Professional Documents
Culture Documents
Satellite remote sensing is playing an increasingly important role in the detection and documentation of archaeological
sites. Surveying an area from the ground using traditional methods often presents challenges due to the time and costs
involved. In contrast, the multispectral synoptic approach afforded by the satellite sensor makes it possible to cover
much larger areas in greater spectral detail and more cost effectively. This is especially the case for larger scale regional
surveys, which are helping to contribute to a better understanding of ancient Egyptian settlement patterns. This study
presents an overview of satellite remote sensing data products, methodologies, and image processing techniques for
detecting lost or undiscovered archaeological sites with reference to Egypt and the Near East. Key regions of the
electromagnetic spectrum useful for site detection are discussed, including the visible near-infrared (VNIR), shortwave
infrared (SWIR), thermal infrared (TIR), and microwave (radar). The potential of using Google Earth as both a data
provider and a visualization tool is also examined. Finally, a case study is presented for detecting tell sites in Egypt using
Landsat ETM+, ASTER, and Google Earth imagery. The results indicated that principal components analysis (PCA) was
successfully able to detect and differentiate tell sites from modern settlements in Egypts northwestern Nile Delta region.
Keywords: Archaeological site detection, remote sensing, satellite imaging, tell, spectral signature, ASTER, Egypt
1. INTRODUCTION
Remote sensing is used in numerous applications ranging from medical imaging to radio astronomy. It can be broadly
defined as the collection of information about an object, area, or phenomenon, without direct physical contact, and where
electromagnetic energy usually acts as the communications link between the sensor and the study area. As defined in this
paper, remote sensing is discussed from an Earth observation perspective and can include both airborne and spaceborne
multispectral and hyperspectral imaging, Radio Detection And Ranging (RADAR), and Light Detection And Ranging
(LIDAR). Ground based geophysical survey techniques, such as resistivity, magnetometry, and subsurface imaging
radar, although often used in archaeological survey, are normally considered in the realm of geophysics and are not
discussed further here.
There are many advantages to conducting a remote sensing survey when compared to a ground campaign. Primarily,
remotely sensed data can be safely collected from a distance, which otherwise may be impossible due to rugged or
waterlogged terrain, regional political instability, or as a result of a site being located on privately held or military owned
land. A remote sensing survey may also prove more economical and efficient, not normally being affected by the usual
time, personnel, and budget constraints. Furthermore, remote sensing offers a non-invasive technique for the
reconnaissance and recording of cultural heritage that ensures the preservation of sites for future generations.
The aim of this paper is to provide an initial approach to the techniques and methodologies used in remote sensing for
detecting ancient Egyptian archaeological sites and features. It is addressed primarily to Egyptian archaeologists and
Egyptologists, although the information may also prove relevant to Near Eastern specialists working on similar arid
environments. It is assumed the reader has an elementary knowledge of the principles and concepts used in remote
sensing. More specifically, some understanding of electromagnetic energy and its interaction with matter, passive and
active remote sensing systems, sensor resolution, and elementary image processing are essential. Some of these concepts
are discussed in this paper. For a more in-depth treatment the reader is directed to Lillesand et al.1
The exploration and cataloging of ancient Egyptian sites from the ground began with the French invasion of Egypt under
Napoleon Bonaparte in 1798 1801. This initial work was followed by several further exploratory surveys, including a
Prussian-led expedition under the German Egyptologist Karl Richard Lepsius2 in 1842 1845, and British led surveys
under Francis L. Griffith and Flinders Petrie in 1891 1910 and 1883 1927 respectively. These and similar campaigns
set out to establish a comprehensive record of the most visible archaeological sites. However, many sites were
completely missed or badly interpreted due to the limited perspective offered by a ground-based approach.
