depth to argillic from GPR values onl y for the field in which GPR da ta was collected. For this reason, a variety of GPR filters and proce ssing techniques were used to maximize the differences between the sand and argillic horizon. In other circumstances, more care may be needed in applying various filters to the GPR data that augments the raw data. Also, this research was focused on the relative cha nge in amplitude, or change from one point to another. The actual values were somewh at irrelevant; the amplitude could be 10 or 100,000. It only mattered what the changes were in those values from one sample to the next. Qualitative Methods Initially, it was planned to develop a completely quantitative method to model the argillic horizon in the field. Because the resu lts of the quantitative methods were not as encouraging as initially hoped, a more qual itative approach was taken. The qualitative nature of the methodology is ev ident in the verification of Model 1 and the generation of Model 3. When validating Model 1, the upper limit of the argillic horizon was subjectively

PAGE 85

74 chosen from the column of data. Though this method of choosing the argillic horizon was not statistical in its nature, a more indi vidualized approach could be applied to each column. Nevertheless, this quantitative validati on is a possible source of error in Model 1. In the generation of Model 3, the depth to argillic for each trace was picked by tracing the interface with a mouse. This qual itative approach has specific benefits and drawbacks. One disadvantage is that the same picked data cannot be exactly repeated. If the same profile were pic ked several times, the differenc e in picked values would be minimal, but there would always be a diffe rence. A second disadvantage is the way in which the profile is interprete d. The picks could above or below the actual interface, depending on how the user interprets th e GPR image. However, by allowing an experienced GPR user to pick the depths to argillic, there are desirable benefits. A specific benefit of allowing an experien ced GPR user to pick the argillic horizon is that noise or inte rference within the GPR profile has much less influence. While noise can be identified and exclude d by a user, a statistical method may have trouble interpreting such a phenomenon and mi stake it for a soil feature. Also, an experienced user may be able to identify an overall trend to the argillic horizon present in the imagery that may not be as evident usi ng quantitative methods. A third advantage of qualitatively picking may be initially overlooked. By picking a horizon, only one point per trace is expo rted. With this amount of data, no compression is needed. Without compression, the full amount of data is availabl e to perform the interpolations, a ten-fold increase over the compressed data. While a quantitative, repeat able approach was initially planned and desired, specific benefits of quali tative methods support the techniques used in this project

PAGE 86

75 Model 1 While Model 1 met the objectives initially defined, the statis tical correlation was not as strong as expected. Ther e are several explanations for this problem. First, the data in the model are categorical, not numerical. This presented problems in the validation procedure that resulted in a qualitative meas urement of the predictions in Model 1. The validation dataset contained de pth to argillic values, and th e data in Model 1 predicted either sand or argillic at va rious depths. Furthermore, rather than picking the nearest column of data to a validation point to retrieve a predicted value, developing an interpolation procedure to take into account the three or four near est columns would be a better method of extracting the predicted valu es. Also, if the model were to predict a numeric value rather than a category, it is believed the model would have been more successful. For example, rather than predic ting sand or argillic, perhaps the model should predict percent clay for each voxel. Though the model was able to be visualized in GIS, the visual quality did not satisfy rese arch objectives. The model took on the characteristics of a pointillist painting. At a distance, the conglomeration of points appeared to represent a coherent image. Howe ver, as the user moves closer, or zooms in, the seemingly coherent image dissipates into individual points, which are much more difficult to interpret. This display problem prompted the generation of the other three models. Model 2 Model 2 was much more visually appeali ng than Model 1, but the end value of a predicted A was not a good variable to model. Though A has a definite correlation with the depth to argillic, modeling A creates problems with validation. While the values were numeric rather than categorical, and therefore easier to correlate with depth

PAGE 87

76 to argillic values, determ ining a relationship between A and depth to argillic was very difficult. The result of the validation was a regression equatio n that proved true only at the sand-argillic interf ace of the soil profile. If you know the depth of the sand-argillic interface, what is the use of the mo del in the first place? Also, A is not a variable with which users are familiar. Because Model 2 did not have a very practical use, the modeling variable was changed for future models But, the raster type of model proved to be a powerful means of display, and was kept for future models. Models 3 and 4 Models 3 and 4 are very similar in gene ration and display. Both models predict the same variable, depth to argillic, and ar e raster models. Because both models are similar in their development, similar problem s are inherent. The la rgest portion of error in the model is probably due to the qual itative method in which the raw data was extracted from the GPR profiles. Though pic king the argillic horizon has desirable benefits, it is still a qualitative practice whic h carries more error. The specific amount of error has not been determined.

PAGE 88

77 APPENDIX SOIL PROFILE DESCRIPTIONS Table A-1 Description of represen tative pedons at the research site. HorizonColor (10YR)Texture Depth (cm)NotesLocation (Field 5a)Ap4\1Fine sand0-15 5\0 Eg16\2Fine sand15-60few 7.5 YR 5\6 redox concentrations Eg27\2Fine sand60-115 Eg38\1Fine sand115-154 Btg17\1Loamy sand154-165 Btg27\1Sandy loam165+7.5 YR 5\6 redox concentrations Ap2\2Fine sand0-19 5\7 Bw15\4Fine sand19-77 Bw27\4Fine sand77-130matrix stripping, faint 7.5 YR 5\6 Bt5\6Loamy sand130-170 Btg7\4Sandy loam170-185common 7\1 redox depletions 2Btg6\4Sandy clay loam185-200many 7\1 depletions Ap 3\2Fine sand0-17 6\8 Bw15\3Fine sand17-35 Bw26\4Fine sand35-85few faint redox concentrations Bw37\4Fine sand85-140common faint redox concentrations Eg8\3Fine sand140-1607/1 redox depletions Eg and Bt8\3Fine sand160-200Eg 6\6Loamy sand160-200Bt Ap4\1Fine sand0-16 7\3 Bw16\3Fine sand16-85matrix stripping, few faint redox concentrations, 2.5 YR 5\6 Bw27\3Fine sand85-105 2Btg16\1Sandy clay loam105-115Hawthorn Formation 2Btg26\1Clay115-200Clay films, 5\1 10 YR depletions The soil profile descriptions listed above are an example of th e types of soils that were found across the research plot area. The de scriptions were made by Ron Kuehle and Michael Tischler in 2001. The location in the field refers to the numbering convention of the sampling locations throughout the field.

PAGE 89

78 LIST OF REFERENCES Atekwana, E.A., Sauck, W.A., and Werkema, D.D. Jr. 1998. Investigations of geoelectrical signatures at a hydrocarbon c ontaminated site. Journal of Applied Geophysics. 44: 167-180. Basile, V., Carrozzo, M.T. Negri, S., Nuzzo, L., Quarta, T., and Villani, A.V. 2000. A ground-penetrating radar survey for archaeol ogical investigations in an urban area (Leece, Italy). Journal of Applied Geophysics. 44: 15-32. Collins, M.E. 1990. Applications of groundpenetrating radar. XVII Reunion Nacional Sobre Edafologia. 24(28): 15-32. Conyers, L.B., and Goodman, D. 1997. Gr ound-penetrating radar: An introduction for archaeologists. Altimira Press, Walnut Creek, CA. Dangermond, J., and Smith, L.K. 1988. Ge ographic information systems and the revolution in cartography: The nature of the role played by a commercial organization. The American Cartographer. 15(3): 301-310. Doolittle, J.A., and Collins, M.E. 1995. Use of soil information to determine application of ground penetrating radar. Geoderma. 33: 101-108. Doolittle, J.A., and Collins, M.E. 1998. A comparison of EM induction and GPR methods in areas of karst. Geoderma. 85: 83-102. Environmental Science Research Institute. 2002. Glossary of GIS terms [Online]. Available at http://www.esri.com/libra ry/glossary/glossar y.html (verified 22 September 2002). GARMIN Corporation. 2001. eTrex Vista™ personal navigator: Owner’s manual and reference guide. GARMIN Corporation, Olathe, KS. Gish, T.J., Dulaney, W.P., Kung, K.-J. S., Daughtry, C.S.T., Doolittle, J.A., and Miller, P.T. 2002. Evaluating use of ground-penetr ating radar for identifying subsurface flow pathways. Soil Sci. Soc. Am. J. 66: 1620-1629. Grandjean, G., Curry, J.C., Bittri, A. 2000. Evaluation of GPR techniques for civilengineering applications: study on a test si te. Journal of Applied Geophysics. 45: 141-156. Grasmueck, M., and Weger, R. 2002. Reassessme nt of local paleocurrent directions in the Miami oolitic limestone with 3D groundpenetrating radar. Proceedings of the

PAGE 90

79 9th International Conference on Ground Pene trating Radar. Santa Barbara, CA, 29 April to 2 May 2002, pp. 211-216. Harris, W.G. 2002. Phosphate minerals. p. 637-665. In Dixon, J.B., and Schulze, D.G. (ed.) Soil Mineralogy with Environmental Applications. Soil Sci. Soc. Am. J., Madison, WI. Hyde, B. 1997. Information Bulletin No. 98-06. USDA, Bureau of Land Management. Available online at http://www.blm.gov/ nhp/efoia/wo/fy98/ib98-06.html (verified 22 September 2002). Isaaks, E.H., and Srivastava, R.M. 1990. An introduction to applied geostatistics. Oxford University Press, Oxford, England. Knoll, M.D. 1996. A petrophysical basis fo r ground penetrating ra dar and very early time electromagnetics: Electrical propert ies of sand-clay mixtures. Ph.D Dissertation. University of British Columbia. Lane, E., and Hoenstine, R.W. 1991. E nvironmental geology and hydrogeology of the Ocala area, Florida. Florida Geological Su rvey Spec. Pub. 31. State of Florida, DNR, Tallahassee, FL. Mark, D.M., Chrisman, N, Frank, A.U., McHa ffie, P.H., Pickles, J. 2002. The GIS history project [Online]. Available at http://www.geog.buffalo.edu/ncgia/gishist/b ar_harbor.html (verified 10 October 2002) Marsal, O., Harri, A.-M., Lognonne, P., Rocard, F., Counil, J.-L. 2000. Netlander: the first scientific lander network on the surface of mars. Available online at http://netlander.fmi.fi/pub/NetLande rMissionDescription.pdf (verified 22 September 2002). National Oceanic and Atmospheric Admini stration. 1994. Time bias corrected divisional temperature – precipitation – drought index [Online]. Available at http://lwf.ncdc.noaa.gov/oa/climate/onlin eprod/drought/readme.html (verified 22 September 2002). Olhoeft, G.R. 1996. Application of ground pe netrating radar. Proceedings of the 6th International Conference on Ground Pene trating Radar. Sendai, Japan, 30 September to 3 October 1996, pp. 1-4. Olhoeft, G.R. 1998. Electrical, magnetic, and geometric propert ies that determine ground penetrating radar performance. Proceedings of the 7th International Conference on Ground Penetrating Radar. Lawrence, Kansas, 27 to 30 May 1998, pp. 177-182.

PAGE 91

80 Olhoeft, G.R. 1999. Applications and frustr ations in using ground penetrating radar. Proceedings of the Ultra Wide Band C onference. Washington, D.C., 20 to 22 September 1999. Olhoeft, G.R. 2000. Ground penetrating radar GRORADAR [Online]. Available at http://www.g-p-r.com/introduc.htm (verified 22 September 2002). Ormsby, T., Napoleon, E., Burke, R., Groessl C., Feaster, L. 2001. Getting to know ArcGIS desktop. ESRI Press, Redlands, CA. Ovalles, F.A., Collins, M.E. 1988. Evaluation of soil variability in northwest florida using geostatistics. SSSAJ. 52: 1702-1708. Radzevicius, S.J., Guy, E.D., and Daniels, J.J. 2000. Pitfalls in GPR data interpretation: differentiating stratig raphy and buried objects from periodic antenna and target effects. Geophysical Research Letters, vol. 27. 20: 3393-3396. Saarenketo, T. 1998. Electrical properties of water in clay and silty soils. Journal of Applied Geophysics. 40: 73-88. Sandmeier, K.J. 2002. REFLEXW. Release 2.5 Sandmeier Scientific Sofware, Inc. Karlsruhe, Germany. Sigurdsson, T., and Overgaard, T. 1996. App lication of GPR for 3d visualization of geological and structural va riation in a limestone form ation. Proceedings of the 6th International Conference on Ground Pene trating Radar. Sendai, Japan, 30 September to 3 Oct. 1996, pp. 39-44. Soil Survey Division Staff. 1993. Soil su rvey manual. USDA Handbook 18. U.S. Gov. Print. Office, Washington, D.C. Sumner, M.E. (ed.) 2000. Handbook of soil science. CRC Press, Boca Raton, FL. Thomas, B.P., Cummings, E., and Wittstruc k, W.H. 1985. Soil survey of alachua county, Florida. Soil Conservation Servic e, U.S. Department of Agriculture. U.S. Gov. Print Office, Washington D.C. Thomas, B.P., Law, L., and Stankey, D.L. 1979. Soil survey of marion county, Florida. Soil Conservation Service, U.S. Departme nt of Agriculture. U.S. Gov. Print Office, Washington, D.C. Tomlinson, R.F. 1988. The impact of th e transition from analogue to digital cartographic representation. The Americ an Cartographer. 15(3): 249-261. Trimble Navigation Limited. 1998. GPS mappi ng for GIS with asset surveyor. Trimble Navigation Limited, Sunnyvale, CA.

PAGE 92

81 United States Department of Agriculture – Natural Resources Conservation Service. 2001. National soil survey handbook, title 430VI [Online]. Available online at http://www.statlab.iastate.edu/soils /nssh/ (verified 22 September 2002). Wackernagel, H. 1995. Multivariate statisti cs. Springer – Verlag, Berlin, Germany.

