TerraSAR-X Spotlight Interferometric Observations of

TerraSAR-X Spotlight Interferometric Observations of
Archaeoastronomical Structures at Nabta Playa, Egypt
Paul A. Rosen(1), Thomas Brophy(2), Masanobu Shimada(3)
(1)
Jet Propulsion Laboratory,California Institute of Technology,4800 Oak Grove Drive,Pasadena, CA
91109, email:Paul.A.Rosen@jpl.nasa.gov
(2)
EMCS Consulting, 1161 Pacifica Place, Encinitas, CA, 92024, email : tgbrophy@gmail.com
(3)
Japan Aerospace Exploration Agency, Earth Observation research Center, Sengen 2-1-1 Tsukuba,
Ibaraki Japan, 350-8505, email: shimada.masanobu@jaxa.jp
Abstract
In this paper, we describe the analysis of the TerraSAR-X high-resolution spotlight mode radar data in
comparison with in situ measurements, optical imagery and ALOS PALSAR L-band radar data, for the
purpose of investigating surface and subsurface features associated with ancient stones at Nabta Playa,
Egypt. Many of the structures observable in the field and from high-resolution satellite optical imagery are
clearly visible in the TerraSAR-X data. The L-band data is coarser resolution, but shows distinctly
different features from the X-band data, possibly indicating greater surface penetration. The availability of
high-resolution TerraSAR-X data has been essential to the understanding of the area.
Introduction
Nabta Playa in Egypt has been a site of archaeological interest since the discovery of ceremonial artifacts,
grave sites, calendar circles and astronomically aligned megaliths throughout the region. At the time of
earliest settlement during the early Holocene, the area was less dry, so settlements could develop,
apparently leading to the growth of Nabta Playa into a regional ceremonial center [1]. Many of the
archaeological structures are highly fashioned megaliths in a variety of arrangements including mound
complexes, carvings in bedrock buried several meters below the desert surface, and linear arrangements of
large megaliths that appear to be aligned with stars. Two groups have measured the locations of the
megaliths and structures using GPS [1,2] as depicted in Fig. 1. Brophy and Rosen [2] also measured the
positions using high resolution optical imagery, finding good agreement.
Figure 1. Left – Map of Egypt with red circle showing location of Nabta Playa. Right – Quickbird image of excavated
complex structures (upper left), unexcavated complex structures (center), and apparently stellar-aligned megaliths
(lower right).
One of the most intriguing aspects of the area is the existence of complex structures both on the surface and
embedded or buried relatively deeply within the surface (Fig. 2). Nabta Playa offers an “unprecedented
longitudinal view” of human cultural development from early Holocene to late Neolithic (circa 11,000 to
6,000 BP)[3], and contains the earliest known archaeoastronomical megaliths. Imaging the area with a
surface penetrating system with fine resolution could detect heretofore unknown structures [2]. As Nabta
Playa is very dry, the techniques of using long wavelength synthetic aperture radar demonstrated with the
Shuttle Imaging Radar A in the early 1980s would appear to be promising for this purpose.
Figure 2. Left – Excavation of sculpted megalith from central complex structure [4]. Right – Exposed bedrock
sculpture from bottom of central complex structure [3].
L-band or longer wavelengths are of strong interest for this site because of the known ability for long
wavelength signals to penetrate the dry sands of the desert. However, the Nabta site is quite remote and
limited ground or satellite observations of the site have been made. We have a single fine-beam ALOS
PALSAR image at roughly 10 m resolution that we can employ to search for possible subsurface features,
several lower resolution single-pol and fully polarimetric PALSAR images of the area, and a 0.6 m
resolution Quickbird optical image that has been corrected planimetrically to match ground GPS
measurements [2]. For the study of anthropogenic sites such as Nabta, the finest possible resolution is
desired, and having radar observations that match the fine resolution of the Quickbird imagery is
potentially beneficial. TerraSAR-X provides an excellent resource because of its observational flexibility
and exceptionally good resolution and radiometric accuracy. Therefore, our approach has been to acquire
and process interferometric high-resolution spotlight mode TerraSAR-X data over the Nabta Playa site to
survey the surface characteristics, and look for interesting natural and human-induced phenomena.