Prior to the Space Age, archaeological remote sensing in Egypt had its origins in aerial photography using camera film
sensitive to the visible and near-infrared wavelengths3. The value of these and other aerial surveys from around the world
were well recognized, given their advantage of providing unrestricted and unbiased views of archaeological features
from the air. However, limited spectral capability, in addition to poor geographical coverage, were limiting factors
reducing their full effectiveness. The use of high-resolution orbital imaging for site detection became fully realized with
the declassification of American (Corona) and Russian (KVR-1000) spy satellite photography, and with the launch of
commercial satellites beginning with IKONOS in 1999. In 1996, Japanese researchers at the Tokai University Research
and Information Center (TRIC) and the Institute of Egyptology at Waseda University, were the first to detect an ancient
Egyptian site using satellite remote sensing. The discovery was made in the pyramid zone on the Niles left west bank,
in an area where it was believed several pyramids lay undiscovered. The site was named Dahshur North and after
excavation was determined to be a mudbrick tomb of the New Kingdom with a suggested date of the late 18th to early
19th dynasties (ca. 1330 1250 BCE). It is the first New Kingdom monument to be found in this region, where most of
the surrounding archaeology is from the Old and Middle kingdoms4, 5, 6. There are several pyramids in this area that are
still unaccounted for, including those of Neferkare (7th 8th dynasty), and Ity (9th 10th dynasty). In 2008, a group at
Cairo University may have rediscovered the missing pyramid of Menkauhor (5th dynasty), which had been originally
discovered by Lepsius in 1842, but later had been lost under shifting sands7.
Important archaeological finds are continuing to be uncovered using satellite imaging. As recently as 2006, Robert
Schiestl, an archaeologist with the German Institute of Archaeology, Cairo, detected the remains of two previously
unknown 13th dynasty pyramids in the South Saqqara region west of the river Nile8 - identified as sites SAK S 3 and
SAK S 7 (Figure 1). The two other 13th Dynasty pyramids the Khendjer pyramid (SAK S 5) and the Unfinished
pyramid (SAK S 6) were discovered in the early 19th Century (Lepsius) and later excavated by Gustave Jquier9 in
1929 1931. Jquiers excavation ramps can still be seen next to the sites. Schiestls recent discovery demonstrates the
utility of using satellite remote sensing as a tool for site detection, even in areas thought to contain no new finds.
Figure 1. SAK S 3 and SAK S7: recent pyramid discoveries in South Saqqara. QuickBird image courtesy of Google Earth.
2.
Regions of the electromagnetic (EM) spectrum present windows of opportunity. Our eyes are sensitive to a narrow band
of visible light that makes up one small part of a much larger spectrum. The regions of energy beyond the visible, while
detectable only by the multispectral satellite sensor, can provide useful information to an archaeological study. For
different earth surface materials, the amount of emitted and reflected radiation varies as a function of wavelength,
resulting in a unique spectral signature. Spectral signatures for different earth surface materials are well known and have
been compiled into spectral libraries, many of which are available online10. Four regions of the EM spectrum are of
potential interest to satellite remote sensing, and include the visible and near-infrared (VNIR), shortwave infrared
(SWIR), thermal infrared (TIR), and microwave (radar). Each region is sensitive to different physical properties and can
provide different information about the same area of coverage.
The majority of archaeological remote sensing work in the Middle East, including Egypt, has been limited to detecting
sites using the visible range of the EM spectrum, i.e., blue (0.4 0.5 m), green (0.5 0.6 m), and red (0.6 0.7 m),
often neglecting a satellites full multispectral capabilities. This has usually involved simply enlarging high-resolution
images by zooming into areas of interest to find sites and performing simple image processing procedures11,12,13.
However, surveying large areas of ground using the highest resolution sensors can be prohibitively expensive.