PAGE 93

82 BIOGRAPHICAL SKETCH Michael Tischler was born to James Tischler, Sr. and Patricia Davis in 1978 in Greensburg, Pennsylvania. At the age of 4, Michael moved to Alexandria, Virginia, where he stayed until 1990. At the age of 13, Michael moved with his mother to Punxsutawney, Pennsylvania, where he stayed until high school graduation. Michael left Pennsylvania for the University of Dubuque, Io wa to begin college as an Environmental Science major. During his sophomore and ju nior year, Michael transferred to North Dakota State University and changed his ma jor to Soil Science, with an emphasis on pedology and morphology. While in North Dakot a, Michael worked as a Soil Scientist Trainee for the Natural Resources Conser vation Service during the summers. Upon graduating with honors from North Dakota Stat e University in 2000, Michael enrolled at the University of Florida to begin a Master of Science degree. In October 2002, Michael left to pursue a position at the NASA Godda rd Space Flight Center as a government contractor


Permanent Link: http://ufdc.ufl.edu/UFE0000806/00001

Material Information

Title: Integration of Ground-Penetrating Radar, Geographic Information Systems, and Global Positioning Systems for 3-Dimensional Soil Modeling
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0000806:00001

Permanent Link: http://ufdc.ufl.edu/UFE0000806/00001

Material Information

Title: Integration of Ground-Penetrating Radar, Geographic Information Systems, and Global Positioning Systems for 3-Dimensional Soil Modeling
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0000806:00001


This item has the following downloads:


Full Text










INTEGRATING GROUND-PENETRATING RADAR, GEOGRAPHIC
INFORMATION SYSTEMS AND GLOBAL POSITIONING SYSTEMS FOR
3-DIMENSIONAL SOIL MODELING














By

MICHAEL A. TISCHLER


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA


2003




























Copyright 2003

by

Micheal A. Tischler




























I would like to dedicate this manuscript to my family and friends. My father and

stepmother; James and Laura Tischler; mother and stepfather, Patricia and Merrill Davis,

and brother; James Tischler, Jr. never wavered in their love and support. To them and the

rest of my family, I am truly and forever grateful.















ACKNOWLEDGMENTS

This research could not have been completed without the input, advice, and

support of dozens of individuals. Although there is neither the space nor the words to

convey my gratitude to everyone who offered support, I would like to express my thanks

to a few. Dr. Mary Collins was an exceptional advisor in every aspect of the word. She

allowed enough freedom for me to forge my own ideas while always guiding me toward

the final goal. I will always be grateful to her for the time and energy that she invested in

my education at the University of Florida. Without the assistance of Dr. Sabine

Grunwald, I would never have had the knowledge to perform the complex spatial analysis

and validation procedures necessary to complete the research. John Schultz and Larry

Ellis both offered more assistance and advice than I could expect anyone not personally

involved to provide. I would also like to thank Dr. Jimmie Richardson and Dr. David

Hopkins of North Dakota State University. They saw the scientist within me before I

knew there was one myself. Finally, I would like to thank my family and friends in full

for their support and kindness. None of them will ever realize how thankful I really am.

"Come to the edge they said. He said: I am afraid. Come to the edge they said.
He came. They pushed him, and he flew. ...."
-Adapted from Guillaume Apollinaire
















TABLE OF CONTENTS


A C K N O W L E D G M E N T S ................................................................................................. iv

LIST OF TABLES ................................................... vii

LIST OF FIGURES ................................ ........... ............................ viii

A B S T R A C T ........................................................................................................ ........ .. x

CHAPTERS

1 IN T R O D U C T IO N ....................................................... ... .................... ... ..................... 2

2 GROUND-PENETRATING RADAR (GPR)...................................................3...

B a c k g ro u n d ..................................................................................................................... 3
T h e o ry ............................................................................................................. .......... 4
S ite S u itab ility ......................................................................................................... 8
D ata S tru ctu re ............................................................................................................... 1 1

3 GEOGRAPHIC INFORMATION SYSTEMS (GIS)...............................................13

H isto ry ........................................................................................................................... 1 3
Georeferenced Data .............. .................... ........ .......... 15
F ile F o rm a ts ................................................................................................................ .. 1 8
Analysis ............................................................... 18

4 GLOBAL POSITIONING SYSTEMS (GPS) ................ ...................................20

5 FIELD SITE A T PIN E A CRE S....................................... ...................... ................ 23

Soils ....................................................... ...... .............. 23
G e o lo g y ......................................................................................................................... 2 7
C lim ate ......................................................................................................... ........... 3 0
T o p o g ra p h y ................................................................................................................... 3 1

6 MATERIALS AND METHODS..............................................................................33

Data Collection ................................................. 33
Ground-Penetrating Radar Post-Processing............................................................ 34


v









Model Generation ................................ .. ......... ............................ 40
Data Preparation .............................. .. .......... .............................. 41

7 RESULTS AND DISCUSSION ......................................................... ................ 46

V ariogram s and K riging ......................................... ............. .................................. 46
Modeling Parameters ............................. .. .......... ................................... 52
M odel 1 ......................................................................................................... .............. 54
M o d e l 2 ......................................................................................................................... 5 7
M o d e l 3 ......................................................................................................................... 5 9

8 M O D EL V A L ID A TIO N .......................................................................... ...............64

M odel 1 ......................................................................................................... .............. 64
M o d e l 2 ......................................................................................................................... 6 7
M o d e l 3 ......................................................................................................................... 6 8
M o d e l 4 ......................................................................................................................... 6 8

9 SUMMARY AND CONCLUSIONS ..........................................................................71

D ata C o lle ctio n ............................................................................................................. 7 1
Ground-Penetrating Radar Processing....................... ................................................. 72
Qualitative Methods ....................................................................................... 73
M odel 1 ......................................................................................................... .............. 75
M o d e l 2 ......................................................................................................................... 7 5
Models 3 and 4 .......................................................................................................76

APPENDIX SOIL PROFILE DESCRIPTIONS ............................................................77

B IO G R A PH IC A L SK E T C H ............................................................................................. 82















LIST OF TABLES


Table page

2-1 D electric constants of various m aterials................................................... ................ 10

2-2 Cation exchange capacities of soil materials............................................................ 11

5-1 Corner coordinates of the research plot..................................................... ................ 26

6-1 Example of the tabular data before any statistical manipulation...............................43

7-1 Tabular GPR data ready for 3D modeling................................................................56

A-i Description of representative pedons at the research site.........................................77















LIST OF FIGURES


Figure page

2-1 Monostatic and bistatic antenna configuration.........................................................5...

2-2 Relative depth of penetration and resolution of various antennas............... ...............6

2-3 Ground-Penetrating Radar wave propagation, scattering and reflection.....................9...

2-4 Ground-Penetrating Radar traces and samples......................................................... 12

3-1 Three types of vector data: points, lines, and polygons...................... ............... 16

3-2 R aster data consisting of equal sized cells................................................. ................ 17

5-1 Location of research site in Marion County, Florida.................................................24

5-2 R research plot at P ine A cres........................................... ......................... ................ 25

5-3 Argillic horizon at Pine Acres Research Facility. .....................................................27

5-4 General Soil Associations -Marion County, Florida................................................28

5-5 M odified Palm er D brought Index ...................................... ...................... ................ 31

5-6 Palm er Z Index of precipitation....................................... ....................... ................ 32

6-1 Simplified interior of a GPR antenna housing unit...................................................36

6-2 Uncorrected depth to surface GPR profile ................................................................37

6-3 Corrected depth to surface GPR profile.. .................................................. ................ 37

6-4 B background filter of G PR profile..................................... ...................... ................ 39

6-5 Spectral w hitening filter of GPR profile.................................................... ................ 40

6-6 Com pression of GPR profile. ................ .............................................................. 41

6-7 Scatterplot of AA versus depth for all GPR data.......................................................44

6-8 Ground-Penetrating Radar profile image described by Figure 6-7 ..............................45









7-1 Range, sill, and nugget value for a variogram ....................................... ................ 47

7-2 Variogram s of GPR data at different depths......................................... ................ 48

7-3 D epth versus Range for several variogram s.......................................... ................ 53

7-4 M ean depth of clustered variables......................................................... ................ 55

7-5 M odel 1, using RockW orks 2002 software........................................... ................ 58

7-6 3D display of M odel 2 as viewed in ArcScene..................................... ................ 60

7-7 Standard error of prediction of M odel 2................................................ ................ 61

7-8 "P icked" argillic horizon ........................................... ......................... ................ 62

7-9 Image of Model 3 .................... .. ........... ............................... 62

7-10 The standard error of prediction for M odel 3...................................... ................ 63

8-1 V alidation point locations....................................... ......................... ................ 65

8-2 Histogram of depth to argillic values for validation points...................................66

8-3 V alidation procedure for M odel 1......................................................... ................ 67

8-4 Regression equation for validation point, Model 2 ..............................................69

8-5 Image of Model 4.................... .. ........... ............................... 70









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

INTEGRATING GROUND-PENETRATING RADAR, GEOGRAPHIC
INFORMATION SYSTEMS, AND GLOBAL POSITIONING SYSTEMS FOR
3-DIMENSIONAL SOIL MODELING

By

Michael A. Tischler

May 2003

Chair: Dr. Mary E. Collins
Department: Soil and Water Science Department

Ground-Penetrating Radar (GPR) has become a useful and efficient instrument for

gathering information about subsurface diagnostic horizons in Florida soils. Geographic

Information Systems (GIS) in are a popular and valuable tool for spatial data analysis of

real world features in a digital environment. Florida has a vast array of freely and readily

available GIS data. Ground-Penetrating Radar can be linked to GIS by using Global

Positioning Systems (GPS). By combining GPR, GPS, and GIS technologies, a more

detailed geophysical survey can be completed for an area of interest by integrating

hydrologic, pedologic, and geologic data. Thus, the objectives of this research were to

identify subsurface soil layers using GPR and their geographic position with a highly

accurate GPS; to develop a procedure to import GPR data into a popular software

package, such as ArcGIS, and; to create 3D subsurface models based on the imported

GPR data. The site for this study was the Plant Science Research and Education Center

in Marion County, Florida. The soils are characterized by Recent-Pleistocene-age sand

over the clayey, marine deposited Plio-Miocene-age Hawthorn Formation which drapes

the Eocene-age Ocala Limestone. Consequently, soils in the research area vary from









deep quartz sands (Typic Quartzipsamments) to shallow outcrops of the Hawthorn

Formation (Arenic Hapludalfs). A GPR survey was performed on a 160 m x 320 m grid

to gather data for processing. Four subsurface models estimating the depth to argillic

horizon were created using a variety of specialized GPR data filters and geostatistical

data analyses. The models were compared with ground-truth points that measured the

depth to argillic horizon to validate each model and calculate error. These models may

assist research station personnel to determine best management practices (including

experimental plot placement, irrigation management, fertilizer treatment, and pesticide

applications).














CHAPTER 1
INTRODUCTION

Ground-penetrating radar (GPR) has proven to be a useful and efficient remote

sensing geophysical instrument for gathering information about near-surface pedologic

and geologic materials. Geographic Information Systems (GIS) provide a means of

storing, manipulating, analyzing, and displaying spatially distributed data in a

two-dimensional (2D) or three-dimensional (3D) view. Combining the efficiency and

practicality of GPR data with the visual appeal, analysis, and interpretive power of GIS is

the next logical step in the evolution of both technologies, and can be accomplished using

Global Positioning Systems (GPS).

Recently, georeferenced datasets providing information about the earth and its

environment have become widely available. The accumulation of this information has

created a demand for 3D geophysical data that can be analyzed in conjunction with these

easily accessible datasets. However, no methods exist for combining georeferenced GPR

data with GIS datasets, (which would reduce time and costs while increasing the

interpretive quality of the information). Developing such a method is the most important

goal of this research.

By transforming graphic GPR data to numeric values and spatial coordinates, the

data can be manipulated, transformed, converted, and integrated into a variety of software

packages capable of accomplishing specific portions of the overall objectives.

Ground-Penetrating Radar analysis software can process data to allow better visualization

of the subsurface features. Statistical software can be used to make mathematical









manipulations of the data to ease interpretation. Modeling software can interpolate areas

where little or no data exists to create a solid 3D diagram. Finally, GIS software can be

used to display the modeled information in combination with already available

information about the earth, environment, and natural resources. This research intends to

bridge the gap between GPR and GIS to allow powerful analysis in detailed studies and

projects.

The specific objectives of this research project are to:

Conduct a 3D GPR survey to obtain information about soil stratigraphy

Create a 3D model from the GPR data

Develop methodology to integrate the 3D GPR data into a GIS


The 3D models created need to meet 3 stringent criteria. They must

Be easily understood. Conventional 3D diagrams can be difficult for
users to comprehend and interpret.

Represent the soil features to an acceptable degree of accuracy.
However, we were more interested in developing a method than in
refining the accuracy and precision of the model.

Have the ability to be combined with GIS datasets.

The most imperative objective of this project is the integration of GPR data to a

GIS platform.

We used GPR information to create both vector and raster models viewable in

GIS. The processing and modeling methodology were developed specifically for this

research and for the site-specific conditions at the data collection site. The methods used

were intended to produce an accurate model of subsurface soil horizons while

maintaining efficiency in their production.














CHAPTER 2
GROUND-PENETRATING RADAR (GPR)

Background

The history of GPR in research spans over 70 years. The first practical use of

GPR was by Sterns in 1929 to determine the depth of a glacier in Austria (Olhoeft 1996).

Advances and uses lay virtually dormant until the United States Air Force became

interested in developing better radar systems in the late 1950s. Several pilots crashed

while attempting to land on icy runways in Greenland because the Air Force radar was

seeing through ice and reading inaccurate altitudes. In 1972, GPR was sent to the moon

on the Apollo 17 mission, using an apparatus similar to that used in Sterns' glacier survey

in 1929. Before 1972, GPR was not commercially available. However, in 1972,

Geophysical Survey Systems Inc. (GSSI) was founded and began to sell the first

commercially available GPR units. Today, nearly 300 patents are associated with GPR

technology, and many universities and private companies use GPR technology (Olhoeft

2000).

Used in such varied disciplines as soil science, geology, archeology, engineering,

construction, mining, military, agriculture, and environmental science, GPR has proven

to be a valuable and diverse instrument. Soil scientists and geologists are able to discern

subsurface anomalies and formations (Collins 1990); archeologists are able to locate

buried artifacts (Basile et al. 2000); engineers and the construction industry are able to

quickly locate utility pipes and buried cables (Grandjean et al. 2000); miners are able to

more efficiently retrieve valuable minerals and ore (Sigurdsson and Overgaard 1996); the









military saves lives by locating buried land mines and unexploded munitions (Hyde

1997); environmental scientists are able to track contaminant leaks and hazard waste

movement in the soil (Atekwana et al. 1998); and hydrologists use 3D georeferenced

GPR transects to identify subsurface flow pathways (Gish et al. 2002). In addition to the

myriad of present uses, an international group of scientists plans to conduct GPR surveys

of Mars in the future (Marsal et al. 2000). Ground-Penetrating Radar has proven to be a

useful, accurate, and efficient means of data collection in countless instances, and

continues to be at the forefront of research and technology.

