GROUND PENETRATING RADAR
RELATED LINKS &
What is Ground Penetrating Radar? |
Data Collection |
Interpretation: Introduction | Interpretation: Grid A | Interpretation: Grid B
Discussion | References Cited
What is Ground Penetrating Radar?
A ground penetrating radar (GPR) unit includes a transmitter, a receiver, and a data collection device (analog printer or digital recorder). The transmitter sends pulses of electromagnetic energy down into the ground, and then a sensor measures the electromagnetic energy reflected back from subsurface anomalies. Such anomalies may include interfaces between soil layers, as well as objects and features with electrical or magnetic properties different from the surrounding soil. Radar information can be displayed and/or recorded as numerical values or as grey scale pixel values, with black and white representing areas of high or low reflections respectively. A major advantage of GPR is that it records vertical soil profiles in detail, rather than generating only horizontal plan maps (as do conductivity and magnetometry). It must be remembered that GPR does not directly image objects in the ground: it produces a two-dimensional record of the three-dimensional waves bouncing off of objects in the ground.
Radar waves propagate out from the transmitter in three dimensions. The use of an antenna shield minimizes the spread of waves up into the air, and limits readings to those coming from below. Radar waves form a cone shape as they move down through the soil (Figure 1). Because a radar transmitter is constantly being pulled along a transect and is constantly generating readings, its conical wave images objects to the front, back, and sides of the antenna. In order to minimize interference from side readings, the long axis of the antenna should be aligned parallel to the direction of antenna movement (which is also parallel to the electric field the antenna generates) (Conyers and Goodman 1997: 36). Linear features which are aligned with the GPR's electrical field will not produce high reflectance values, however. This means that GPR is good at distinguishing linear features only if they run perpendicular to the path of the antenna (Conyers and Goodman 1997: 36).
The Nathan and Polly Johnson house has yard space on all sides, but geophysical prospecting was undertaken only in the back yard for several reasons. The small size of the front and side yards, combined with the modern sheds, walkways, and bushes in these areas, severely reduced the potential efficacy and maneuverability of the prospecting equipment. There is likely more, and more varied, evidence of site occupation in the back yard simply because of its size. Because the side and front areas are so small (only a few meters wide and/or long), archaeological testing alone should intersect any former structures, foundations, pits, plantings, task areas, or other modifications to these spaces.
Even with potential trace stacking at 2x or 4x, resolution is still one trace
every 6 cm or 8 cm (Patrick Ryan Williams, personal communication 2000);
objects 10 cm or larger should be resolved by this antenna. Transect lines
were walked in both north-south and east-west directions on both grids in
order to detect more easily any linear anomalies; all walkovers were begun in
the southwest corner of a grid. Transect spacing was 50 cm. Since the
estimated radar footprint diameter is also 50 cm, any object capable of being
resolved by the 500 MHz antenna should have fallen within the radar's
reflectance cone and should be visible in the data set (Patrick Ryan Williams,
personal communication 2000). It was necessary to turn up range gains in the
far- and, particularly, the mid-field ranges to get easily readable results
(necessary when dealing with analog data, since post-field processing is
limited). It is possible these values were ultimately set too high, however,
since readings seem to be very dense in several areas. This is consistently
true where the mid- and far-ranges interface, at a time-depth of approximately
Transect lines were printed out on a roll of paper rather than recorded digitally because we did not have a digital recorder for the radar unit. This output method produces an immediate visual record of recorded anomalies that is organized spatially, and so can be read and grossly interpreted in the field. Rolls of paper printouts are far from an ideal method of data preservation, however. Further data processing (in the form of filtering and resampling) is highly desirable when dealing with radar information. This is because GPR does not directly image data, but records variable wave reflection intensities. The application of mathematical algorithms to digital data can, in essence, "enhance" it and make it easier to relate to physical anomalies. Data can be computer-processed using filters, resampling, trace stacking, and interpolation, for example. Studies show that appropriate computer processing can overcome problematical ground conditions (wet or highly conductive soils, for example) to produce meaningful results (Conyers and Goodman 1997: 16).
The ability to manipulate data using various computer programs and display technologies is crucial to the interpretation and effective communication of GPR results. Printouts from the GSSI-SIR3 were scanned into a computer at 100 dpi resolution in 256 shades of grey and saved as tagged image format (TIF) files. These raw image files are available on this web site. Anomalies of various shapes, sizes, intensities, and depths are readily apparent in the 50 profiles generated by this radar study, and so minimal processing was necessary to interpret these data in at least a basic way; the results are discussed below in the "Interpretation" section. A research goal of the larger geophysical prospecting project is the generation of comparative datasets, however. Both magnetometry and conductivity results are measured and plotted in the horizontal plane, not in vertical profiles. It was therefore necessary to generate horizontal plots of anomalies from the vertically-sampled radar data.
The conversion of paper printouts to computerized data sets requires innovative processing strategies. Digitizing the images, as described above, was the first step. Several strategies for accomplishing this were discussed. The ideal method would be converting pixel values to numerical values by (1) translating images from TIF to text files, (2) entering these in a spreadsheet program, (3) performing averaging functions on them, and (4) translating them back to TIF files. Because of time considerations, because GPR's relatively high "information density" allows for a good deal of approximation when generating views to compare with the magnetometry and conductivity used, and because the primary interpretative power of GPR lies in its detailed profiles rather than its resampled time-slices, this researcher opted for a less accurate yet visually satisfying, simpler, and less time consuming method. I relied on resampling functions in the image processing program itself to average selected ranges of pixel values.