TerraSAR-X and ALOS PALSAR Observations
To date, we have acquired and processed six interferometric pairs surrounding the Nabta site – three at a
30-degrees look angle and three at 50 degrees. Because of concerns of possible targeting uncertainty in the
employment of spotlight mode acquisition, images were acquired in three overlapping areas as shown in
Fig. 3. The acquisition mode was finest resolution, 300 MHz range bandwidth, yielding imagery roughly 1
m x 1 m planimetrically, depending on incidence angle.
Figure 3. Illustrating the targeting strategy for TerraSAR-X spotlight mode imaging. The center coordinate of the
center frame has the coordinates of Nabta Playa (22.51oN, 30.73oE). The other frames were chosen to overlap the
center frame by roughly 30%. It turned out that the targeting was perfect, and future acquisitions will be made only
with the center frame.
Because of poor understanding of the acquisition request webtool, the interferometric pairs were ordered
with 33 day separation rather than the intrinsic 11-day repeat cycle separation (Each of the three spotlight
frames takes a separate pass to acquire since they are overlapping. To minimize temporal separation of
interferometric pairs, the center frame should have been acquired as a pair on two successive passes,
followed by a pair for the northern or southern frame, followed by a pair for the remaining frame.
However, we ordered the three frames first, followed by a repeat of those three frames later.)
The data were converted from TerraSAR-X COS format to raw single-look complex format using the gdal
geospatial utility software suite [5]. Gdal also allows for conversion to geotiff format and a number of
other transformations.
Interferometric processing was performed in the natural coordinates of the delivered single-look complex
imagery. Because the data were acquired in high-resolution spotlight mode, the Doppler varies in a near
continuous fashion along-track in the image. As a result, the standard interferometric methodology of
resampling the second image to match the first required a slight modification to accomplish the resampling
with an azimuth interpolation kernel centered in the spectral domain on the Doppler bandpass appropriate
to the along-track position. Once the Doppler variation was properly accounted for, the interferogram
formation was straightforward, with good fringes resulting over much of the imaged area.
Geographic tie points provided with the TerraSAR-X data allowed for convenient transformation to the
UTM frame of the reference Quickbird image, and subsequent mosaicking of the frames, using the gdal
utilities. Fig. 4 presents the mosaicked imagery for the two incidence angles acquired. Note that each of the
three mosaicked frames was acquired 33 days apart, yet the correlation estimate is essentially consistent
from one frame to the next (brightness variations from frame to frame on the right are not due to correlation
changes). This would not be expected if correlation loss were due to temporal effects such as changes in
wind-blown sand. Thus we can draw a preliminary conclusion that correlation variability is primarily
driven by signal-to-noise ratio effects: smooth surfaces at the 3 cm radar wavelength scale and/or
penetration of the radar signal into the surface with no return due to scattering or absorption lead to low
signal backscatter in low correlation areas. Nabta Playa is the low correlation region at center and right in
the mosaic, comprising primarily compressed playa sediments, mixed between exposed ablated and sandcovered surfaces. The sand alternates between fossilized sand dunes and localized large and dynamic sand
dunes.
Figure 4. Mosaicked image backscatter for the 30o incidence angle observation (left). Correlation mosaicked images
(with radar backscatter brightness modulation) for the 30o incidence angle (center) and 50o incidence angle (right)
observations. Area depicted at left and center is roughly 10 km across and 20 km vertical. Note generally lower
brightness and correlation at larger incidence angles, even in stable, rocky areas, indicating an intrinsic signal-to-noise
ratio limitation at large incidence angles.
A number of ALOS PALSAR L-band acquisitions have been examined and compared to the TerraSAR-X
images. Fig. 5 shows some of these data, including a full frame of single-pol (HH) fine-beam data
(processed at JPL), and partial frames of polarimetric channels (HH, HV, VV) over the Nabta site.
Figure 5. ALOS PALSAR data acquired over Nabta Playa. The top panel depicts the fine-beam (FBS) HH polarization
data acquired in Feb 2007, acquired at 34o incidence angle. The rectangle in the top panel indicates the portion of the
polarimetric data shown in the lower three panels. The lower left panel is HH, center HV, and right VV polarimetric
data, acquired at 22o incidence angle. The rectangle in the lower left panel indicates roughly the area covered by the
TerraSAR-X mosaic.