Theory

The function of GPR is similar to reflection seismology, a technique traditionally

used by exploration geophysicists. By sending electromagnetic energy into the earth,

information about the lithology of a site can be determined. Ground-penetrating radar

uses a transmitting and receiving antenna in bistatic or monostatic mode to send a

variable frequency radar signal into the ground. The signal is reflected and sent back to

the antenna. In bistatic mode, the transmitter and receiver antennas are held apart at a

constant distance by a rigid frame. Frequently, antennas in bistatic mode are at zero-

offset, meaning that the point of transmission is also the point of reception, or that both

antennas are held at the same point. In monostatic mode, the transmitting and receiving

antennas are independent of each other and can be used in a variety of arrays to gather

information. Figure 2-1 illustrates the differences between bistatic and monostatic data

collection modes.

Depending on the desired output and soil conditions, several antenna options

exist. Generally, two antenna properties determine the use of the equipment at a

particular site: depth of penetration and resolution. The depth of penetration is the










Monretatic or "zero-offset" position


Pedologic or Geologic Feature




Figure 2-1. Monostatic and bistatic antenna configuration. In monostatic collection
configuration, the transmitting and receiving antenna are located in the same
position. In bistatic collection configuration, the transmitting and receiving
antennas are separated by an offset distance determined by the user.



deepest depth at which a GPR signal will be reflected back to the surface. Resolution is

the ability to differentiate features of different sizes from the surrounding media. Smaller

features can be identified by an antenna providing high resolution than by an antenna

providing low resolution. While depth of penetration and resolution are very strongly

correlated with antenna frequency, their relationships are different. As antenna frequency

increases, depth of penetration decreases and resolution increases (Figure 2-2).

Conversely, as frequency decreases, depth of penetration increases and resolution

decreases. As the objectives of a project change, so might the choice of radar equipment

for that objective.


Bistatic position












U U


i0 I IH-

400 MHz

500 MHz


Figure 2-2. Relative depth of penetration and resolution of various antennas, based on a
scale of 1 to 10. Generally, antenna frequency is directly proportional to
resolving capability and inversely proportional to depth of penetration.



Gathering as much information about the site a priori allows for an informed

equipment selection. Specifically, knowing the depth of possible features of interest and

existing soil conditions that may inhibit radar penetration will aid in the selection of

proper equipment.

Ground-penetrating radar antennas collect information from the subsurface using

electromagnetic waves. A voltage pulse is sent into the ground by the antenna, beginning

the propagation of a radar wave that travels downward through the soil. As changes in

electrical properties of the soil occur, part of the electromagnetic energy is reflected back

to the surface, where the receiving antenna amplifies the signal and stores the data

(Radzevicius et al. 2000). The output is a measure of the strength of the









electromagnetic wave, or amplitude. Ground-Penetrating Radar is able to measure the

time between wave propagation and reception, which can be converted to a depth below

the surface. In this fashion, amplitude values can be assigned a specific depth below the

transmitting antenna. If the data collection process is repeated over a short distance, an

image of the subsurface can be constructed from the amplitude values and their

associated depths stored and displayed by the GPR equipment (Knoll 1996).

There are two main parameters to be entered by the user for GPR equipment:

dielectric constant; and two-way travel time, or range. The depth of penetration for an

antenna can be greatly augmented by proper determination and entry of these user

parameters. Materials that are dielectricc" have a low dielectric constant and will permit

electromagnetic energy to pass through without dissipation. The more

electromagnetically conductive a material, the less dielectric it is. For maximum depth of

penetration, material should be highly dielectric with low electrical conductivity

(Conyers and Goodman 1997).

The range of the GPR is the amount of time the system will record received

reflected signals after propagation (Geophysical Survey Systems Inc. 1999). For

example, a wave propagated at time 0 is partially backscattered at 5 nanoseconds (ns)

after transmission, and reaches the receiving antenna at 10 ns. The unscattered portion

continues downward, is again partially backscattered at 10 ns, and reaches the receiving

antenna at 20 ns. If the range of the GPR equipment is set at 15 ns, only backscattering

that reaches the receiving antenna in less than 15 ns will be recorded. As the range of the

GPR is increased, more time is allowed for the wave to penetrate and reflect, thus

increasing the total depth of penetration. However, a point occurs at which increasing the

range no longer provides any usable data, because the entire signal has been









backscattered or lost. A generic example of GPR scattering is illustrated in Figure 2-3.

The dielectric constant of a material is a physical property of the soil and the

environmental conditions at the time the survey is conducted. It must be properly set

during data collection to achieve accurate results. Dielectric constants of various

materials are shown in Table 2-1. The dielectric constant, dependant mostly on soil

moisture, is a specific value that must be determined or approximated at each site.

A common method used to determine the dielectric constant in the field is to bury

a target at a given depth and position the radar antenna directly over the target. Then,

adjust the dielectric constant of the GPR until the target appears at the correct depth in the

GPR profile. In the absence of a target, contrasting soil horizons, such as argillic

horizons, can be used if their exact depth is known.

Site Suitability

The suitability of a site for GPR study is quite variable. While climate, geology,

relief, and general feasibility play a large role, the most important aspect of site suitability

is the soil. Soils that have a high electrical conductivity dissipate the radar signal and do

not provide a good medium for GPR studies. In general, the soil properties that affect the

electrical conductivity of a soil are the cation exchange capacity (CEC), amount and type

of clay, amount of dissolved salts, and moisture content (Doolittle and Collins 1995)

The CEC of a soil is the sum of the total exchangeable cations that can be held by

a soil (Thomas et al. 1985). Since most cation reactions take place on the surfaces of

colloidal particles and organic matter, soils that are high in clay or organic matter tend to

have a higher CEC. Sandy soils typically have a low CEC because coarse-textured soils

are necessarily low in clay and organic matter. As the CEC of a soil increases, the

suitability of the site for GPR drastically decreases, due to a decrease in depth of less than













Transmitter Receiver


Fine Sand







Sandy Loam





Clay


Figure 2-3. Ground-Penetrating Radar wave propagation, scattering and reflection. As a
GPR wave encounters horizons of differing electrical properties, the wave is
partially scattered. Parts of the backscattered energy are lost to the
environment, but portions are also reflected back to the receiving GPR
antenna. As more energy is scattered, less energy is available to penetrate
into the ground. sy is the dielectric constant.









Table 2-1. Dielectric constants of various materials

Air 1.0 Wet Sandstone
Snow 1.5 Wet Granite
Dry Loamy/Clayey Soil 2.5 Travertine
Dry Clay 4.0 Wet Limestone
Dry Sands 4.0 Wet Basalt
Ice 4.0 Tills
Coal 4.5 Wet Concrete
Asphalt 5.0 Volcanic Ash
Dry Granite 5.0 Wet Sands
Frozen Sand & Gravel 5.0 Wet Sandy Soils
Dry Concrete 5.5 Dry Bauxite
Dry Limestone 5.5 Saturated Sands
Dry Sand & Gravel 5.5 Wet Clay
Potash Ore 5.5 Peats
Dry Mineral/Sandy Soil 6.0 Organic Soils
Dry Salt 6.0 Sea Water
Frozen Soil or Permafrost 6.0 Water


6.0
6.5
8.0
8.0
8.5
11.0
12.5
13.0
15.0
23.5
25.0
25.0
27.0
61.5
64.0
81.0
81.0


less than 1 meter of penetration is possible (Olhoeft 1999). This is also the case with

soils high in other smectites.

Different clay minerals have vastly different CEC values. Typically, smectites

have a high CEC and severely retard the depth of GPR penetration. Smectites, which are

2:1 expanding phyllosilicates, exhibit a high CEC due to isomorphic substitution of

Mg+2 for Al+3 in the dioctrahedral layer. This creates a negative charge that must be

satisfied by cations. Smectites also have an inherently high internal surface area,

allowing for more cationic exchanges. In contrast, soils with kaolinite clay do not pose

as much of a problem for GPR survey. Kaolonite is a 1:1 nonexpanding phyllosilicate,

does not undergo isomorphic substitution, and has a small surface area (Sumner 2000).

Clay types and their CEC values are listed in Table 2-2.

The amount of salts dissolved in the soil solution is a major contributing factor to

the electrical conductivity of a soil. Dissolved salts such as calcium carbonate and

sodium chloride dramatically increase the electrical conductivity of soils (Doolittle and









Table 2-2. Cation exchange capacities of soil materials.

Smectite 80-120
Vermiculite 100-180
Fine Mica 15-40
Chlorite 15-40
Kaolinite 2-5
Humus 100-550


Collins 1995). The correlation is so significant that electrical conductivity is used as a

measure of salt content for site assessment and the determination of saline and sodic soils

(United States Department of Agriculture Natural Resources Conservation Service

2001).

Data Structure

It is important to understand the fashion in which GPR data are organized in order

to proceed with processing and quantitative analysis steps. A GPR profile is divided into

"traces" and "samples". A "trace" of data is made at a specific X and Y location each

time the GPR generates an electromagnetic pulse. A "sample" is a data point along the Z

(depth) axis of each trace at an interval of depth. The GPR can either record 512 or 1024

samples per trace. Depth, X, Y, and amplitude are all attributes associated with each

"sample". When filters are applied to GPR data, they can be applied to a group of traces

or a selection of samples. A graphic representation of "traces" and "samples" is

illustrated in Figure 2-4.













AMPLITUDE
-32768
1 -28672
-24576
-20480
-16384
-12288
S-8192
-4096
0
4096
8192
12288
16384
20480
24576
28672
32768

Figure 2-4. Ground-Penetrating Radar traces and samples. A GPR profile (left) can be
divided into traces and samples (right). In the electromagnetic waves (right)
each black line represents a trace, and each red square represents a sample.
A trace exists at the X, Y location at which the GPR sends an
electromagnetic pulse. A defined number of samples are taken for each
trace, either 512 or 1024. Each sample contains X, Y, Z, and amplitude
information.














CHAPTER 3
GEOGRAPHIC INFORMATION SYSTEMS (GIS)

A GIS is a system of computer software, hardware and data, and personnel to help

manipulate, analyze and present information that is tied to a spatial location (Ormsby et

al. 2001). Real world features with spatial coordinates are represented by layers and may

be displayed in combination with any other spatially referenced data.

History

Modem GIS began in the 1960s, with the help of emerging computer technology

and technicians. However, the history of GIS is unique in that it developed nearly

concurrently by separate research teams from different locations and backgrounds. One

of the earliest accounts of a computerized GIS is the Canada Geographic Information

System (CGIS) developed in the early 1960s. Meanwhile, the Harvard Laboratory for

Computer Graphics and Spatial Analysis (HLCGSA) created their automated mapping

application, SYMAP, which served as the training ground for many of the scientists that

developed and created the precursors to the popular consumer GIS packages used today.

At the same time, the Census Use Study (CUS) in New Haven, Connecticut was

beginning "computer mapping experiments" to use in the 1970 U.S. Census. (Mark et al.

1988).

The CGIS was the first GIS to be used and began with The Federal Department

of Agriculture land inventory describing agriculture, forestry, wildlife, and recreation

capabilities across the lower one-third of Canada. The materials used to create the

digitized maps were the first of their kind, and designed specifically for the land









inventory. The drum scanner used in the project to digitize aerial photography, for

example, is now in the National Museum of Science and Technology in Ottawa

(Tomlinson 1988).

The HLGCSA was important for the academic development and experience it

offered the students. Though the laboratory was situated at Harvard University,

collaborators from nearby institutions such as Yale University and the Massachusetts

Institute of Technology contributed in the research and directives of the center. Because

these institutions did not have an organized geography department, the early focus of the

HLGCSA research was primarily directed toward practical applications in landscape

architecture, urban and regional planning, and resource management.

To prepare for the 1970 U.S. Census, the newly created Census Advisory

Committee on Small-Area-Data established the CUS in 1966 to develop methods of

computer mapping. The programmers associated with this project developed the method

to incorporate topology, the spatial relationships between connecting or adjacent vector

features (Environmental Science Research Institute 2002), into municipal data that

became the core of current spatial data formats such as USGS Digital Line Graphs,

Spatial Data Transfer Standard (SDTS), and the polygon vector layers used in Arc/Info.

While the 1960's served as the decade of development for GIS, the 1970s were

years of "lateral diffusion" (Tomlinson 1988). More universities and government

agencies became interested in the technology, expanding the user base of GIS. Adding to

this user base was the Environmental Systems Research Institute (ESRI) in Redlands,

California. Founded in 1969 by Jack Dangermond, formerly of the HLGCSA, ESRI has

become arguably the most widely used GIS package in use. While ESRI was primarily a

consulting group that happened to develop software in the 1970s, computers became









cheaper, smaller, and more powerful. By the time the mid-1980s arrived, users were

"pounding at ESRI's door" requesting their software. In response, ESRI released

ARC/INFO, the GIS package upon which ESRI's current models are based.

(Dangermond and Smith 1988). The success of ARC/INFO turned ESRI into the

internationally known GIS developer and manufacturer that it is today.

Georeferenced Data

Raw data must be digitized and georeferenced to be incorporated into a GIS

platform. Digitizing is the process of encoding geographic features in digital form as x,

y, and z coordinates (Environmental Science Research Institute 2002). Digital formats

may be either vector or raster. Vector data represents information about the earth in the

forms of points, lines, or polygons. For example, a sinkhole location may be represented

as a "point", interstate highways as "line", and a lake as a "polygon". Figure 3-1

illustrates the three types of vector data. Raster data is information about the earth

represented by equal-sized cells. For example, population density, elevation, and land

use are frequently represented as raster data. Satellite and aerial photography is also

considered raster data. Figure 3-2 is an example of elevation represented as raster data.

Georeferencing is the process of referencing real world data to geographic

coordinates, such as longitude and latitude. These known coordinates will place the

feature at the correct location on the earths' surface. Frequently, Global Positioning

Systems (GPS) are used to correctly georeference the datasets. Once correctly

georeferenced, a dataset can be imported to a GIS to be part of any spatial analysis

procedures. Furthermore, spatially referenced data can be analyzed for interpolation by

using kriging and other geostatistical methods (Ovalles 1988).