A 27.5-32.5 nS time-depth window was chosen because these signals were well below the area of near-field interference and the majority of anomalies seemed to pass through this range. Based on an assumed soil RDP of 4-9, this loosely corresponds to a physical depth of 1.24-1.5 meters. The image section corresponding to this time-depth range was (1) selected, copied, and (2) pasted into a new file. Then this image was (3) resized in order to (3a) standardize y-axis distances and obtain averaged values approximately every 10 cm (which is also the antenna resolution in the y direction, the direction in which it was pulled). For example, for the WE walkover of grid A, which had a y-axis value of 4 meters or 400 cm, the image "width" was set at 40 pixels. At the same time, the time-depth slice (the image "height" corresponding to readings between 27.5 and 32.5 nS) was (3b) averaged to a value of 1 pixel. Resampling for this image size change was bilinear. The resulting transect line of values was (4) pasted into another file and aligned with its neighboring transects. This process was repeated for each transect in a data set, resulting in a data resolution of 10 cm in one direction but only 50 cm (the transect resolution) in the other direction. In order to produce a proportionate image, (6) the image "width," which is dependant on the number of grid transects, was multiplied by 5 and resampled bicubically. This resulted in a very small (40 x 90 pixels for grid A) image which was (7) resized and resampled to produce a meaningful plan view (Figures 4 and 5).
Fifty GPR profiles were generated on the Johnson site. Because of the large amount of data, discussion will be limited to the identification and description of major anomalies in grid A and grid B. Synthetic remarks, assessment of research goals, and directions for future research are in the "Discussion" section below.
Because the RDP of the soil is not known, and its estimated range is so large (4-9, as discussed
above in "Data Collection"), correlating time-depth and spatial depth is difficult. A rough value of
a .25 m minimum real-depth for 5 nS time-depth may be derived using the equation depicted in Figure 1
(Patrick Ryan Williams, personal communication 2000). On the radar
profiles, faint horizontal lines mark these intervals every
5 nS beginning at 2.5 nS. Even though time-slice plans for the depth of
1.25-1.5 meters are very approximate representations of in-ground
anomalies, they are a good starting point for interpretation because they
do compare favorably with the radar profiles.
Thumbnails of all profiles from the grid A south-north walkover and the grid A east-west walkover are available on this website. The largest anomalies in grid A seem to occur near the yard margins; no comparably large disturbances are evident in the center of this grid.
Thumbnails of all profiles from the grid B south-north walkover and the grid B east-west walkover are available on this website. Grid B seems to have numerous distinct anomalies with narrow vertical readings, several of which are reminiscent of hyperbolas (Figure 8). This wave shape is typical of reflections from buried objects. Large anomalies that involve different types of soil and fill (pits or cellars, soil stratigraphy) result in linear or irregular reflectance wave patterns.
Profiles from across the Johnson site show a dense and variable assortment of anomalies. At a (very roughly estimated) .50-.75 meters depth, just below the near-field region (time-depth 12.5 nS), some irregularities are apparent (Figure 12). Their location at this interface makes interpretation problematic, however. The first hundred centimeters of soil are particularly vulnerable to a wide range of disturbances and modifications resulting from the actions of people, plants, animals, the weather, errosion, and other geological processes.
It is unfortunate that a higher frequency antenna, 900 MHz for example, was not available for GPR testing at this site. Higher frequency testing provides better resolution in the top soil layers. The target dates of the site are relatively recent, from ca. 1820-1890, and so not much soil accumulation is expected on top of the 19th century living surface. It is significant that no historical occupation occurred on this site prior to 1820, however. The numerous features below one meter may of course represent prehistoric occupations. Given their high reflectance values (very dark, possibly metallic), I believe it likely at least some of these "anomalies" result from the Johnsons' occupation.
The GPR results will be integrated into the archaeological excavation plan at the site. Test units will be located to intersect the anomalies discussed above, focusing primarily on property margins and foundation areas. Units will be placed so as to intersect the two anomalies noted near the northern boundary of Grid A (along the western boundary and at the northwest corner), the regularly spaced anomalies noted in both grids, at least one possible point-source anomaly in Grid B, and the anomaly that stretches across the southern boundary of Grid B. The anomaly noted by both conductivity and magnetometry in the southwestern corner of Grid A also appears as dark patches in GPR transects; this location will be tested archaeologically. The depth of disturbances seen in GPR profiles from the site falls below the expected depth of cultural deposits. Archaeologists, therefore, will be prepared to excavate deep into at least some test units in order to investigate these results.
Anomalies seen in magnetometry and conductivity results at the Johnson site that can be attributed to EM interference are not apparent on GPR readings. This suggests that in bounded urban areas with numerous potential sources of EM interference, GPR may be the best method of geophysical prospection. High frequency radar antennas are also more maneuverable than boom-type receivers in small, fenced areas. Some anomalies that were detected by magnetometry and conductivity, particularly along the west side of grid A and the north side of grid B, do coincide with those detected by radar. These features likely have electrical and magnetic properties distinguishable by the conductivity and magnetometry meters that are not obvious on GPR profiles. It is therefore advisable to combine methods of geophysical prospection in order to obtain information about feature composition as well as location.
Conyers, Lawrence B. and Dean Goodman