11th International Conference on Ground Penetrating Radar, June 19-22, 2006, Columbus Ohio, USA

GPR Overlay Analysis for Archaeological Prospection

Dean Goodman1,John Steinberg, Brian Damiata2, Yasushi Nishimura3, Kent Schneider4,Hongo Hiromichi, Noriaki Higashi5

1Geophysical Archaeometry Laboratory, 20014 Gypsy Ln, Woodland Hills, CA 91364, gal_usa_goodman@msn.com

2UCLA Cotsen Institute of Archaeology, A-210 Fowler Museum Los Angeles, CA 90095-1510, jmstein@ucla.edu, bdami-ata@ucla.edu

3Cultural Heritage Protection Cooperation Office, Asia/Pacific Cultural Centre for UNESCO(ACCU), nyasushi@accu.or.jp4USDA Forest Service, Heritage Program, 1720 Peachtree Rd NW, Atlanta, GA  30309, krschne@bellsouth.net5Saitobaru Archaeological Museum, Miyazaki Prefecture, Japan, hongo-hiromichi@pref.miyazaki.lg.jp, higashi-noriak-i@pref.miyazaki.lg.jp

Abstract- A method to synthesize reflections located at vari-ous travel times in the radar record onto a single composite time slice map is developed. Overlay analysis is used to cre -ate comprehensive time slice maps that show the strongest reflectors at a site. The method of synthesizing reflections is a binary process in that it involves overlaying only the rela-tive strongest reflectors at each depth in the radar record. Overlay analysis is shown to be yield much better imaging results than doing simple map additions for several archaeo-logical sites studied.

Keywords – GPR, overlay analysis, archaeology,time slice

I. INTRODUCTION

Often at survey sites, the continuous subsurface structures of interest are not buried at a level depth in the ground. Thin time slices made at these sites may show only parts of these continuous structures on any give single image. To “collect” all these reflections made at different parts of the time slice dataset to try and make a complete image of these structures, a method referred to as “overlay analysis” is introduced. In overlay analysis, the relative strongest reflector at each depth is considered in the composite time slice image to be generated. In overlay analysis, the strongest reflector and the weakest reflectors at each time slice level are assigned colors in a uniquely defined color transform at each level. These are individually set by ei-ther adjusting the points on the data histogram where the colors are to be applied across the data, and/or developing a unique transform curve over which the colors are ap-plied to various levels in the time slice data values.

The process can be easily visualized by examining Figure 1. The reflection components from each time slice map

that are to be included in the composite image are first chosen. The relative strongest reflector at each depth is then progressively displayed on the screen. In Figure 1a, the complete time slice image for the first level to be in-cluded in the overlay is displayed. In the next level to be included in the overlay (Figure 1b), the illuminated areas include all the pixels that essentially have stronger reflec-tions (colors) than the previous map. In the third image (figure 1c), all the areas on this map are higher in reflec-tions than the previous 2 maps. This process is continued down to whatever desired level in the time slice record is desired.

Although examination of the individual maps showing the overlay components from each level separately, does not usually provide a useful display for interpretation of the data, when the images are overlaid on top of each other the final image can be very useful. Shown in Figure 2 are progressive overlay images containing reflections from the surface down to a depth of 68 centimeters. The images were synthesized for the Nanaojo Castle located in Ishikawa Prefecture, Japan. The site is a medieval castle which was once one of several western outposts for the Tokugawa Shogun. The reflections identified on the overlay time slice image indicate several subsurface re-flections that are associated with castle fortifications. In one area near y=25m, 4 strong reflections located along a line are identified which are about 1.8 meters apart. These reflections are believed to be large flat stones that once acted as base supports for building pillars.

The time slice information displayed is made from the square amplitude of background filtered, migrated, and Hilbert transform radargrams. Migration was imple-mented using a nominal microwave velocity of 7cm/ns

11th International Conference on Ground Penetrating Radar, June 19-22, 2006, Columbus Ohio, USA

found from hyperbolic fitting of small points source tar-gets recorded on the site. The thickness of individual slices was made at 4 ns intervals. A total of 87 slices were made across the complete depth range in the radargram, with the time slices having overlap with adjacent levels. Time slices were generated using Krigging or inverse dis-tance interpolation following (Goodman et al, 2004; Conyers and Goodman, 1997; Goodman et al, 1995). The site, because some archaeological features were believed to be very shallow, required making survey lines very dense. The shallower the targets, the finer the spacing must be between survey lines in order not to miss any shallow reflections. For the Nanao Castle, a 0.25m pro-file spacing was used to try and gather a complete reflec-tion record of the shallow structures at the site.

Figure 1. Overlay time slices showing the individual compo-nents of the relative strongest reflectors at each depth. The maps indicate the strongest reflectors which are present at each progressive level. (For example, map 1c, shows only those reflections that are stronger than reflections found on all the shallower maps 1a-1b, etc).