Depending on the objective of a project, several GIS layers may be used in





























zI^-'


Legend


- Interstate
Cities
Conservation Land


112.5


225


450 Kilometers


A


Figure 3-1. Three types of vector data: points, lines, and polygons. In this image,
cities are represented as points, interstates as lines, and conservation land as
polygons.
















































Elevation (m)
Value N
High : 94.5 0 0 2 0.4 0.8 Kilometers

Low: -5.3

Figure 3-2. Raster data consisting of equal sized cells. Each cell contains a single value
representing real world data. In this case each cell contains the average
elevation for the cell area. The elevation values are then color coded to
enhance the visual display. The elevation data are a subset of the entire
state.









combination to create a visual output to help in the interpretation of the site. The state of

Florida is unique in the amount and availability of free, digital, georeferenced data easily

importable to a GIS. For this reason, data which can be imported to a GIS becomes much

more useful because it can be combined with other data from a wide array of scientific

disciplines. For example, soil data can be instantly combined with elevation, land use,

geology, vegetation, civic, and political information for an area of interest.

File Formats

Vector and raster data are stored in a variety of file formats. The two main types

of vector files are shapefiles and coverages. While coverages necessarily contain

geographic projection information, shapefiles do not. Coverages also contain topology,

i.e. the spatial relationships between connecting or adjacent features. Shapefiles contain

no topologic information. Raster data can be stored in several formats, e.g. discrete of

continuous rasters. Imagery such as satellite or aerial photography is usually stored in

completely different formats than raster data representing elevation or landuse. JPEG,

TIFF, and Mr. Sid are all commonly used file formats for imagery.

Analysis

Multiple analysis procedures can be performed on both raster and vector data

entered into a GIS. Selection of spatial features by querying specific characteristics of

the data, called attributes, is a widely used procedure. For example, all well-drained soils

can be selected and removed from a dataset containing soil mapping units for a county.

Spatial analysis procedures commonly used include proximity analysis and

geoprocessing steps. Spatial analysis procedures act upon the geography of a feature

rather than its specific attributes. Proximity analysis, or buffering, is a frequently used

spatial analysis procedure that creates a zone of specified distance around a vector feature






19


of interest (Environmental Science Research Institute 2002). For example, proximity

analysis can determine the features of a dataset that are within one mile of a selected

interstate. Geoprocessing steps use two or more themes to create a new theme. For

example, a shapefile of soils for the state of Florida can be clipped by a Marion County

boundary shapefile to create a new dataset containing soil data for only Marion County.














CHAPTER 4
GLOBAL POSITIONING SYSTEMS (GPS)

Global Positioning Systems (GPS) provides a link between GPR data and GIS.

The Global Positioning System is composed of a constellation of 24 satellites that

provide accurate spatial coordinates to users worldwide. Satellites are convenient

because the user does not need to have line of sight to the satellite, only a clear view of

the sky. The GPS system is owned and operated by the United States Department of

Defense (DoD) and can be accessed and used by civilians for free (Trimble Navigation

Limited 1998). To access the system, a GPR receiver must be used to communicate with

the satellites.

Initially, the DoD scrambled the GPS satellite signals, creating a large error. Only

military GPS units were able to decode the error, called selective availability, and receive

the decoded and more accurate measurement. The error associated with selective

availability is approximately 100 meters. During the Gulf War, the United States military

used GPS units to track troop and equipment movements. However, not enough military

GPS units were available to satisfy demand, so civilian GPS units were employed. In

order for these commercial units to achieve similar accuracy to the military units,

selective availability was turned off. After the Gulf War was finished, President Clinton

ordered selective availability to be turned off permanently. This action allowed civilians

to use the military satellites to receive accurate measurements anywhere in the world.

Today, many users use handheld GPS receivers to access the satellites and receive

spatial coordinates. While handheld receivers are inexpensive and portable, their margin









of error, particularly in the vertical plane, is too large for detailed research projects.

Typically, the vertical error in a GPS measurement is 1.5 to 3 times as great as the

horizontal error. The Garmin etrex VistaTM handheld GPR receiver, for example, has a

horizontal error of approximately 15 meters (GARMIN Corporation 2001). If a more

accurate measurement is necessary, a GPS receiver capable of differential correction

should be used.

Differential correction is the precise measurement of the relative positions of two

receivers tracking the same GPS signal. One of the GPS receivers must be placed over a

known coordinate, and is termed the base station. The base station determines what error

is inherent to the satellite signal and transmits corrections to the other GPS receiver,

called the rover. When using Real-Time Differential Correction (RTDC), the corrections

are sent from a base station to a rover via FM radio signal. Many base stations are

maintained by government agencies such as the coast guard. Differential correction can

also be accomplished by post-processing the GPS data gathered in the field. Base station

corrections can be downloaded and applied to the rover data by matching the exact time

of the measurements, which is sent by the satellite. By applying differential correction,

whether in real-time or post-processing, the error in measurement can be reduced to less

than 1 meter (Trimble Navigation Limited 1998).

The error in a GPS measurement can be attributed to four sources: atmospheric

delay, obstructions, mulitpath, and human error. As the satellite signal travels through

the ionosphere and trophosphere, the signal can be bounced. The time the signal takes to

reach earth is altered by this bouncing, and is called atmospheric delay. Obstructions

obscure satellite signals by blocking the line of sight to the sky and can be large trees,

buildings, or other solid objects. Multipathing occurs when a satellite signal is reflected






22


from an object, such as a building wall, and is still received by the GPS receiver. Human

error occurs in equipment configuration, equipment setup, and equipment use (Trimble

Navigation Limited 1998).














CHAPTER 5
FIELD SITE AT PINE ACRES

The research site used for this project is located at the University of Florida

Institute of Food and Agricultural Science Plant Science Research and Education Unit

located near Citra, FL in Marion County, and commonly referred to as Pine Acres. The

facility contains over 400 hectares of land for university research in plant science,

turfgrass science, soil science, agronomy, and other related disciplines. The location of

Pine Acres is shown in Figure 5-1, the specific plot in 5-2, and the coordinates in Table

5-1.

Soils

The site used for this research is part of a larger plant science research facility,

and detailed information about the soil has been gathered in recent years. Properties of

the soil determined by mapping and laboratory analyses show the soils to have suitable

properties for GPR use (Doolittle and Collins 1995). These properties include

The texture of the non-argillic horizon soil is very sandy, with very little
clay content (Thomas et al. 1979). In several places, a fine sand texture
exists for several meters.

The soil morphology is distinguishable by use of GPR. Argillic horizons
can be separated from overlying sandy horizons when proper conditions
exist (Doolittle and Collins 1995).

The water table is quite deep at this location in Marion County, in part due
to the drought of recent years. In this condition, the water table cannot be
confused with a soil horizon.

The amount of dissolved salts is low in these soils.

Fortunately, a large scale soil survey had been completed on the Pine Acres










































N
0 115 230 460 Kilometers

Figure 5-1. Location of research site in Marion County, Florida. Pine Acres is located
in northern Marion County, south of Orange Lake.






































0.5 0 0.5 1 Miles
I Im


N


W E


Figure 5-2. Research plot at Pine Acres. Located in field 5a, the research plot was
approximately 320 meters x 160 meters.


= Research Plot








Table 5-1. Corner coordinates of the research plot. The projection system used was
UTM, NAD 1983, Zone 17 North.


Southeast 386666 3253871

Southwest 386988 3253875

Northeast 386667 3254032

Northwest 386990 3254030


property. Alfisols, Ultisols, and Entisols occur on the property. The argillic horizon, as

in Figure 5-3, is quite variable in its depth, texture, and base saturation, providing for the

presence of both Alfisols and Ultisols (M.E. Collins, personal communication, 2001). In

areas where no argillic horizon is present, Entisols exist, due to the lack of other

diagnostic subsurface horizons. Soil profile descriptions taken from the study area are

located in the Appendix.

The site used for this research is located in the Arredondo-Sparr-Tavares

Association, Figure 5-4, which is characterized by nearly level to sloping soils, and sandy

to a depth of less than 100 cm. Arredondo is a well-drained loamy, siliceous, semiactive,

hyperthermic Grossarenic Paleudult, Sparr is a somewhat poorly drained loamy,

siliceous, subactive, hyperthermic Aquic Arenic Paleudult, and Tavares is a moderately

well drained hyperthermic, uncoated Typic Quartzipsamment. Depending on drainage

and the presence of an argillic horizon, the soils at the field in which the GPR transects

were collected are Sparr, Millhopper, or Adamsville. Millhopper soils are loamy,

siliceous, semiactive, hyperthermic Grossarenic Paleudults, and Adamsville soils are

hyperthermic, uncoated Aquic Quartzipsamments (Thomas et al. 1979).

Geology

















-- ~rj
e ~


Figure 5-3. Argillic horizon at Pine Acres Research Facility. The presence of an argillic
horizon in this profile places this soil in the Ultisols or Alfisols soil order,
depending on the base saturation of the soil. The red line indicates the
approximate top of the argillic horizon.














Research Site




























General Soil Association Map Unit Names
ARENTS-MATLACHA-HYDRAQUENTS POMONA-EAUGALLIE-MALABAR
ARREDONDO-SPARR-TAVARES POMONA-MYAKKA-WAUCHULA
BLICHTON-FLEMINGTON-KANAPAHA SAMSULA-HONTOON-EVERGLADES
CANDLER-ASTATULA-TAVARES SMYRNA-IMMOKALEE-BASINGER
CANDLER-TAVARES-ASTATULA SMYRNA-MYAKKA-IMMOKALEE
FELDA-CHOBEE-KALIGA TAVARES-ZOLFO-SATELLITE
FLORIDANA-RIVIERA-TERRA
HYDRAQUENTS-UDORTHENTS-ARENTS TERACEA
MILLHOPPER-SPARR-LOCHLOOSA WAAE
^ WATER


0 12.5 25 50 Kilometers

Figure 5-4. General Soil Associations Marion County, Florida. The Pine Acres
research site is located on the Arredondo-Sparr-Tavares Association
(Thomas et al. 1979).









The geology of Marion County has been well documented and mapped. Sands

deposited in the Pleistocene epoch cap the Hawthorn Group Formation (HGF), a clay-

rich Plio-Miocene-age deposit. In many areas the Ocala Limestone of Eocene age is

present directly beneath the HGF. In many cases, the argillic horizons present in soils of

this area are the upper portion of the HGF, forming a lithologic discontinuity with the

Pleistocene sands.

Frequently, phosphate nodules are present at the interface between the Pleistocene

sands and the HGF. Ranging in size from less than 1 cm to over 50 cm in diameter, the

phosphate nodules are sand grains cemented together by phosphatic minerals, primarily

wavellite (Harris 2002). Though large phosphate nodules are frequently identified at

Pine Acres, none were encountered at the specific site of research.

The HGF consists of mixed lithology. Consisting of interbedded phosphatic clay,

sand, dolomite, and limestone, the formation is frequently referred to as undifferentiated

in origin. The primary diagnostic feature of the HGF is the presence of phosphatic

material in the sediment (Lane and Hoenstine 1991). Frequently, the HGF exhibits a

blue-gray hue in the field, differentiating it from a pedogenic argillic horizon.

The nature of the geology and the soils provide an excellent opportunity for GPR

research. The clayey sediments contrast the overlying sands. Clay content increases by

as much as 20% or more across the interface. Furthermore, the water holding capacity of

the clay is much higher than the coarse-textured sand (Thomas et al. 1979). In addition,

smectite is a typical component in the clay fraction of the HGF. While higher water

content and presence of smectite do not provide great depth of penetration, they provide a

significant contrast in electromagnetic properties from the overlying coarse-textured

pleistocene sand material.









Climate

Climatic conditions of the past few years in Marion County are a departure from

normal, especially with regard to hydrologic issues. The Modified Palmer Drought Index

(MPDI) in Figure 5-5 illustrates the climatic pattern over the last 4 years. The MPDI is a

modification of the Palmer Drought Index indicating the severity of a wet or dry spell.

This index is based on the principles of a balance between moisture supply and demand.

The index generally ranges from -6 to +6, with negative values denoting dry spells and

positive values indicating wet spells. PMDI value ranges can be interpreted to mean 0 to

-0.5 = normal; -0.5 to -1.0 = incipient drought; -1.0 to -2.0 = mild drought; -2.0 to -3.0 =

moderate drought; -3.0 to -4.0 = severe drought; and greater than 4.0 = extreme drought

(National Oceanic and Atmospheric Association 1994). As illustrated in Figure 5-5,

Marion County has been under drought conditions for most of the past 4 years. Though

the MPDI does indicate relatively normal conditions for the past 12 months, the climatic

impact on the soils at Pine Acres is still one of drought.

The precipitation pattern, represented by the Palmer Z index of precipitation

(PZP) in Figure 5-6, of the Pine Acres area for the last 4 years provides more data to

support the drought conditions experienced in the field. The PZP is a monthly

generated value that can be expressed as the "Moisture Anomaly Index."

Each monthly Z value is a measure of the departure from normal of the moisture

climate for that month. This index can respond to a month of above-normal precipitation,

even during periods of drought (National Oceanic and Atmospheric Association 1994).

Over the course of the GPR investigation, several auger holes were dug to a depth of 2

meters of more. In no instance was a near surface or perched water table encountered.

This absence of a soil water table in the research area provided less interference for the










GPR signal.



7000 -

6 000
5 000
4 000

3 000
2 000
1 000 III I
Sooo I I


-2 0000 I 1


-3 000
-4 000
-5 000
-6 000


Month and Year

Figure 5-5. Modified Palmer Drought Index. This data represents the MPDI state
division number 3 for the state of Florida from 1998 to 2002.




Topography

The amount of relief at the Pine Acres property is minimal. Agricultural

production in past years has produced a fairly flat to slightly rolling landscape. However,

topographic changes are present in microrelief, slight variations in the height of a land

surface that are too small or intricate to delineate on a topographic or soils map at

commonly used map scales (Soil Survey Division Staff 1993). The microrelief is

particularly pronounced in areas of karst activity, such as sinkholes.











8 000
7 000
6 000
5 000
4 000
3000
2 000 |
1 000 I II I II



S200011 =IIIIIIII-hIIII I1
-3 000
-4 000
-5 000
-6 000
-7 000


Month and Year

Figure 5-6. Palmer Z Index of precipitation. From 1998 to 2002, the majority of Z
Index values were below 0, indicating a water deficit for each month. This
data represents state division number 3 for the state of Florida from 1998 to
2002.