Researchers have found that using a combination of lower resolution imagery, such as Landsat ETM+ and ASTER14,15,
in combination with higher resolution products, such as Corona, for confirmation of site locations, can offer a more cost
effective solution. Declassified spy satellite photography from the American Corona16 and Russian KVR missions17,
offer a valuable resource for archaeologists, especially in the Middle East where data availability and imagery costs may
be a problem. Collected in the 1960s and 1970s using black and white film, and delivered as a single panchromatic
channel, Corona can be used in the same way as an aerial photograph and holds enormous potential as a source of
archival material for visualizing archaeological sites and landscapes that have changed over time. Imagery from the
Corona KH-4B series camera (operated between 1967 1972) holds the most potential for work in Egypt due to its high
spatial resolution (~1.8m), wide geographic coverage of Egypt, and low cost. When used in combination with other data
products, it has been instrumental in identifying such features as hollow ways, canals, and archaeological sites in Iraq18,
and the mapping of ancient tell sites in Egypt19.
The NIR region (~ 0.7 1.2 m) provides useful spectral information for areas rich in vegetation cover, such as the Nile
Delta, floodplain, and desert oases, due to its sensitivity to vegetation (biomass) and health (vigor). It has been shown
that for regions covered by more than 75% vegetation the two most promising EM regions for detecting archaeological
features are at the chlorophyll peak of 0.56 m, and the red edge at 0.67 0.72 m20. Healthy vegetation reflects
strongly in the near-infrared when compared to the visible wavelengths, making this region useful for detecting
archaeological features, such as crop or soil marks. Crop marks appear due to the differential growth rates of vegetation
cover and can be indicative of a surface or subsurface archaeological site. The NIR has also been used to increase image
contrast in the pyramid zone to locate new pyramids, and in the ancient city of Tanis for the highlighting of mudbrick
structures21.
While the VNIR bands are sensitive to vegetation cover, the SWIR and TIR channels (~1.6 m 3 m and ~3 20 m
respectively) are more sensitive to soil, mineral, and rock geology18, and can help distinguish materials that may be
spectrally similar in the VNIR channels. Exploiting the channels of higher spectral resolution sensors, such as ASTER,
may help to improve the spectral signature collection for archaeological features.
The TIR is often used in geological remote sensing for the identification and mapping of various rock types and rock
forming minerals and is particularly sensitive to silicate composition22. It can also be a valuable tool for detecting
archaeological sites, where differential heating may be an indicator of archaeological features within a site. This is
because archaeological materials at or below ground level can affect the temperature of their surroundings (i.e., different
materials cool at different rates). Although most TIR data is of coarse spatial resolution, it may be useful for detecting
lineaments such as roadways, canals, and walls, in addition to larger scale features. Most thermal imagers make use of
the 8 14 m range, with a secondary window occurring at 3 5 m (atmospheric windows). Other regions cannot be
used due to atmospheric absorption caused by water and atmospheric gases, such as carbon dioxide. Many remote
sensing systems include at least one thermal band, which are typically displayed in grayscale, where bright areas
represent warm regions, and dark areas represent cooler regions. ASTER is unique in having five thermal infrared
channels - covering 8 to 12 m at a spatial resolution of 90 m (in contrast, Landsat ETM+ has only one thermal band at
60 m). At a resolution of 90 m, ASTER TIR is not ideal for locating buried structures, but its increased spectral
capabilities should prove useful in locating archaeological features on a larger scale.
Unlike optical sensors, which passively receive energy reflected from or emitted by an object, Radio Detection and
Ranging (RADAR) provides its own energy source by actively transmitting pulses of microwave energy that are
reflected off a distant surface or object. Radar has a side looking geometry, illuminating the ground scene at an oblique
angle and enhancing variations in both roughness and texture23. Radar offers several advantages to optical remote
sensing in that it can penetrate through both cloud and rain, in addition to operating during both day and night,
independent of illumination from the sun. A major benefit of radar at longer wavelengths is its ability in arid
environments to penetrate up to 2 meters of sand in optimal conditions, i.e., extremely dry, fine-grained, clay-free sand24.
This can be very useful for detecting subsurface archaeological structures in desert regions and for locating
palaeochannels, also known as radar rivers (see Section 4).