Fig. 5 demonstrates that this region of the southern Egyptian desert is in general quite radar dark at 34
degrees incidence angles, indicating relatively smooth surfaces and/or penetration. The polarimetric data
acquired at steeper incidence angle have brighter return, but can still be quite dark. The cross pol channel
has a very low SNR.
Interpretation
The purpose of this study is to look for evidence of man-made large-scale artifacts from ancient times.
These coarse resolution L-band data (10-20 m) would only be useful for detecting large features. We can
however use the TerraSAR-X and optical data to compare fine-scale features at the meter level to these
coarser data. Fig. 6 overlays the optical data on top of the TerraSAR-X data, further on top of the ALOS
HH fine-beam data. The transparency of the optical data is set so that the TerraSAR-X data can be seen
also. Only the edges of the PALSAR data can be seen where the data are exposed due to shifting of one set
relative to another. Note that each data set was geocoded independently, and it was necessary to shift the
data by several meters relative to each other to align them (source of uncertainty unknown at this time).
Figure 6. Overlay of Optical, TerraSAR-X, and ALOS-PALSAR HH data, showing large dunes arranged vertically left
of center and other surface features. Transparency of optical data allows some of the dark regions in the TerraSAR-X
data show through. In particular the dark bands south of each large dunes are the dune itself, which has migrated south
by some 40 meters in the six years between times when the optical and radar data were acquired.
Note that in Fig. 6, there are bands at the south of the major dunes that appear black. These black regions
are the signature of the dune itself in the X-band data, where very little signal is reflected back to the radar.
Thus, this is an indication of dune migration over the time interval between image acquisitions. This is
shown more explicitly in Fig. 7, which shows the dune edge in the optical and in the X-band data close up.
Figure 7. Explicit measurement using hand-held GPS at the site (red mark) confirms that the dune has migrated
southward. Quickbird image of center of Nabta Playa, December 31, 2002 (left). The red mark shows the handheld
GPS coordinate of the edge of the dune, taken on location April 17, 2008. The edge of the dune moved 45 meters
south, as is verified in X-band satellite image, taken the next day, April 18, 2008 (right).
Fig. 8 highlights many of the surface features that can be seen on the ground, including megaliths and
complex structures. The buried objects at one of the complex structures were removed to a museum, but
there is a likelihood that other unexcavated buried structures exist. Many of the features observable in the
optical satellite image are discernable in the radar data as well.
Figure 8. Similar to Fig. 6, but optical and X-band data shown side by side to highlight common features.
Features seen in the optical image such as megaliths and complex structures are readily seen in the X-band images.
Sandy areas appear bright in the optical and dim in the radar. Exposed surfaces appear dark in the optical and bright in
the radar image. Note the prevalence of fresh tracks leading to and from the campsite situated at the south side of the
north-most dune (labeled “dune-edge closeup”).
Fig. 9 is the L-band version of Fig. 8, comparing HH polarization data acquired in 2007 from two separate
modes, the fine-beam HH-pol mode (FBS) and the lower resolution full-polarization mode (PLR). The
large dunes seen clearly in Fig. 8 are barely discernable in the left panel of Fig. 9, but are clearer at right,
presumably due to increased SNR at the steeper incidence angle. An intriguing feature just south of the
north-most dune in the left panel of Fig. 9 is the very bright set of pixels. These pixels are not bright in the
right panel taken two months later, nor are they particularly bright in the X-band data shown in Fig. 8. It is
possible that these pixels are bright due to backscatter from vehicles used by researchers who visit the site
every year. The image was acquired in February 2007, which is a typical time for an expedition. This
particular location is typically used by researchers as a base camp, and tracks to and from the location are
easily seen in the X-Band image of Fig. 8.
Another intriguing feature is visible just left of center in the right panel of Fig. 9. Here, there is a clear
pattern of linear features roughly in the shape of an inverted bell rotated about 10 degrees clockwise. This
feature is not present in the left panel, which is nearly devoid of any backscatter signatures, nor is it present
in the X-band or optical data shown in Fig. 8. To be sure it is a very tenuous feature. We note that it could
possibly be a set of lineaments on or below the surface that may warrant further investigation. This large
possible subsurface feature is situated east of the central “Complex Structure A” of megaliths, from which
the stellar aligned megaliths described by numerous authors radiate [1-4]. The putative subsurface feature
is collocated with a group of surface megaliths located a similar distance from Complex Structure A as
other megalith lines. These surface megaliths were apparently referenced by Malville et al. (Nature
1998)[6] as aligning precisely east “with azimuths of 90.02 degrees”. We identified them in the field and
in the Quickbird image as actually aligning a few degrees north of due east, and labeled them “Newly
Identified Megaliths” [2]. The playa sediments were deposited gradually during the Holocene periods of
known human activity. It would be very interesting if an anthropogenic origin to this possible large
subsurface feature similar to the known surface megaliths could be inferred.