CHAPTER 6
MATERIALS AND METHODS

Data Collection

The GPR transects at Pine Acres were taken in a grid formation, 40 meters apart.

The extent of the grid was 160 m x 320 m. Eight transects were collected south to north,

and four transects were collected east to west. Four additional transects were collected

from the corners of the grid to the center of the opposite baseline.

The GPR used was a SIR-2000 control unit (Geophysical Survey Systems,

Incorporated, North Salem, NH) with a 300 MHz ground-coupled monostatic antenna.

To collect the transects, the antenna was pulled behind a field vehicle running at

approximately 5 miles per hour. The range was set at 60 nanoseconds, with a dielectric

constant of 5. The dielectric constant was estimated by calibrating the GPR to a known

depth to the argillic horizon. Descriptions of soil profiles located within the study area

are located in the Appendix.

GPS measurements were taken with a Pathfinder series differerential GPS

Receiver (Trimble Navigation, Ltd., Sunnyvale, CA). By using real-time differential

correction when collecting GPS measurements, the horizontal error of measurement was

decreased to approximately 1 meter.

Several software packages were used in the manipulation and calculation of GPR,

GIS, and GPS data. ReflexW 2.5.x (Sandmeier Scientific Software, Karlsruhe, Germany)

was used for all GPR post-processing. Tabular manipulation and calculation was

completed in SPSS 11.0 (SPSS, Inc., Chicago). The primary GIS software packages used









were ArcGIS 8.1 and ArcGIS 8.2 (Environmental Systems Research Institute, Redlands,

CA). All raster interpolations were completed within ArcGIS with the use of the Spatial

Analyst, 3D Analyst, and Geostatistical Analyst extensions. RockWorks 2002

(Rockware, Inc., Golden, CO) was the secondary GIS software package employed for 3D

solid modeling. Variogram analysis was completed in S-Plus 6.0 (Insightful Corporation,

Seattle, WA) using the Spatial module. GPS data was imported and converted to GIS

vector data with GPS Pathfinder Office (Trimbe Navigation, Ltd.)

Ground-Penetrating Radar Post-Processing

"There is no single processing, visualization, and interpretive strategy that is
applicable to all datasets."
-Dr. Alan Green, 2002 International Conference on Ground-Penetrating Radar

The data generated from this research are an exemplary case of the quote above.

The desired result was an easily understood 3D soil model representing only two main

horizons: E or Bw and argillic. The nomenclature and designation of the sandy non-

argillic horizon is subjective, depending upon the soil scientists describing the site and

their theory of soil formation. Whether the horizon is called an E or Bw is irrelevant to

GPR data collection. In this study the focus was not on soil classification but the

electrical and textural properties of the underlying soils. The electrical and textural

properties of the soil are physical and do not change with the taxonomy or name of the

soil or diagnostic horizons.

The post-processing steps used to prepare the GPR data for the model generated

an easily understood model, but are not recommended for studies in which accuracy and

precision are integral to the success of the project. At most, the data were processed by

six ordered procedures before they were exported to the modeling software.

A linear correction to the GPR profile was the first process completed. The linear









correction step is needed because the GPR antenna cannot be pulled at a constant speed

across a transect. If the antenna could be pulled at a known constant speed, distances

could be calculated from the elapsed time from start for a given point along each transect.

However, the GPR antenna is pulled at a variable rate. To correct for this inconsistency,

intermediary "marks" were created at 40 meter intervals along each GPR transect line

when collecting data. These marks can be displayed in the GPR software. Knowing the

distance between intervals allows the GPR software to stretch or compress the sections

between the interval marks to create the correct distance.

Next, the GPR profiles were corrected to the proper surface depth. This step was

performed early because radar features typically removed or augmented by other

processing steps are used in determining the correct depth to the soil surface. In the GPR

analysis software, time 0 ns is assumed to be equivalent to a depth of 0 meters,

interpreted as the soil surface. However, the actual soil surface is typically at a depth

equivalent to a time value greater than 0 ns. This phenomenon occurs because of the

architecture of the antenna, as illustrated in Figure 6-1. An example GPR profile before

and after the depth correction has been applied is illustrated in Figures 6-2 and 6-3. The

wave is propagated from the transmitting antenna inside the fiberglass housing. But,

while the transmitting antenna is located inside the fiberglass shell, it is at some distance

above the surface of the earth. This allows the GPR wave to begin propagation through

the air sealed within the fiberglass housing, yielding a very signature uniform reflection

on the GPR output image. The problem becomes more serious with lower frequency

antennas that are physically larger. As the size of the housing increases, the antenna is

removed further from the soil surface. Since the wave is traveling through a

comparatively homogeneous material (air) the GPR reflection is quite uniform, in










contrast to the undulating and varied reflections seen as GPR waves travel through the

soil. The interface between the uniform air reflection and non-uniform soil reflection can

be easily identified and corrected in the post-processing software. In the GPR profiles

associated with the research, the soil surface was at a depth equivalent to 6 ns.



GPR antenna
housing


GPR Antenna


Air filled/
interior










Figure 6-1. Simplified interior of a GPR antenna housing unit. The actual GPR
antenna is located at some point above the actual soil surface. At point A
indicated in the figure, the GPR wave begins propogation. This point is also
designated time 0 by the GPR unit, and becomes the perceived soil surface.
However, the actual soil surface is located at point B, a known time later
than point A. The radar wave travels through the uniform air interior of the
antenna and antenna housing before reaching the actual soil surface.
Analysis software was used to correctly identify the actual soil surface. The
picture is not drawn to scale. The activity of the GPR wave is not intended
to be accurate in wavelength, frequency, or amplitude.



After the soil surface is correctly determined, a background filter, as in Figure

6-4, is applied to all transects. Typically, the upper part of the GPR profile contains

fairly uniform horizontal bands that are of no further interpretive use and are attributed to

radar attenuation or "noise." By applying the background filter, the bands are removed

allowing the analyst a better view of the features within the GPR profile. The filter works











C(4rAWN *WE1I!
iM M0 -C A3 i M0


(Sl 7ta (


M 71


-w -.


I 9*,


(, -


gAP Am
'J A91




.4'fl


Uncorrected depth to surface GPR profile. This profile has not been
corrected to indicate the correct depth to soil surface. The actual soil
surface is indicated by the arrow on the right hand side of the image. The
yellow and orange bands at the top of the GPR profile aid in identifying the
correct surface depth. Because the GPR wave is traveling through static air
inside the antenna housing before entry to the soil, the GPR will show very
uniform horizontal bands while wave is still traveling through air. Once the
wave enters the soil, the signal becomes much less uniform and more
variable. The soil surface exists at the interface between the uniform orange
band and variable yellow band in this picture.


0 10 20


S TAt:E 0EEP[
JSO ___ 4f] ____ 60 ______ 0_


Corrected
oo soil
surface,
depth 0


S.AI, 9, 'j

#AADE
< ^ i1kAAW


AMPLITUDE

,28672


-)27
-24576
-20480
-16eM
-1228
8192
-1092

40%


Figure 6-3. Corrected depth to surface GPR profile. This figure illustrates a GPR profile
corrected to indicate the actual soil surface. In this case, there was a time
difference of 6 nanoseconds between the perceived soil surface and actual
soil surface. The correct soil surface is indicated by the 0.0 depth on the
right hand axis.


Actual soil
surface,
depth 0


AMPLITUDE
-32768
-28672
-24576
-20480
-16364
*12288
8192
-40%
. ..... ._ 0
40%
8192
12288
163684

24576
28672
3276A


Figure 6-2.


I
3"


70 __


i


|









on each profile individually. The software determines the time and amplitude range of

the radar "noise" that appear continuously throughout the profile. That noise is then

subtracted from the traces within the GPR profile. In this fashion, horizontally

continuous energy patterns and radar attenuation features are removed (Sandmeier 2002).

It is important to note that this does not remove horizontally continuous soil features such

as pedogenic horizons. The background filter is a valuable tool able to visualize soil

features that may be initially hidden or suppressed by radar features of no interpretive

use.

The fourth processing step applied was a spectral whitening filter, illustrated in

Figure 6-5. In order to accurately model the argillic horizon from GPR data, the division

between the sand and argillic horizon had to be very distinct. The spectral whitening

filter accentuated and highlighted the argillic horizon, while further defining the

sand-argillic interface. The spectral whitening filter works by applying a series of

complex algorithms to the GPR profile involving narrow overlapping band-pass filters,

amplitude decay curves, and spectral flattening procedures (Sandmeier 2002). This filter

helps correct the scattering of the GPR signal deeper in the profile. Again, while these

processing steps were ideal for this research, they may not applicable to all datasets. The

next processing step was employed primarily to speed calculations and computation of

the data. The sheer volume of data was so immense that further compression of the data

was absolutely necessary to accomplish modeling in a timely fashion. A compression

rate of 10x in both the X and Z directions was decided upon after a trial and error period

of various compressions rates. The 10x rate retained an acceptable level of resolution

while decreasing the data to a manageable size, as seen in Figure 6-6. After compression,

a data observation was recorded every 8 cm in the x-direction, and 10 cm in the










z-direction. After the data were compressed, they were ready to be georeferenced.


1 C \GPRDATA\PROCDATA\PROCDATA\FILE2689.01T / traces 1261 / samples: 512
DISTANCE [METER]
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 it
AMPLITUDE
-28672
-24576





2 C \GPRDATA\PROCDATA\PROCDATA\FILE2689 DOT / races. 121 / samples. 512 0
DISTANCE [METER] 4096
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 1E

10 16384







Figure 6-4 Background filter of GPR profile. This illustration depicts the process of
background filtering, as applied in this research. In the top profile, several
horizontal bands mask the underlying information. By applying a
background filter, a clearer picture of the subsurface becomes available in
the bottom profile.



The starting and ending points of each GPR transect were georeferenced in the

field using differential GPS. However, the process of actually georeferencing the entire

transect occurred in the GPR analysis software. Real world coordinates were entered into

the file header for each GPR transect as the start and end locations. The relative X and Y

coordinates of the file then became absolute x and y coordinates, placeable in the real

world. The direction in which the transect was run (x or y) was also indicated in the file

header. Since linear correction was already performed, as soon as a real world coordinate

was entered into the file header, the entire transect became georeferenced. This approach

will not work for coordinate systems in which angles of latitude and longitude are the

chosen unit of measurement.










0 1.0 to 40 5o 0o 70o a0 so 100 110 120 130 140 150
S....... ... .. Before AMPLITUDE
1 -3276





-1096
mto Ie "0Ici t T o x


.After 8192
0 1 122288






Figure 6-5. Spectral whitening filter of GPR profile. The Spectral whitening filter was
applied to accentuate and highlight the argillic horizon in many profiles.
The filtered output made differentiation of the horizons much easier and
more efficient to model. A scaling factor of approximately 10 was used
when applying this filter.analysis demanded compression of the data. A raw
data file recorded a data observation every 8 mm in the x-direction and
every 1 cm in the z-direction. A single profile of 320 m contained about 2
million records.



All data were exported as an ASCII text file, with four columns. Each row

represented a single observation, with X,Y, Z, and amplitude values, respectively. The

data were exported in this format to ease transfer to statistical and 3D modeling software

packages. Each transect was exported separately, creating 18 different tables of data.

Once the data were correctly analyzed and exported, preparation for 3D model generation

could begin.

Model Generation

Three specific goals were to be obtained from the generation of a 3D model.

First, the model must be easily interpreted. A person with average knowledge of GPR

has difficulty comprehending and interpreting 3D GPR models. Second, the model must

represent the soil features to an acceptable degree of accuracy. Third, and most










sth [ t6j AMPLITUDE
0 10 20 30 40 0 50 65 0 70 t00 10 1 20 130 140 150 ISO

o B Afterf





... 96
0 10 20 30 40 50 leo 10 z 0 100 110 120 130 140 150in the X
and Z directions, as in this gure, provided a compromise between le size92
12288

120480

28672
V2768

Figure 6-6. Compression of GPR profile. A compression procedure was necessary to
minimize the file size of each transect. A compression rate of lOx in the X
and Z directions, as in this figure, provided a compromise between file size
and resolution.



important, the data must be transferable to a GIS software package for display. These

objectives helped to define the modeling procedures and methods for the post-processed

GPR data.

Four different models were created using GPR data. Model 1 consists of

categorical data that predict the presence of sand or an argillic horizon based on GPR

amplitude readings. Model 2 is a numerical raster model that predicted GPR amplitude

values continuously throughout the field. Model 3 and Model 4 are both numerical raster

models that predict the depth to argillic horizon, but differ in their methods of generation.

The GPR processing steps taken, specific objectives, and the quantitative analysis

performed on the data, differentiate all the models.

Data Preparation

The four column ASCII data exported from each GPR profile were first imported

to SPSS 11.0 Statistical Software for Windows using the proprietary import procedure.









Each transect contained approximately 15,000 points of data. A fifth column was added

to indicate the file of origin. However, in order to create the 3D model, all the GPR data

needed to be part of the same table, as in Table 6-1. Using SPSS, the 18 separate GPR

profile tables were combined into one large table containing the data from all GPR

profiles. This table consisted of approximately 260,000 points of data. Once the data

were combined to a common table, a depth conversion was performed on the time

column (Zns) of the table. This step was necessary to create a depth scale (meters) rather

than a less intuitive time scale (ns). An example of the data at this point is displayed in

Table 6-1.

The values in the Zns column were converted from time to depth using equation

6-1. This conversion assumes an average velocity through the soil medium. The average

velocity was calculated from the dielectric constant used while collecting the data in the

field. Equation 6-2 was used to determine the velocity of the medium for use in the depth

conversion. By using both equations, the z dimension is no longer a time scale (ns) but a

depth scale (meters).

D =tp (0.15 (1/s/y) (6-1)

Where D is the maximum depth of GPR signal penetrations, tp is the two-way travel

time, in nanoseconds, Fy is the dielectric constant (Collins 1990)

v = c / V/y (6-2)

Where v is the velocity of the wave in meters/s, c is the speed of light in a vacuum, 3 x

108 m/s, and ay is the dielectric constant (Olhoeft 2000)

In addition to creating a column for depth, a column containing the relative

amplitude difference, or change in amplitude, was created. For each trace, the amplitude

change, AA, is the absolute difference in amplitude between neighboring samples. The










Table 6-1. Example of the tabular data before any statistical manipulation. X and Y are
the horizontal real world coordinates for each point of data, Zns is the time-
depth of the data point in nanoseconds, and Amplitude is the amplitude value
as exported from ReflexW.

