3.1. ALOS PALSAR Polarimetric Analysis
Following the workflow described in
Section 2.2, via comparison of the information, it was possible to derive an initial hint of the different types of scattering mechanisms present over the archaeological structures. The Pauli RGB image presented the sum of all the backscattering contributions coming from the illuminated targets. It provided a first knowledge about the three principal scattering mechanisms of single bounce, double bounce and volume scattering. By overlapping SAR Pauli RGB decompositions with optical imagery (
Figure 2), the pieces of archaeological evidence was located and, together with them, some well-defined scattering mechanisms were identified.
By analyzing SAR Pauli image and Kompsat-2 data, it was noticed that the major contribution was given by the larger archaeological structures, as well as by urban and agricultural areas. Unfortunately, due to the insufficient spatial resolution of PALSAR data, it was only possible to identify the major structures, and for this reason, the ALOS PALSAR investigation has been focused in the field of pyramids in the western area of the site.
In
Figure 12, we can easily distinguish the two groups of royal pyramids (yellow squares), and in the image acquired in 2009, the modern infrastructure close to the site (white arrow, right image). However, a third backscattering ‘anomaly’ has been identified close to the NW group of pyramids (
Figure 12, red ellipse). The qualitative analysis performed by observing satellite optical data (KOMPSAT-2 and Google Earth acquisitions) revealed that there was no surface structure in the area where this backscatter anomaly was detected. Moreover, the available cartographic materials, dating back to 1995, did not seem to record archaeological evidence at that point.
In order to investigate the possible influence of sand humidity on radar wave penetration and possible absorption/loss of the signal we consulted archives of Weather Online [
26]. In both dates of ALOS acquisitions and in the period directly preceding the acquisitions there was no precipitation event, thus the sand was dry and we can conclude that the humidity did not have direct influence on the radar response and couldn’t cause this type of anomaly.
Once this first qualitative analysis was completed, additional polarimetric descriptors were analyzed in order to understand and validate the nature of the scattering mechanisms upon the already known archeological structures, the general morphology surrounding them, and upon the area identified in the qualitative analysis.
Among the several polarimetric decompositions analyzed, the Yamaguchi 4-component decomposition [
24] turned out to be the most meaningful one for the purpose of the present research. Yamaguchi decomposition starts from the assumption that in nature reflection with symmetric conditions is not very common, as previously stated by Freeman decomposition [
22]. Due to the presence of urban and cultivated areas where symmetry conditions are not valid, Yamaguchi 4-component decomposition was performed with a sliding window of 3 × 3 (
Figure 13). This decomposition, shown in
Figure 13, produced an RGB image where the red channel corresponds to the double bounce scattering, the green channel corresponds to volume scattering and the blue channel corresponds to single bounce scattering. This allowed an interpretation of the physics behind the color representation that can be linked to the most common classes of backscattering (urban areas, vegetated areas, and surface).
By observing the RGB image of the three Yamaguchi decomposition descriptors, more detailed scattering mechanism discrimination was noticed in Yamaguchi_G4U1 (
Figure 13, bottom). In this decomposition, it can be noticed that a part of the city of Karima presented a double bounce mechanism (yellow arrow), typical for buildings, while in Yamaguchi_Y4O (
Figure 13, top) the backscattering coming from the city was volume scattering, typical for vegetation. This effect was present mainly due to the orientation of buildings. In fact, in this kind of decomposition, vegetated and urban areas were still presenting a similar contribution. What we noticed is that the cultivated area in the southern part of the Jebel presented volume scattering (
Figure 13, orange arrow) as well as the major part of the buildings in Karima in the NE part of the image. In addition, the two groups of royal pyramids (
Figure 13, red arrow), as well as the Jebel itself, were represented as a combination of single bounce and volume scattering contribution. Following the same consideration, in the area close to the NW group of Pyramids the strong backscattering visible in this area is represented by double bounce mechanism in Yamaguchi_ Y4R and Yamaguchi_G4U1 (
Figure 13, middle, bottom), generally more representative for urban areas, while it appears as a combination of single bounce and volume scattering in Yamaguchi_Y4O (
Figure 13, top). White arrows in
Figure 13 refer to the rocky topography of the area, which presented a similar behavior in both ALOS acquisitions and in the three Yamaguchi RGB images.
In order to deepen the analysis of Yamaguchi_G4U1 decomposition, in which the archaeological and topographic features seemed to be better discriminated, the single channels relating to different scattering mechanisms were investigated. The recognition of the strong backscatter in the single channels was not as straightforward as in the qualitative study carried out on the RGB images. For this reason, we used the latitude and longitude coordinates of the area of interest (18°32′15″N 31°49′14″E WGS84) in order to locate and verify as precisely as possible the correspondence of the well-defined scattering mechanism between SAR data products.