Another application of the overlay analysis was imple-mented for the study of a Viking age site in northwest Ice-land. The site studied was believed to be the location of a fallen Viking age turf house which was once erected from large pieces of drift wood, collected on the coast of an otherwise barren – glacier landscape. The larger pieces would be used as beams for homes; and the material be-tween the beams was turf. The Viking age turf homes can be uniquely identified in age by the use of turf that had a fine white volcanic tephra layer that was from volcanic

eruptions on Iceland dated to 1104 A.D. Outside of the thin tephra layer which is found in the Viking age turf, lit -tle else other than soil color is available to distinguish an-cient turf with modern turf – in essence, the target con-trasts are small.

The site was surveyed with a 400 MHz antenna in both x and y directions with a 05 meter profile separate. Several time slice images were first tried before overlay analysis was implemented. Comprehensive time slice datasets from both combined x and y lines were generated. These images contained some “checkerboarding” noises which were caused by small mosaic noise between data lines recorded on separate days. When there is a small constant offset gain difference between x and y lines, combining binned data used in time slice interpolation can show more fine noises that give the appearance of a “checkerboard” since the interpolation algorithm tries and matches the re-flection at each point used within the search radius (GPR-SLICE v5.0 Software User Manual, 2006). If adjacent or nearby points from x and y surveys have significant gain changes caused by mosaic gain noises, then the resulting time slice images will have the checkerboarding noise. Without pre-amplifying either one of the datasets in the x or y directions, overlay images can be used to synthesize the important reflection contained in both directions with-out incorporating these noises. To do this, time slice im-ages were first created separately for profiles that were parallel to the x axis of the grid, and separate time slices were made for profiles that were parallel to the y axis of the grid. Next, the individual transforms for each dataset are generated and then overlay of the relative-strongest-re-flector for time slice in x and y are generated.

A comprehensive overlay image for the Glaumbaer Viking age site is shown in Figure 3. Some of the reflections identified from partial excavations at the site revealed fallen turf walls. An iron smithy could be inferred from small fragments of slag found near this southernmost re-flection structure.

II. CONCLUSIONSAn important distinction of overlay images implemented in this study are that each individual transform for each time slice used in the overlay weights the importance of reflections at those individual levels. The overlay process provides for user controls on what subsurface levels are important to contain in the overlay images. Comparison of single – thick – time slices computed over the same time window that the overlays are generated have been compared. The results indicate that overlay images pro-vide for better image control and higher contrasts of sub-surface targets recorded. In addition, the overlay image process allows levels in the time slice record where strati-

11th International Conference on Ground Penetrating Radar, June 19-22, 2006, Columbus Ohio, USA

graphic or other geological targets, unrelated to the sub-surface archaeology, can be easily removed from the over-lay. Also, overlay analysis implemented in this study, naturally contains “auto” gaining of the time slice record. The applied (color) transforms at each level are equally weighted when the overlay is created, and only the strong-est value (color) at each location in the time slice depth record is displayed.

ACKNOWLEDGMENTS

Taemoto Izumi and Sumiya Yamamoto from the Bunka Center in Nanao City, Ishikawa prefecture, assisted in the field survey. The surveys in Iceland were funded through research grants from the National Science Foundation.

REFERENCES[1] Conyers, L. B., and D. Goodman, Ground Penetrating

Radar: An Introduction for Archaeologists, Alta Mira Press, Sage Publications, ISBM 0-7619-8927-7, (1997).

[2] Goodman, D., Nishimura, Y, and Piro S., Discovery of a 1st Century AD Roman Amphitheater and Other Structures at the Forum Novum by GPR, JEEG, 9(1), 35-41, (2004).

[3] Goodman, D., Nishimura Y., Rogers, J. D., GPR time slices in Archaeological Prospection, Archaeological Prospection, 2, 85-89, (1995).

Figure 2. Overlay time slices computed for the Nanaojo Castle in Ishikawa Prefecture, Japan. The overlay images of the rela-tive-strongest-reflector at each level in the time slice dataset are progressively added onto the overlay images. The images shown here are only a portion of a much larger continuous overlay dataset computed from the individual time slices. The larger reflector in the southern portion of the area is a filled in well. 4 base stones for ancient pillar supports, and suspected foundations from a building are also identified.

wellfoundations

base stones

11th International Conference on Ground Penetrating Radar, June 19-22, 2006, Columbus Ohio, USA

Figure 3. Overlay image comprised of individual x and y time slice images, computed for the Viking age site in Glaumbaer, Iceland. Photograph at the top shows, an excavation of a turf house wall; a stratigraphic cut shows a white tephra layer that is also found in the excavation. Several other structures, including a iron smithy and a possible shell midden could be identified from the overlay image.