File X Y Zns Amplitude
2689 386786.844 3253873.75 0.836 1899
2.203 1981
3.57 -7260
4.938 -9854
6.305 -7353
7.672 -4281
9.039 -6703
10.406 -14475
11.773 2005
13.141 -5364
14.508 4375
15.875 -21584
17.242 -9853
18.609 -32760
19.977 -32760
21.344 -32760
22.711 -32760
24.078 -32760


calculation of amplitude difference intended to capture the rapidity of GPR signal change

when the argillic horizon was encountered. Figures 6-7 shows the distribution of AA

with depth and the relativity to the GPR image in Figure 6-8. It is important to note at

what depth large values of AA begin to appear. Only at depths deeper than the expected

depths to argillic, approximately 1.5 m, do AA values greater than 30,000 appear.Based

on the assumption that changes in soil morphology occur concurrent with large AA

values, AA became the modeling variable. However, AA is not a value with which

average GPR users are familiar.












0-





2 20 -E



0E.


40E






60E

I I I I
0 20000 40000 60000

delta a


Figure 6-7. Scatterplot of AA versus depth for all GPR data. An example illustrating the
distribution of AA with depth for a selected GPR profile. Each point
represents one sample of processed GPR data. The Y axis is the time below
the surface in nanoseconds, and the X axis is AA. Notice that large values
of AA (> 30,000) do not occur at less than 20 nanoseconds below the
surface. The expected interface between the sand and argillic horizon is
between 20 and 40 nanoseconds below the surface. AA was used for model
generation because it captures the interface between the sand and argillic
horizon. The two spikes in AA above and below 0 nanoseconds are residual
noise left from background filtering and depth conversions, and are seen in
Figure 6-8.












AMPLITUOE
32766
-28672
-24576
204380
-16389
-12288
0 10 20_30 40 50 60_70 _8090 100 110_120 _130 _140 150 )2

.10.:. ... .- ... *.,.,., .. .. :: : .










Figure 6-8. Ground-Penetrating Radar profile image described by Figure 6-7.
In this profile the alternating lavender and dark blue lines represent the
argillic horizon from a depth of approximately 1.5 meters to greater than 4
meters. The interface is located between 1.5 and 2.75 meters. The depth
distribution of large AA values in Figure 6-7 quantifies this relationship.














CHAPTER 7
RESULTS AND DISCUSSION

Variograms and Kriging

To create a model of the argillic horizon at Pine Acres, depth to argillic values for

areas between data points collected by GPR needed to be estimated. The procedure used

to fill in these values, or interpolate, was ordinary kriging. Kriging is a family of

interpolation procedures that use both the distance and degree of variation between

known data points to estimate new values. Kriging also provides an estimation of error at

each interpolated point, creating a measure of confidence in the model. Ordinary kriging,

the most popular kriging method, assumes the mean of the data is unknown.

The ordinary kriging procedure made use of variograms that were calculated for

each subset used in Model 2. A variogram summarizes the relationship between

differences in pairs of measurements and the distance of the corresponding points from

each other. A variogram is constructed by plotting the semivariance, or squared

difference in value, between every pair of points in a dataset against the distance the two

points are apart from each other in space. (Wackernagel 1995).

For GPR data, data points physically located close to each other should have AA

values that are numerically close. As comparative points become spaced further apart,

their differences in value will be greater. At some point, however, the increase in

distance no longer causes an increase in the semivariance, and the variogram graph

plateaus. The spatial distance at which this occurs is called the range. It is assumed that

beyond the range, the data does not have any autocorrelation. Autocorrelation is the











degree to which a variable is correlated with itself. The variogram value (semivariance)

at the range distance is called the sill. Theoretically, at a distance of zero the

semivariance is zero. And as distances approach zero, the semivariance should also

approach zero. This is rarely the case. Sampling error and short-scale variability cause

dissimilarities in values that are separated by very small distances. This causes a vertical


"jump" at the origin of a variogram, and is called the nugget-effect (Isaaks and Srivastava

1990). Knowing the sill, range, and nugget allows for a more accurate interpolation

procedure in 3D modeling. The range, sill, and nugget are illustrated in Figure 7-1.


Sill
0


(00






0
0a 0 n







Ca
0-
-objective 4.292475e+015




0 0 50 100 150
o o












semivariance, gamma, at the point at which the variogram graph plateaus.
The range value is the distance at which the variogram plateaus. The nugget
is the Y-intercept value at extremely small distances.




To produce variograms for this dataset, the data was stratified by depth.

variogram graphs of the GPR data were constructed at approximately 10 cm intervals for







48


the research site. AA values were plotted against the distance in y for x,z pairs. Figure


7-2 show the variogram graphs for selected depths of the survey area.


160000090.E00


000


OJ.EOO
000
0100



*E


64000001 EOO



3200000L EOO


~oso
0000
00


o


=70253232
=65180241
=17.13


20 40 60 80 100


8000000.EOO


64000001


0000
0 o


IEOO/"


4800000 .EOO



3200000 .EOO



1600000 .EO0


N=53093394
S=17317353
R=16.65


0 I I I I I
0 20 40 60 80 100
Figure 7-2. Variograms of GPR data at different depths. Red dots are indicative of low
AA, and blue dots are indicative of high AA. N is the nugget value, S is the
sill value, and R is the range value. The x-axis is lag distance, and the y-
axis is semivariance. A) Variogram of GPR data at 30 cm. B) Variogram of
GPR data at 50 cm. Variograms of GPR data at different depths. C)
Variogram of GPR data at 73 cm. D) Variogram of GPR data at 115 cm. E)
Variogram of GPR data at 145 cm. F) Variogram of GPR data at 177 cm.
G) Variogram of GPR data at 250 cm. H) Variogram of GPR data at 302
cm.


1280000(



9600000(











8000000.E00


64000001


0o O


.EDO


4800000 .E00


3200000 .E00


16000000.EO0


2000000a


16000009


120000000.EOO


80000001.EO0


40000000.EO0


N=124243756
S=69140121
R=43.74


Figure 7-2. Continued


Q,- 0 0 %.I-..2
0 o


fnn C


=55680609
15164752
=30.56


000


I !


0.EO0 0o 0o o o



EO0
1D.EOO 0r

.*^


000 000a










400000090.E00


0000000.


' *0 00 Q


32000000. E00


240000000.E00


160000010.E00


N=273107206
S=78817138
R=57.26


8000000a.EO0


80000000.EO0


0 0E000000046


n-o n -


'. lt.. .'F'
F


48000000. E 00


32000000. E00


N=566319393
S=58282864
R=49.62


160000000.E00


Figure 7-2. Continued


I I I I I I I I


.00











1000000000.E00


8000000O


o0 0o fnl


SU 00 ---


9a0 -
0.Ea


U 00*U'*


600000 .EOO


400000000.EO0


N=666772977
S=192783837
R=2.642


200000010.EO0


1000000900. E00


nlo ,- -n nr -


8000000E


G000000o.EOO


e* H* -


N=791339195
S=68515194
R=3 843


40000000.E00


200000000.E00


Figure 7-2. Continued


S00


-









The variograms of the GPR data have particular features in common. First, a

definite pattern in fluctuation of covariance exists at distances greater than the range.

This cyclicity, or hole effect, may be due to heterogeneity of the soil material, and is

particularly evident in the variograms with depth less than 100 cm. Second, the ranges of

the variograms show a steady increase range with depth followed by a sharp decline in

range, as illustrated in Figure 7-3. The range of the variogram steadily increased from

the surface down to approximately 150 cm, at which depth the range drastically

decreased. The depth of 150 cm correlates with the presence of a shallow argillic

horizon, suggesting that autocorrelation is small within the argillic horizon. More

interestingly, the distance for which autocorrelation exists increases from the surface

downward to the top of the argillic horizon, suggesting an increased homogeneity with

depth, until the argillic horizon is reached.

The factor of anisotropy must also be considered when interpolating between

points using ordinary kriging. Anisotropy is a directional influence on the data. For

example, if a pollutant is dumped into a stream, the change in concentration of the

pollutant would be greater from streambank to streambank than upstream to downstream,

due to the flow of the water. The geostatistical analyst has the ability to account for such

anisotropy when performing ordinary kriging calculations. Anisotropy is statistically

manifested within a directional variogram, a variogram that looks in a specific direction,

rather than globally. If the variogram changes rapidly when direction is changed,

anisotropy exists. If no dramatic changes in the variogram exist, the data is isotropic, and

presents no directional influence.

Modeling Parameters

When modeling AA or depth to argillic, specific parameters where entered to













70


60


50


40


0 30


20


10


0--
0


Figure 7-3.


Depth (cm)

Depth versus Range for several variograms. When the range of a variogram
is plotted against the depth at which the data points were collected, an
interesting pattern emerges. The range of the variogram steadily increased
to approximately 150 cm. At deeper depths, the range of the variogram
decreases sharply. The depth of 150 cm corresponds with shallow
occurances of the argillic horizon. The pattern of the range values may be
due to increased homogeneity of the soil material between the surface and
the top of the argillic horizon, and heterogeneity within the argillic horizon.


accurately predict or estimate values, or to achieve a desired result. Model 1 used a

K-means clustering technique to group the data. A K-means cluster analysis designates a

value for each cluster center based on the spread and distribution of the dataset. The user

specifies the exact number of clusters to group the data. For a two cluster example, each

data point is placed into one of two groups based on the numerical distance of the specific

value from the cluster center. As a result, each data point is given an additional attribute

indicating its membership in a cluster. Model 1 used the closest point interpolation

procedure for solid modeling available in RockWorks 2002 for its ability to retain the


* *









integer classification values.

Models 2, 3, and 4 used ordinary kriging to interpolate between data points. The

geostatistical extension of ArcGIS was used to calculate kriged raster surfaces. A

logarithmic transformation was performed on the data to normalize the distribution of the

data before modeling. A constant order trend was removed from the data to allow a more

global interpolation scheme. When evident, anisotropy was accounted for in the

modeling procedure.

Model 1

To continue with the goal that the model be easily interpretable, a classification

procedure needed to be employed to separate the data. This method resulted in the

production of Model 1, a categorical model predicting the presence of an argillic horizon.

Rather than having amplitude or AA with their wide ranges in value as the primary

variable to be modeled, a number indicating a data point as belonging to either a sand

horizon or argillic horizon was desirable. A K-means clustering technique was

performed on the AA values to classify each data point into two groups: sand and argillic

horizon. Figure 7-4 shows that points with a membership in cluster 1 are located closer

to the surface than points with a membership in cluster 2. An example of data that has

gone through all the tabular manipulations and is ready for the modeling procedure is

shown in Table 5.

The construction of the 3D model began after the data had been analyzed,

augmented, and clustered. RockWare 2002 (Rockware, Inc., Golden, CO) was used to

create a solid 3D model from the GPR table containing over 260,000 data points. The

table was saved in SPSS in *.dbf format and imported to RockWare 2002 in order to

comply with the import procedure. At this point, no more tabular manipulations were













o0





-1 .

Mean depth cluster 1
1.71 meters
-2

Mean depth cluster 2
2.61 meters
-3



ol -4
.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2

Cluster Number of Case
Figure 7-4. Mean depth of clustered variables. The mean depth values for the two
cluster classes illustrate their relationship to the soils in the field. Cluster 1,
representing sand, has a mean depth value much nearer to the surface than
Cluster 2, representing the argillic horizon.



performed. The data were ready to become a usable model.

RockWare 2002 creates 3D solid models from X, Y, Z, and attribute data by

creating a 3D grid of equal-sized 3D cells called "voxels". The dimensions of the voxels

can be changed by the user for optimal resolution. The voxel centers, or "nodes", all

have specific coordinates as well as attribute values.

The closest point interpolation method was used to create the 3D solid model

from the cluster membership values. This choice was very important to the output of the

model. If a different model was used, the cluster values would not be retained as an










Table 7-1. Tabular GPR data ready for 3D modeling. FILE designates the original GPR
file, X, and Y are the absolute GPR coordinates, Z_NS is the depth in
nanoseconds for each point, AMP is the amplitude value for each point,
AMPDIFF is AA, Z_M is the depth in meters, and CLUSTER is the cluster
membership value. To create a dummy variable for multivariate analysis,
subtract 1 from the cluster value.


FILE X Y Z NS
2680 386989.6 3253991.00 -4.63
-3.27
-1.9
-0.53
0.84
2.2
3.57
4.94
6.3
7.67
9.04
10.41
11.77
13.14
14.51
15.88
17.24
18.61
19.98
21.34
22.71
24.08
25.45
26.81


AMP
-6
9
-5
9
21
522
796
33
2437
-368
1733
520
1188
-619
659
-7928
-4076
-6644
6325
-2304
717
-32088
5310
32760


AMPDIFF
6
15
14
14
12
501
274
763
2404
2805
2101
1213
668
1807
1278
8587
3852
2568
12969
8629
3021
32805
37398
27450


ZM
0.31
0.22
0.13
0.04
-0.06
-0.15
-0.24
-0.33
-0.42
-0.51
-0.61
-0.7
-0.79
-0.88
-0.97
-1.07
-1.16
-1.25
-1.34
-1.43
-1.52
-1.62
-1.71
-1.8


CLUSTER
1


integer. For example, a K-means cluster value of 0 suggests the data point belonged to

the "sand" cluster, and a value of 1 suggests the data point belonged to the "argillic

\horizon" cluster. With this numbering convention, a decimal value such as 0.65 would

have no connection to the real world; the value 0.65 is neither a member of the "sand"

cluster nor the "argillic horizon" cluster. It was imperative that an interpolation

procedure be used that retained the integer nature of the cluster value. The only method

to provide this function was the closest point algorithm. The closest point algorithm

assigns a value to each voxel that is equal to the nearest data point. Since every data

point used in the model was an integer, the only possible voxel values would be integers.









The end result was a solid model with a value of 0 or 1 for each voxel, representing sand

or argillic horizon, respectively.

The closest point method of 3D solid modeling created a model capable of

interactive viewing in RockWorks, Figure 7-5. However, the model still needed to be

integrated to a more popular GIS package, such as ArcGIS. The entire model was

imported to ArcGIS as a large array of points, and will be discussed in detail in later

sections.