In
Figure 14, Yamaguchi decomposition channels corresponding to the double bounce (top), single bounce (middle), and volume scattering mechanisms (bottom) for the two PALSAR acquisitions are shown. By identifying the latitude/longitude position of the backscattering noticed in the RGB images illustrated above, it was possible to note the amplitude values recorded in that point for each channel of the same polarimetric decomposition in the two PALSAR acquisitions (
Table 3). As a result, the strong backscattering was identified as having high contribution of single bounce and a low contribution of double bounce which was recorded only for this target, while it was not detected in volume scattering channel (
Figure 14, blue cross). Comparing this response to both archaeological and urban features, what arises from a wider analysis is that the urban area of Karima (
Figure 14, red arrows), as well as the palm plantations following the Nile river in the southern part of the site (green arrows), present different major contributions than the one observed for the target analyzed, that is, a volume scattering mechanism.
This kind of response is typical for vegetated areas, however it may also appear in the cities due to the specific orientation of buildings. In the longitudinal SW/NE portion of the city, in the northern part of the images the main backscattering mechanism detected was double and single bounce, as could be expected for buildings and their roofs (red arrows). The two groups of royal pyramids, due to their orientation with respect to the incidence angle of the electromagnetic wave, exhibit some contribution of the volume scattering, but with the major contribution of the single bounce.
As mentioned before, the scattering mechanism detected close to the NW group of pyramids was characterized by the major single bounce scattering with a lower contribution of the double bounce. This target presents, however, the same principal scattering mechanism as the one coming from surface archaeological structures and from the light surrounding morphology, but is distinguished by a double bounce backscatter contribution that was not recorded for either existing archaeological features or for topographic elements.
As we know, low frequency L-band wavelength present a deep penetration capability in very dry environments, as in the case of Gebel Barkal. Therefore, all abovementioned considerations led to the assumption that the detected target might be placed under the surface, thus not visible in KOMPSAT-2 and Google Earth acquisitions, as well as in the available cartographic materials. In order to verify or confute this hypothesis, four RADARSAT-2 imagery acquired with a similar incidence angle of ALOS PALSAR data were further analyzed. The results from this analysis are presented in the following section.
3.2. RADARSAT-2 Polarimetric Analysis
Similarly, to what was done for ALOS data (
Section 3.1) we wanted to obtain an overall visualization of the different types of scattering mechanisms appearing over the archaeological structures, the surrounding morphology, and the vegetated and urban areas in RADARSAT data. For this purpose the RADARSAT-2 polarimetric Pauli RGB decomposition was overlaid with one of the available Google Earth acquisitions (November, 2012), for a first crossed qualitative analysis (
Figure 15).
By observing the four C-band Pauli RGB decomposition images, each of them overlaid with the Google Earth acquisition, we noticed that the archaeological structures and the morphology of the site were more easily recognizable thanks to the higher spatial resolution of RADARSAT-2.
However, the higher spatial resolution posed a problem of less clear backscattering discrimination compared to what was observed in ALOS PALSAR data, highlighting a more detailed distribution of backscattering in the urban area and in the surface morphology to which C-band is more sensitive. Nevertheless, well-localized backscattering (
Figure 15, yellow ellipse) was detected in the area close to the NW group of royal pyramids (
Figure 15, white squares), confirming, at a first level analysis, the persistence of an important scattering mechanisms in the same area identified in ALOS PALSAR data (
Figure 12, red ellipse).
Considering the absence of surface archaeological structures recorded in the cartographic sources and in optical data, also in this case we looked at the meteorological information regarding precipitation. In fact, a C-band target detection in that point could be easily affected by even light precipitations or by a contrast of humidity in the soil. Thanks to the archives of WeatherOnline, as previously, the absence of precipitations and a very low percentage of humidity was noted in the days of the acquisitions and days directly preceding them. This again confirmed dryness of the sand in the area [
26].
In order to observe the backscattering behavior of targets analyzed, and to compare it with the backscattering noticed in the ALOS PALSAR data, the Yamaguchi G4U1 decomposition was performed.
This polarimetric descriptor showed, also in RADARSAT-2 case, a significant amount of information for detection of archaeological structures (
Figure 16).