SATELLITE SENSOR
SUBSYSTEM with
number of bands
SPATIAL
RESOLUTON (m)
COST (US $)
Landsat 7 ETM+
VNIR (4)
SWIR (2)
TIR (1)
PAN (1)
30
30
60
15
VNIR (3)
SWIR (6)
TIR (5)
PAN (1)
VNIR (4)
SWIR (1)
Visible (1)
Visible (1)
VNIR (4)
PAN (1)
VNIR (4)
PAN (1)
VNIR (4)
PAN (1)
VNIR (8)
PAN (1)
VNIR-SWIR (220)
Synthetic Aperture
Radar (SAR)
L Band Synthetic
Aperture Radar
15
30
90
2.5 - 5
10
20
1.8
2-3
4
1
1.65
0.5
2.4
0.6
1.84
0.5
30
30
Free
10-100
ASTER
SPOT-5
Corona 4B
KVR-1000
IKONOS
GeoEye-1
QuickBird-2
WorldView-2
Hyperion
Shuttle Imaging Radar
(SIR)
PALSAR
ASTER GDEM
DEM
SRTM
DEM
20 vertical;
30 horizontal
15 vertical;
30 horizontal
Free
Free
Remotely sensed imagery can be purchased and downloaded from the following satellite imagery providers:
1. The Global Land Cover Facility (GLCF) of the University of Maryland: http://glcf.umiacs.umd.edu/.
2. The Land Processes Distributed Archive Center (LP DAAC) of the United States Geological Survey (USGS):
https://lpdaac.usgs.gov/.
3. The Warehouse Inventory Search Tool (WIST) of the USGS: https://lpdaac.usgs.gov/lpdaac/get_data/wist/.
Declassified high-resolution photography can be ordered from:
1.
2.
Corona: http://eros.usgs.gov/#/Find_Data.
KVR-1000: http://www.sovinformsputnik.com/.
The cost of acquiring large amounts of commercial high-resolution satellite imagery for regional survey purposes can be
prohibitive. Although data costs have come down, pricing is still beyond the reach of most archaeological budgets. For
example, the cost of the WorldView-2 Basic Bundle, consisting of the full complement of spectral bands
(1 panchromatic and 8 multispectral) costs over US$8,000 (correct as of July 2010) for imagery that is currently archived
and does not require the re-tasking of the satellite. In the past, this has meant that researchers have had to rely on lower
resolution imagery, i.e., Landsat ETM+ and similar datasets, and order higher resolution data only when essential.
The use of high-resolution imagery delivered through the Google Earth (GE) application (http://earth.google.com/) is
causing a mini-revolution in the way that scientists, professionals, and private individuals are collecting and using
satellite data of the Earth. Imagery from an external Google server is streamed over a broadband connection to a remote
user anywhere in the world. The wide coverage of the Earth with high-resolution satellite imagery via GE can be very
useful for archaeological prospection25. For the first time both professional and amateur archaeologists have free access
to large amounts of high-resolution satellite data. The imagery is provided at various spatial resolutions with the majority
being one to three years old. Currently most land areas of the Earth are made available as base imagery in GE at 15 m,
using pan-sharpened Landsat ETM+ data. The base imagery is gradually being replaced with higher resolution data such
as 2.5 m SPOT and 0.6 m (60 cm) QuickBird-2 (QB-2), in addition to aerial photography and imagery from other
sources. Currently about 70% of Egypt is covered by the QB-2 satellite (http://www.digitalglobe.com/) at a spatial
resolution of 60 cm. Areas that do not yet have QB-2 coverage are covered by the French satellites - Systme Probatoire
d'Observation de la Terre (SPOT, http://www.spotimage.com/) at resolutions ranging from 2.5 m to 10 m.