Model 2

Model 2 was constructed as an alternative to the categorical nature and display

problems associated with Model 1. The K-means hierarchical clustering technique and

closest point interpolation method were not used in the development of Model 2 in order

to provide a numerically based model. However, AA remained the predictive variable.

The output of Model 2 is a series of raster surfaces rather than a 3D array of point

values,as in Model 1. However, the values calculated and displayed in model 2 are less

intuitive than the categorical Model 1. Rather than predicting either sand or an argillic

horizon for each data point, Model 2 predicts a AA value for a designated depth. Though

this provides for a much more dynamic and aesthetically pleasing model, the values are

not easy for a person of average GPR knowledge to understand.

To create Model 2, the master table of data values was divided into subsets based

on depth. A new subset was created at 8 cm intervals for the depth of the GPR survey,

resulting in 37 subsets of data. Next, each subset table was added to ArcGIS to begin the

interpolation procedure. Then, ordinary kriging was performed on each subset to

calculate a predicted surface of AA values, creating a cell based raster model for each

selected depth.








































Figure 7-5. Model 1, using RockWorks 2002 software. The red areas in Model 1 are
voxels with a value of 1, indicating the argillic horizon. The purple areas in
Model 1 are voxels with a value of 0, indicating sand.



Ordinary kriging is the most used kriging method and attempts to estimate a value

at a point of a region for which a variogram is known, using data in the neighborhood of

the estimation location. (Wackernagel 1995). The interpolated surfaces give Model 2 a

visually pleasing appearance, Figure 7-6, and do not use categorical data, such as the

clusters in Model 1. However, a model that predicts AA is not that well understood. The

standard error of prediction associated with the kriging procedure of Model 2 is in Figure

7-7.









Model 3

Model 3 was constructed to combine aspects of Model 1 and Model 2. Model 1

predicted an easily understood variable, presence of an argillic horizon, but did not have

the visual appeal of Model 2. Model 3 was designed to not only predict the depth to

argillic horizon, rather than AA, but also provide a model that was easy to interpret.

Model 3 accomplished these goals by reprocessing the GPR profiles. Rather than

exporting a massive amount of data and quantitatively deciding where the argillic horizon

begins, the horizon interface was "picked" in Reflex. By "picking", a horizon or

interface can be traced with the mouse, as in Figure 7-8, and have those traced values

stored. Each stored data point can contain X, Y, Z, and amplitude attribute data. In this

case, the argillic horizon of each GPR profile was "picked", and the stored values

exported to a table. Since the original GPR profile had already been georeferenced, the

picked data points were georeferenced by exporting the X, Y, and Z coordinates.

Each "picked" profile contained approximately 3,000 points, depending on the

length of the transect. The points from each profile were combined to form a master

table, upon which spatial analysis and interpolation could be conducted. But first, the

points were imported to ArcGIS and converted to a shapefile. The Z value associated

with each point was the "picked" depth to argillic at that location. By interpolating

between "picked" points, a surface could be calculated that predicted depth to argillic

across the field. This interpolation procedure was conducted using the kriging function

available in the spatial analyst extension. The result of Model 3, as shown in Figure 7-9,

is a single raster model predicting depth to argillic for the research field. Figure 7-10 is

the standard error of prediction associated with the kriging procedure for Model 3.










Depth


30 cm

80 cm

122 cm

156 cm


223 cm


306 cm
Delta A
348 cm
Value
High :65520

Low : 35


Figure 7-6. 3D display of Model 2 as viewed in ArcScene. Each layer is a kriged
interpolation of AA values from the corresponding depth. A total of 36
layers from 20 cm to 357 cm were calculated. 7 layers are shown in this
figure.













I75 150 I
75 150 300 Meters


Figure 7-7. Standard error of prediction of Model 2. The standard error of prediction of
Model 2 for the kriging procedure is largest in the areas between GPR
transects, and lowest in the areas closest to the transects.


55 cm


156 cm


80 cm


223 cm


105 cm


248 cm


0
N








DISTANCE [METER]
3253901 3253902 3253903 3253904 3:
3253880 3253890 3253900 3253910 3253920 3253930_ 53940









Figure 7-8. "Picked" argillic horizon. A) The argillic horizon has been "picked" in
post-processing software. "Picking" is selecting a series of data points (B)
by tracing a horizon or feature using a mouse.


"" 9 .


r



U


Depth to Argillic (cm)
Value
SHigh 330


0 30 60 120 Meters


Low 68


Figure 7-9. Image of Model 3. Model 3 is a continuous surface of argillic horizon depths
interpolated from "picked" argillic horizon interface depths from GPR
profile data.












I I I I I I I !
0 25 50 100 Meters


.I.


4


+&


+


Standard Error
Value
High : 0.158329
Low: 0.154636



i.<


*


*


*


Figure 7-10. The standard error of prediction for Model 3. As for Model 2, the areas
closest to the GPR transect lines showed a smaller standard error than
those farther away from the GPR transect lines.














CHAPTER 8
MODEL VALIDATION

All Models created were statistically validated using a validation dataset from

Pine Acres. Test holes (Total 36) were evaluated throughout the research field to

ascertain the true depth to argillic and are illustrated in Figure 8-1. At the validation

locations, the depth to argillic ranged from 103.0 cm to 400.0 cm, with a mean value of

210.6 cm and standard deviation of 64.7 cm. This histogram of the values is shown in

Figure 8-2.

Model 1

In Model 1, each data point is either a 0 or 1. A validation method to correlate

these categorical data with numerical depth to argillic measurements had to be developed.

By "reading" the nearest values to a validation point, a qualitative measurement of

argillic depth could be determined. This value could then be compared with the

validation dataset.

In ArcScene, the "column" of points closest to each validation point was

selected using a spatial query to be interpreted. Each "column" of points represented the

soil at that location in O's (sand) and l's (argillic). The depth to argillic was determined

by evaluating each series of integers to locate the depth at which l's were uniformly

presentThe validation procedure for Model 1 is shown in Figure 8-3. The predicted

argillic depths had a mean value of 192.1 cm and a standard deviation of 64.7 cm. The

Pearson correlation coefficient between the observed and predicted argillic values was

0.663 and significant at a = 0.01






































Legend N
O Validation Points o An an ian ho.f. /


* GPR Transects


Figure 8-1. Validation point locations. The validation points were intended to be placed
at locations not covered by GPR transects.










10



8



6



4



2
Std. Dev = 64.72
Mean = 210.6

0 N = 36.00
100.0 150.0 200.0 250.0 300.0 350.0 400.0
125.0 175.0 225.0 275.0 325.0 375.0


DEPTH (cm)

Figure 8-2. Histogram of depth to argillic values of validation points. The histogram has
a skewness of 0.798 and kurtosis of 0.768. The Shapiro-Wilk statistic is
0.950.









Column of Data Model 1

0



0



0 225
0 \CI





Figure 8-3. Validation procedure for Model 1. The column of data closest to each
validation data point was evaluated to determine the depth at which integer
values representing argillic (1) became uniform. The depth at which this
occurred was extracted from the attribute table of Model 1 and correlated
with the depth to argillic for the respective validation point.


Model 2

Model 2 consisted of a series of raster models that predicted AA at depth intervals

from 20 cm to 357 cm. Validating the model required developing a method to compare

AA values with depth to argillic values. Through the use of raster analysis techniques

and regression statistics, a comparison was calculated.

Each validation point overlay a single raster cell for each raster layer of varying

depths. The raster cells contained the interpolated AA value for that depth. First, each

raster cell in each raster layer beneath each validation point was extracted and organized.

By doing this, a table of data was created that listed the predicted AA values at depth

intervals for each validation point. Then, AA was plotted against depth, and a regression

curve was calculated that fit the data distribution. The majority of the validation point









data were fit best by a 3rd order polynomial regression function, with a mean coefficient

of determination of 0.9103. The regression curve was used to calculate the predicted AA

for Model 2, termed AAmod2, at the depth to argillic determined for each validation

point. This process is illustrated in Figure 8-4.

The AAmod2 from each validation point was then plotted against the depths at

which the AAmod2 values were predicted. A 2nd order polynomial curve with an

adjusted r2 = 0.792 and standard error of 5856 was calculated to fit the data distribution.

The interpretation of this equation is somewhat complicated. The regression equation

will return a predicted AA only for the depth at which clay first appears. Likewise, a

depth can be calculated only from a AA value at the depth at which clay first appears.

The data do not support predicted AA values at any depth in the soil other than that at

which clay first appears.

Model 3

In Model 3, a singe raster layer was interpolated from a "picked" depth to argillic

measurements for each GPR profile. A comparison of the predicted depth to argillic

measurements for each validation point was evaluated with the observed depth to argillic

measurement at each location in the raster layer, Dmod3, to determine the degree of

correlation. The Dmod3 value was extracted from the raster cell associated with each

validation point. The Pearson correlation coefficient for the values was 0.630 and

significant at a = 0.01.

Model 4

The validation procedure for Model 4 differed from the previous three methods.

This is because the model was generated from a regression equation developed from the

validation procedures for other models (Equation 8-1). While validating Models 2 and 3,










a statistical relationship was discovered that predicted the depth to argillic at the

validation points with a higher degree of accuracy than either model independently. The

relationship was linear regression Equation 8-1 that used the AAmod2 values from Model

2 and Dmod3 from Model 3.




70000


60000


50000 *


40000

S Argillic at 204 cm
0 30000 AA,,-,,,j = -38500


20000


10000 .

S***
0 50 100 150 200 250 300 350 400
Depth (cm)


Regression equation for validation point, Model 2. For each validation point,
the corresponding raster cells values (AA) from layers at depth intervals
were extracted and plotted against their respective depths. An equation was
calculated that fit the distribution of the data. A predicted AA for the sand-
argillic interface was calculated using the regression equation. For this
point, the sand-argillic interface was determined through ground verification
to be at 235 cm. The corresponding AA value at this depth calculated using
the regression equation is 48610. This procedure was performed for each
validation point.


Depth to argillic= 143.210 + (-.484 Dmod3 ) +

(.00002224 (Dmod3 AAmod2 ))


(8-1)


For this model, adjusted r2 = 0.840 and the standard error of the estimate = 25.85.


Figure 8-4.










To perform the raster calculation, an interpolated raster layer of AAmod2 values was

created using kriging and multiplied by the Model 3 raster layer. Equation (8-1) was

applied to Model 3 and the newly created Dmod3 AAmod2 layer using the raster

calculator function in the Spatial Analyst extension to create Model 4, illustrated in

Figure 8-5.


Depth to Argillic (cm)
Value
High 505


30 60


12o Meters


SLow 190


Figure 8-5. Image of Model 4. Model 4 was created by applying a regression equation to
Model 3. Model 4 accomplishes all the goals desired; the model is
understandable, GIS compatible, accurate to an acceptable degree, and
visually appealing.Every MS Word document uses styles to format
information. To help prevent the unnecessary copying of different styles
into your dissertation, follow these rules of thumb when copying
information:














CHAPTER 9
SUMMARY AND CONCLUSIONS

Data Collection

If a more accurate model is desired, the first change should be in data collection.

The GPR transects for this project were gathered 40 meters apart. Several GPR analysts

have been successful in 3D modeling by using a much tighter spacing. Grasmueck and

Weger (2002) created 3D data from transects collected 10 cm apart. However, the area

covered by these denser grids is much smaller in size than the research field at Pine

Acres. The transect spacing distance is a function of several factors. First, what is the

size of the feature you are attempting to model? Second, how much time are you willing

or able to spend in data collection and processing? Third, do you have the financial

resources to justify the added expenses associated with more data collection and

processing?

This research intended to model the argillic horizon a continuous feature of

infinite size. In addition, the resolution of the model could remain somewhat large. The

final cell size of Models 3 and 4 was 2 m. This may initially seem to be a coarse

resolution. However, it is assumed that the argillic horizon has a uniform depth over a 2

m area.

If more detailed data are required, the amount of time invested in the project

increases accordingly. A previously unstated goal of this project was to determine an

efficient methodology for GPR data collection, processing, and modeling. The goal of

efficiency may have been overshadowed by more quantitative aspects of the project, but









an efficient plan was always a goal. To collect a larger, more detailed amount of data, the

project no longer becomes efficient. For example, if the transect spacing were to be

decreased to 2 m, a total of 240 transects would need to be collected, not including any

transects to be collected at angles through the collection grid. This equates to 4 x 1010

points of data for a 320 m x 160 m field. When conducting a GPR survey, it is important

to collect all the data for a grid at the same time in the field, keeping the soil conditions

uniform for all transects. The soil conditions would undoubtedly change in the amount of

time it takes to collect 240 transects of data. Furthermore, the processing time has been

increased immeasurably. While conducting detailed and time-consuming processing

steps on 18 transects is feasible and quite possible efficient, performing these same tasks

on the amount of data proposed by smaller transect spacings would take an unpredictable

amount of time. Furthermore, it is frequently necessary to revise the collection

methodology and recollect data for the entire field. A total of 4 complete surveys were

completed for this project over a year and a half. Fortunately, each complete survey

could be completed in less than one day. Collecting 240 transects multiple times would

not be an efficient or practical use of time and resources.

While resolution, time, and money are definite variables to be considered before

developing a methodology, increased sampling is most likely the best way to generate a

more accurate model. However, the methods used in data collection for this project were

satisfactory to all involved, and met the initial goals of the research.

Ground-Penetrating Radar Processing

Unfortunately, error is an inescapable variable of GPR collection and post

processing. A considerable source of error is the way in which depth is calculated from

time. To calculate a depth from time, the velocity of the wave must be known. The GPR









allows only a single dielectric constant to be assigned for the entire GPR transect, no

matter how many different soil types or horizons occurred within the transect. The

dielectric constant is directly related to the velocity of the radar wave through the soil.

By allowing only a single dielectric constant to be assigned to a transect, you are

assuming only a single velocity for the radar wave through a variety of different soils.

Though no study determining the amount of error this situation incurs has been

completed, it is certainly a factor when analyzing GPR data by depth.

It is important to understand the context in which this research was completed,

and the overall goals. The research was intended to create a model that predicted depth

to argillic from GPR values only for the field in which GPR data was collected. For this

reason, a variety of GPR filters and processing techniques were used to maximize the

differences between the sand and argillic horizon. In other circumstances, more care may

be needed in applying various filters to the GPR data that augments the raw data. Also,

this research was focused on the relative change in amplitude, or change from one point

to another. The actual values were somewhat irrelevant; the amplitude could be 10 or

100,000. It only mattered what the changes were in those values from one sample to the

next.