The overall observation of the Yamaguchi decomposition RGB images lead to a similar conclusion derived for ALOS PALSAR data (
Figure 13). In fact, both the palm cultivations in the southern part of the site and the urban area of Karima presented strong volume scattering, due to the orientation of buildings in the central part of the city. Observations made for modern buildings of Karima were not valid concerning the royal pyramids’ backscattering (
Figure 16, white squares), for which a contribution of all the scattering mechanisms was observed, with a higher percentage of single bounce mechanism.
Indeed, being C-band more sensitive to the surface characteristics, the sum of all the contributions can be caused not only by varying walls’ inclination presented by the two royal cemeteries, but also by the reciprocal orientation of pyramids. The orientation varies for each pyramid, resulting in differently oriented walls with respect to the incident wave. In a deeper analysis, it was possible to notice how the recorded single bounce backscattering from pyramids was related to the inclination of the pyramids’ walls, varying from 68° to 60° (
Figure 17). Moreover, the C-band sensitivity to surface characteristics generated also single bounce scattering mechanism with some double bounce contributions coming from the morphology of the site.
Analyzing the strong backscatter individuated in Pauli RGB and in Yamaguchi G4U1 decomposition RGB images, the nature of the scattering contributions was investigated in each channel of Yamaguchi G4U1 decomposition (
Figure 18).
In particular, the single channels of the decomposition were analyzed for each RADARSAT-2 acquisition to understand if the typology of the noticed backscattering, apparently recorded in Yamaguchi G4U1 RGB image as a double bounce, could also be due to other scattering contributions. The exact location of the backscattering was indicated, once again, by latitude and longitude coordinates (
Figure 18, blue cross). In addition, the backscattering coming from the NW group of pyramids (
Figure 18, red arrow) and the central royal cemetery (
Figure 18, yellow arrow) was analyzed, as well as the scattering contribution coming from the surrounding morphology.
Once the strong backscattering was localized in all the four RADARSAT-2 acquisitions, the corresponding amplitude values of each Yamaguchi_G4U1 decomposition channel were examined. As reported in
Table 4, concerning the strong backscattering close to the NW group of pyramids (
Figure 18, red arrow), a high value of single bounce scattering has been registered, with a low contribution of double bounce and a lower contribution of volume scattering.
By comparing these values to the ones recorded in ALOS PALSAR data, a strong response in the single bounce mechanism was identified in both SAR datasets, as well as a low contribution of double bounce, which however, seemed not to be recorded for the morphology of the site in each RADARSAT-2 acquisitions.
In fact, being C-band more sensitive to the surface topography, several surface scatterers have been detected also in the surrounding portion of the site, in which the L-band seemed not to record any backscatter. By comparing the recorded backscattering close to the NW group of pyramids to that of the surface archaeological structures, we noticed that amplitude values related to the pyramids present the same percentage of scattering contributions, with an exception for volume scattering contribution that was higher for the central group of pyramids (yellow arrow) due to their orientation.
Considering the acquisition time of multi-frequency dataset composed by ALOS PALSAR and RADARSAT-2 data, the persistence of the target was noted over seven years (2006–2013) in the same area.
In the cartographic documentation derived from UNESCO reports on the analyzed area, no surface archaeological evidence was registered in the point corresponding to the strong backscattering noticed in ALOS PALSAR data (
Figure 13). It has to be remembered that the morphology of the site is composed of sand and sand-stone rocks, which show backscattering similar to the one noticed close to NW group of royal pyramids. This was particularly evident in the qualitative analysis of RADARSAT-2 data (
Figure 15).
Nevertheless, supposing this backscattering was due to the sandstone topography of the site, we could expect to have the same responses for all the morphological evidences in the area. This could be true when observing RADARSAT-2 imagery, in which higher spatial resolution allowed discriminating several strong backscattering signals related to the morphology of the site, as well as the scattering mechanism close to the pyramids (
Figure 15). However, this was not confirmed in ALOS PALSAR data, in which we did not noticed specific backscatter related to the morphological surface evidence (
Figure 13).
Moreover, it is important to consider, from a technical point of view, the typology of polarimetric acquisitions selected for this analysis: different frequency and similar incidence angle configurations. In fact, being respectively a 26.20°/23.10° (ALOS PALSAR) and 27.06° (RADARSAT-2) incidence angle, we can assume that ALOS PALSAR L-band wave deeply penetrated sand detecting a target in the ground, as reported in both acquisitions, while RADARSAT-2 C-band lightly penetrated sand, demonstrating a higher sensitivity to the general morphology of the site.
Considering the absence of meteorological events that could affect wave interaction with targets on each acquisition date, it is important to point out that using the higher frequency RADARSAT-2 C-band it was still possible to discriminate this point target, despite the lower penetration capability of the C-band and thanks to the narrow observation incidence angle.