GPR surveys were done over an artificially constructed site with known features to test how the physical properties of the buried materials change with regard to their radar reflectivity in differing conditions. The map of known features at this site is seen below:

Results:

400 MHz:

Dry conditions: Surveys generally had poor resolution. Dessication cracks in the soil caused many point source hyperbolas, thus skewing the "appearance" of other reflections.

Wet (saturated) conditions: Radar only reflected the clay floor, while the other three floors were "invisible".

This suggested that water, and how it is distributed and retained, is the primary factor in whether anything will generate radar reflection. In dry conditions, the compacted floors likely had enough water to distinguish them from the overlying material. In wet conditions, the compacted floors and overlying material retained the same amount of water. This lack of distinction in material type caused the floors to be "invisible" on radar. The clay floor, however, pooled water and showed up quite well.

Frozen conditions: Showed similar results as dry conditions, with somewhat better resolution. All four floors are visible, but in general the feature resolution is poor. This slice map was constructed from 400 MHz profiles.

Data collected in frozen conditions with the 900 MHz antenna, however, showed a much better resolution of the floor features. Along with these floor features, many unknown features were also reflected. It is predicted that radar is reflecting the changes in water content of the soils overlying the floors.

Spatial Statistical Analysis

Spatial statistical analysis was performed on maps to determine the spatial correlation between models and actual maps. A correlation was calculated between what the model showed was there and GPR did not pick up, and between what GPR picked up that was not actually there.

In the test with the 400 MHz antenna during dry conditions the R2 value of .56 shows a good correlation, but not great. In general, the outline of the floors is clearly visible and some of the more distinct floor features. Therefore, if a similar feature was present in an archaeological site with the same ground characteristics, it would be easily mapped with GPR when the ground was dry.

Wet conditions using the 400 MHz antenna had an R2 value of .3705. When the CATS floor features were imaged during wet ground conditions, very different results were obtained. The GPR slice map shows high amplitudes only of the second floor, which is the earthen floor that was burned prior to burial. Some of the floor features produced high amplitude reflections in the other three floors, but the floors themselves were not visible. A low R2 value was obtained for the floors overall (.37). It is likely that in similar conditions at an archaeological site only the burned floors would be visible in GPR maps.

This conclusion has very important implications for the use of GPR in archaeological mapping, and suggests that perhaps multiple surveys should be conducted in areas that are prone to alternating wet and dry conditions. If data are collected when the ground is very wet, the survey should be then performed again in dry conditions, as very different features will potentially be visible. The importance of water, and its retention and placement in the ground, is dramatically illustrated in this example from CATS.

The floor features at CATS were even less visible when the ground was frozen using the 400 MHz antenna, which is very much different than was expected. A maximum correlation of only .27 was obtained for the frozen ground at CATS, which was the poorest of all the GPR data. This suggests that in clay-rich areas frozen ground may disperse radar reflections, perhaps because of the fractures in the clay. If the underlying clay is relatively dry, even if the ground surface is frozen, conditions in the ground are un-knowable, and difficult to model correctly.

The 900 MHz antenna was also used to collect data at CATS during the frozen conditions and produced a better correlation than the 400 MHz in the same conditions. Using the 900 MHz antenna the floors are barely visible in frozen conditions, but the correlation is somewhat better than with the 400 MHz antenna, which is purely a function of better feature definition with high frequency antennas.

Topographically Corrected Data

Another feature at CATS was studied, but not in the same fashion as the house floors discussed above. As part of the same feature burial process a simulated burial mound was constructed by placing two pig carcasses in “tombs” and then piling earth over them. Reflection data were collected over the pig tombs in the same way as the house floors, but for these data a topographic correction was applied to the data in the mound to develop a fast way to correct data for elevation. The usual way that GPR data are corrected for topography is to survey each transect about every meter. Each profile must then be “static corrected” manually, by inputting the data for each profile

A surface map was created using a theodolite and stadia rod. This map was gridded and smoothed so that it represented the mound as it existed in the field (see figure on left). Amplitude slice maps were then generated from the reflection data at every depth level (see figure on right), which proved to be an efficient slicing method. When this was done the “pig crypts” that were placed in the ground were easily visible, as seen in the slice in the figure on the right, which crosses the carcasses.