Qualitative Methods

Initially, it was planned to develop a completely quantitative method to model the

argillic horizon in the field. Because the results of the quantitative methods were not as

encouraging as initially hoped, a more qualitative approach was taken. The qualitative

nature of the methodology is evident in the verification of Model 1 and the generation of

Model 3.

When validating Model 1, the upper limit of the argillic horizon was subjectively









chosen from the "column" of data. Though this method of choosing the argillic horizon

was not statistical in its nature, a more individualized approach could be applied to each

column. Nevertheless, this quantitative validation is a possible source of error in

Model 1.

In the generation of Model 3, the depth to argillic for each trace was "picked" by

tracing the interface with a mouse. This qualitative approach has specific benefits and

drawbacks. One disadvantage is that the same "picked" data cannot be exactly repeated.

If the same profile were "picked" several times, the difference in "picked" values would

be minimal, but there would always be a difference. A second disadvantage is the way in

which the profile is interpreted. The "picks" could above or below the actual interface,

depending on how the user interprets the GPR image. However, by allowing an

experienced GPR user to "pick" the depths to argillic, there are desirable benefits.

A specific benefit of allowing an experienced GPR user to "pick" the argillic

horizon is that noise or interference within the GPR profile has much less influence.

While noise can be identified and excluded by a user, a statistical method may have

trouble interpreting such a phenomenon and mistake it for a soil feature. Also, an

experienced user may be able to identify an overall trend to the argillic horizon present in

the imagery that may not be as evident using quantitative methods. A third advantage of

qualitatively "picking" may be initially overlooked. By "picking" a horizon, only one

point per trace is exported. With this amount of data, no compression is needed. Without

compression, the full amount of data is available to perform the interpolations, a ten-fold

increase over the compressed data. While a quantitative, repeatable approach was

initially planned and desired, specific benefits of qualitative methods support the

techniques used in this project









Model 1

While Model 1 met the objectives initially defined, the statistical correlation was

not as strong as expected. There are several explanations for this problem. First, the data

in the model are categorical, not numerical. This presented problems in the validation

procedure that resulted in a qualitative measurement of the predictions in Model 1. The

validation dataset contained depth to argillic values, and the data in Model 1 predicted

either sand or argillic at various depths. Furthermore, rather than picking the nearest

"column" of data to a validation point to retrieve a predicted value, developing an

interpolation procedure to take into account the three or four nearest "columns" would be

a better method of extracting the predicted values. Also, if the model were to predict a

numeric value rather than a category, it is believed the model would have been more

successful. For example, rather than predicting sand or argillic, perhaps the model should

predict percent clay for each voxel. Though the model was able to be visualized in GIS,

the visual quality did not satisfy research objectives. The model took on the

characteristics of a pointillist painting. At a distance, the conglomeration of points

appeared to represent a coherent image. However, as the user moves closer, or zooms in,

the seemingly coherent image dissipates into individual points, which are much more

difficult to interpret. This display problem prompted the generation of the other three

models.

Model 2

Model 2 was much more visually appealing than Model 1, but the end value of a

predicted AA was not a good variable to model. Though AA has a definite correlation

with the depth to argillic, modeling AA creates problems with validation. While the

values were numeric rather than categorical, and therefore easier to correlate with depth









to argillic values, determining a relationship between AA and depth to argillic was very

difficult. The result of the validation was a regression equation that proved true only at

the sand-argillic interface of the soil profile. If you know the depth of the sand-argillic

interface, what is the use of the model in the first place? Also, AA is not a variable with

which users are familiar. Because Model 2 did not have a very practical use, the

modeling variable was changed for future models. But, the raster type of model proved

to be a powerful means of display, and was kept for future models.

Models 3 and 4

Models 3 and 4 are very similar in generation and display. Both models predict

the same variable, depth to argillic, and are raster models. Because both models are

similar in their development, similar problems are inherent. The largest portion of error

in the model is probably due to the qualitative method in which the raw data was

extracted from the GPR profiles. Though "picking" the argillic horizon has desirable

benefits, it is still a qualitative practice which carries more error. The specific amount of

error has not been determined.
























APPENDIX

SOIL PROFILE DESCRIPTIONS


Table A-i Description of representative pedons at the research site.


Ap 4\1 Fine sand 0-15 5\0
Egl 6\2 Fine sand 15-60 few 7 5 YR 5\6 redox concentrations
Eg2 7\2 Fine sand 60-115
Eg3 8\1 Fine sand 115-154
Btg1 7\1 Loamy sand 154-165
Btg2 7\1 Sandy loam 165+ 7 5 YR 5\6 redox concentrations

Ap 2\2 Fine sand 0-19 5\7
Bwl 5\4 Fine sand 19-77
Bw2 7\4 Fine sand 77-130 matrix stripping, faint 7 5 YR 5\6
Bt 5\6 Loamy sand 130-170
Btg 7\4 Sandy loam 170-185 common 7\1 redox depletions
2Btg 6\4 Sandy clay loam 185-200 many 7\1 depletions

Ap 3\2 Fine sand 0-17 6\8
Bwl 5\3 Fine sand 17-35
Bw2 6\4 Fine sand 35-85 few faint redox concentrations
Bw3 7\4 Fine sand 85-140 common faint redox concentrations
Eg 8\3 Fine sand 140-160 7/1 redox depletions
Eg and Bt 8\3 Fine sand 160-200 Eg
6\6 Loamy sand 160-200 Bt

Ap 4\1 Fine sand 0-16 7\3
Bwl 6\3 Fine sand 16-85 matrix stripping, few faint redox concentrations, 2 5 YR 5\6
Bw2 7\3 Fine sand 85-105
2Btgl 6\1 Sandy clay loam 105-115 Hawthorn Formation
2Btg2 6\1 Clay 115-200 Clay films, 5\1 10 YR depletions


The soil profile descriptions listed above are an example of the types of soils that


were found across the research plot area. The descriptions were made by Ron Kuehle and


Michael Tischler in 2001. The location in the field refers to the numbering convention


of the sampling locations throughout the field.














LIST OF REFERENCES
Atekwana, E.A., Sauck, W.A., and Werkema, D.D. Jr. 1998. Investigations of
geoelectrical signatures at a hydrocarbon contaminated site. Journal of Applied
Geophysics. 44: 167-180.

Basile, V., Carrozzo, M.T. Negri, S., Nuzzo, L., Quarta, T., and Villani, A.V. 2000. A
ground-penetrating radar survey for archaeological investigations in an urban area
(Leece, Italy). Journal of Applied Geophysics. 44: 15-32.

Collins, M.E. 1990. Applications of ground-penetrating radar. XVII Reunion Nacional
Sobre Edafologia. 24(28): 15-32.

Conyers, L.B., and Goodman, D. 1997. Ground-penetrating radar: An introduction for
archaeologists. Altimira Press, Walnut Creek, CA.

Dangermond, J., and Smith, L.K. 1988. Geographic information systems and the
revolution in cartography: The nature of the role played by a commercial
organization. The American Cartographer. 15(3): 301-310.

Doolittle, J.A., and Collins, M.E. 1995. Use of soil information to determine application
of ground penetrating radar. Geoderma. 33: 101-108.

Doolittle, J.A., and Collins, M.E. 1998. A comparison of EM induction and GPR
methods in areas of karst. Geoderma. 85: 83-102.

Environmental Science Research Institute. 2002. Glossary of GIS terms [Online].
Available at http://www.esri.com/library/glossary/glossary.html (verified 22
September 2002).

GARMIN Corporation. 2001. eTrex VistaTM personal navigator: Owner's manual and
reference guide. GARMIN Corporation, Olathe, KS.

Gish, T.J., Dulaney, W.P., Kung, K.-J. S., Daughtry, C.S.T., Doolittle, J.A., and Miller,
P.T. 2002. Evaluating use of ground-penetrating radar for identifying subsurface
flow pathways. Soil Sci. Soc. Am. J. 66: 1620-1629.

Grandjean, G., Curry, J.C., Bittri, A. 2000. Evaluation of GPR techniques for civil-
engineering applications: study on a test site. Journal of Applied Geophysics. 45:
141-156.

Grasmueck, M., and Weger, R. 2002. Reassessment of local paleocurrent directions in
the Miami oolitic limestone with 3D ground-penetrating radar. Proceedings of the









9th International Conference on Ground Penetrating Radar. Santa Barbara, CA,
29 April to 2 May 2002, pp. 211-216.

Harris, W.G. 2002. Phosphate minerals. p. 637-665. In Dixon, J.B., and Schulze, D.G.
(ed.) Soil Mineralogy with Environmental Applications. Soil Sci. Soc. Am. J.,
Madison, WI.

Hyde, B. 1997. Information Bulletin No. 98-06. USDA, Bureau of Land Management.
Available online at http://www.blm.gov/nhp/efoia/wo/fy98/ib98-06.html (verified
22 September 2002).

Isaaks, E.H., and Srivastava, R.M. 1990. An introduction to applied geostatistics.
Oxford University Press, Oxford, England.

Knoll, M.D. 1996. A petrophysical basis for ground penetrating radar and very early
time electromagnetics: Electrical properties of sand-clay mixtures. Ph.D
Dissertation. University of British Columbia.

Lane, E., and Hoenstine, R.W. 1991. Environmental geology and hydrogeology of the
Ocala area, Florida. Florida Geological Survey Spec. Pub. 31. State of Florida,
DNR, Tallahassee, FL.

Mark, D.M., Chrisman, N, Frank, A.U., McHaffie, P.H., Pickles, J. 2002. The GIS
history project [Online]. Available at
http://www.geog.buffalo.edu/ncgia/gishist/bar harbor.html (verified 10 October
2002)

Marsal, 0., Harri, A.-M., Lognonne, P., Rocard, F., Counil, J.-L. 2000. Netlander: the
first scientific lander network on the surface of mars. Available online at
http://netlander.fmi.fi/pub/NetLanderMissionDescription.pdf (verified 22
September 2002).

National Oceanic and Atmospheric Administration. 1994. Time bias corrected
divisional temperature precipitation drought index [Online]. Available at
http://lwf.ncdc.noaa.gov/oa/climate/onlineprod/drought/readme.html (verified 22
September 2002).

Olhoeft, G.R. 1996. Application of ground penetrating radar. Proceedings of the 6th
International Conference on Ground Penetrating Radar. Sendai, Japan, 30
September to 3 October 1996, pp. 1-4.

Olhoeft, G.R. 1998. Electrical, magnetic, and geometric properties that determine
ground penetrating radar performance. Proceedings of the 7th International
Conference on Ground Penetrating Radar. Lawrence, Kansas, 27 to 30 May
1998, pp. 177-182.









Olhoeft, G.R. 1999. Applications and frustrations in using ground penetrating radar.
Proceedings of the Ultra Wide Band Conference. Washington, D.C., 20 to 22
September 1999.

Olhoeft, G.R. 2000. Ground penetrating radar GRORADART [Online]. Available at
http://www.g-p-r.com/introduc.htm (verified 22 September 2002).

Ormsby, T., Napoleon, E., Burke, R., Groessl, C., Feaster, L. 2001. Getting to know
ArcGIS desktop. ESRI Press, Redlands, CA.

Ovalles, F.A., Collins, M.E. 1988. Evaluation of soil variability in northwest florida
using geostatistics. SSSAJ. 52: 1702-1708.

Radzevicius, S.J., Guy, E.D., and Daniels, J.J. 2000. Pitfalls in GPR data interpretation:
differentiating stratigraphy and buried objects from periodic antenna and target
effects. Geophysical Research Letters, vol. 27. 20: 3393-3396.

Saarenketo, T. 1998. Electrical properties of water in clay and silty soils. Journal of
Applied Geophysics. 40: 73-88.

Sandmeier, K.J. 2002. REFLEXW. Release 2.5. Sandmeier Scientific Sofware, Inc.
Karlsruhe, Germany.

Sigurdsson, T., and Overgaard, T. 1996. Application of GPR for 3-d visualization of
geological and structural variation in a limestone formation. Proceedings of the
6th International Conference on Ground Penetrating Radar. Sendai, Japan, 30
September to 3 Oct. 1996, pp. 39-44.

Soil Survey Division Staff. 1993. Soil survey manual. USDA Handbook 18. U.S. Gov.
Print. Office, Washington, D.C.

Sumner, M.E. (ed.) 2000. Handbook of soil science. CRC Press, Boca Raton, FL.

Thomas, B.P., Cummings, E., and Wittstruck, W.H. 1985. Soil survey of alachua
county, Florida. Soil Conservation Service, U.S. Department of Agriculture.
U.S. Gov. Print Office, Washington D.C.

Thomas, B.P., Law, L., and Stankey, D.L. 1979. Soil survey of marion county, Florida.
Soil Conservation Service, U.S. Department of Agriculture. U.S. Gov. Print
Office, Washington, D.C.

Tomlinson, R.F. 1988. The impact of the transition from analogue to digital
cartographic representation. The American Cartographer. 15(3): 249-261.

Trimble Navigation Limited. 1998. GPS mapping for GIS with asset surveyor. Trimble
Navigation Limited, Sunnyvale, CA.






81



United States Department of Agriculture Natural Resources Conservation Service.
2001. National soil survey handbook, title 430-VI [Online]. Available online at
http://www.statlab.iastate.edu/soils/nssh/ (verified 22 September 2002).

Wackernagel, H. 1995. Multivariate statistics. Springer Verlag, Berlin, Germany.















BIOGRAPHICAL SKETCH

Michael Tischler was born to James Tischler, Sr. and Patricia Davis in 1978 in

Greensburg, Pennsylvania. At the age of 4, Michael moved to Alexandria, Virginia,

where he stayed until 1990. At the age of 13, Michael moved with his mother to

Punxsutawney, Pennsylvania, where he stayed until high school graduation. Michael left

Pennsylvania for the University of Dubuque, Iowa to begin college as an Environmental

Science major. During his sophomore and junior year, Michael transferred to North

Dakota State University and changed his major to Soil Science, with an emphasis on

pedology and morphology. While in North Dakota, Michael worked as a Soil Scientist

Trainee for the Natural Resources Conservation Service during the summers. Upon

graduating with honors from North Dakota State University in 2000, Michael enrolled at

the University of Florida to begin a Master of Science degree. In October 2002, Michael

left to pursue a position at the NASA Goddard Space Flight Center as a government

contractor