A hint of another intriguing feature is in both panels of Fig. 9, just to the northeast of the central sand dune.
This possible subsurface feature spans only a few large L-band pixels, but it seems to persist in both
images. It is located in an area of playa where field surveys and Quickbird imagery show the surface to be
flat and essentially featureless.
Figure 9. Left panel shows ALOS PALSAR fine-beam mode HH L-band data acquired February 19, 2007 (34o) in
roughly the same aspect and area as Fig. 8. Right panel shows corresponding ALOS PALSAR full-pol mode HH Lband data acquired April 30, 2007 (22o).
Summary and Conclusions
We are still in the early stages of this investigation, acquiring both L-band and X-band data as quickly as
possible. It is clear from the preliminary analysis that higher power systems would be desirable for
imaging areas like this that are SNR limited. PALSAR observations at the finer resolution would be
preferable to existing and planned imaging modes. Because planimetric resolution degrades as the
incidence angle is reduced toward nadir, an angle of no less than about 25 degrees is best.
Interesting changes in the surface are observed in the six years between Quickbird data acquisition and
TerraSAR-X and ALOS imaging in 2007/8. Several large dunes are observed to migrate, and surface
features due to modern people seen to appear and disappear. Some intriguing features appear in the L-band
data, but no clear signal of subsurface structures is apparent in the data. Fine resolution for observation of
human-scale features on the surface is critical, and TerraSAR-X data have been instrumental in
understanding and interpreting the area.
There is much more to be done with this site using existing data, and future satellite acquisitions. A
combination of high-resolution spotlight mode data from TerraSAR-X and Cosmo-Skymed may help
improve SNR, resolution, and feature detectability. Similarly, further time series analysis of polarimetric
L-band data or more fine resolution data would improve feature detectability, should such data be acquired.
Further, we note that in very remote areas of the region there are other sites possibly archaeologically
related to Nabta Playa, that are as yet little studied. For example, in April 2008, one of us visited
“Bagnold’s Circle” named for British officer and physicist Ralph Bagnold who discovered it in 1930[7]
(Fig. 10). Located 450 km west of Nabta Playa and just above the current tropic line, Bagnold’s circle has
some similarities to the “Calendar Circle” at Nabta – it is about twice as big around, made from similar
sized stones, and probably of similar antiquity, and located in a flat shallow basin probably an ancient
playa. The site is so remote and little visited that until the April 2008 expedition, even basic measures of
the circle had not been made. Thus, the Bagnold site may be ideal for future comparative study with Nabta
Playa by surface penetrating systems.
Figure 10. Bagnold’s Circle, April 2008.
References
1. Malville, J., R. Schild, F. Wendorf, and R. Brenmer (2007), Astronomy of Nabta Playa, African Skies,
No. 11.
2. Brophy, T.G., and P. A. Rosen (2005), Satellite Imagery Measures of the Astronomically Aligned
Megaliths at Nabta Playa, Mediterranean Archaeology and Archaeometry, v. 5, No. 1, 15-24.
3. Wendorf, F., and Schild, R. (1998) Nabta Playa and its Role in Northeastern African Prehistory, J.
Anthropological Archaeology, v.17, pp. 97-123.
4. Schild, R., and Wendorf, F. (2004) Mysteries of the South Western Desert, Academia pp.10-15,
No.1(1).
5. http://www.gdal.org/
6. Malville, J.M., Wendorf, F., Mazar, A.A, and Schild, R. (1998) Megaliths and Neolithic Astronomy in
Southern Egypt, Nature pp.488-491, Vol.392, April.
7. Bagnold, R.A. (1931) Journeys in the Libyan Desert 1929 and 1930, The Geographical Journal Vol.
LXXVIII, No.1.
Acknowledgments
This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under
contract with the National Aeronautics and Space Administration. Copyright 2008. All rights reserved.