GE comes with several useful features, including the ability to use historical (multi-temporal) imagery to view regions of
the Earth at earlier dates. This can be useful for several reasons. Firstly, if many archived images of the same area are
available they can give an idea of how the landscape has changed over time. For example, historical imagery can be used
in change detection studies for modeling the effects of urban expansion on a landscape. Secondly, as discussed earlier,
the Delta and floodplain can exhibit unique vegetation patterns, such as crop marks, that may assist in detecting
archaeological remains. Depending on the time of year that the satellite image was acquired, the crop mark may not be
visible. GE allows imagery to be examined at different dates to find a more appropriate image. A second useful feature
of GE is the ability to change the angle of the sun and the surface lighting. The time of day that the scene was acquired
can be an important factor when examining archaeological features. For example, tell sites can often be spotted based on
their characteristic shape and shadowing, which is dependent on the position of the sun. Digital Elevation Models
(DEMs) such as NASAs SRTM or ASTER GDEM, can be added as a separate layer to GE to assist in the
identification and visualization of tells and mound-type burial structures by giving a more life-like three-dimensional
perspective. Previous workers have used DEMs combined with mathematical modeling to assist in the identification of
tell sites in the Near East26, 27.
There are a number of disadvantages to using GE as a primary imagery provider: 1) It provides no multispectral
capability, other than in the visible part of the EM spectrum. The channels in the near, shortwave, or thermal infrared that
are normally included in a commercial release have been truncated. This limits the user to applications in the visible
region only; 2) The visible channels have been merged into a single band and displayed in true color. This forces the
user to examine the imagery under a specific band combination; 3) Imagery from GE (or Google Maps) can only be
saved as a jpeg file and at a limited resolution up to 4,800 pixels for the professional version, compared to 1,000 pixels
for the free version. Once saved, the image needs to be imported (with geo-referencing) into a remote sensing or a GIS
package for further processing; 4) The spatial and spectral resolution of the imagery has been degraded due to the use of
compression algorithms for reducing the file size when streaming the imagery over the internet.
Although GE is very good at examining imagery at very high resolution in the visible bands, it cannot substitute for the
full spectral resolution of imagery offered by commercial providers. Unlocked commercial imagery includes many more
bands that can be easily manipulated with digital image processing operations.
GE is freely available from http://www.google.com/earth/index.html. The product is currently offered in two versions:
GE and GE Pro. GE is free to download, while GE Pro is a commercial product providing extra features, including the
ability to save and print at higher resolutions, import vector files, and calculate 3D measurements. Competing products
offering similar functionality include: NASAs World Wind (http://worldwind.arc.nasa.gov/), USGS EarthExplorer
(http://earthexplorer.usgs.gov/), ArcGIS Explorer (http://www.esri.com/software/arcgis/explorer/), and SkylineGlobe
(http://www.skylinesoft.com/). These products are all proprietary, except for World Wind, which is open-source only
using data that is freely available in the public domain.
features, named radar rivers, are detectable using microwaves (radar), but are invisible from the ground and
undetectable using optical sensors, such as Landsat. The presence of many present day oases, including the Kharga,
Dakhla, Fayyum, and Siwa of the Western Desert can be attributed to these palaeo-drainage networks buried deep under
the sand. Many of these oases served as the crossroads for caravan routes transporting luxury items, including ivory,
spices, in addition to animals and slaves. It is hoped that a future study to map the full extent of radar rivers will lead to
the discovery of further ancient water sources and their associated archaeological sites.
In addition to using microwave remote sensing for locating radar rivers, VNIR imagery has been successfully used to
identify ancient roads and trading routes (hollow ways). First identified by Wilkinson and Tucker in the North Jazira,
Iraq36, hollow ways represent shallow linear depressions in the landscape resulting from a millennia of human and
animal foot traffic. They can be identified through satellite imagery from their soil, moisture and vegetation
characteristics18. Although there have been many studies of hollow ways in the Near East, e.g., Iraq, Iran, Syria18,37,38,
there has been no definitive study of them in Egypt. The lack of moisture in a desert environment may not be conducive
to their preservation. However, the locations of many ancient trading routes in Egypt are well known. For example, the
Darb el-Arbain, part of which is now paved and still in use today, was a trading route dating from the Old Kingdom,
connecting Middle Egypt and northwestern Sudan and passing through the Kharga Oasis. Further studies to detect such
features in Egypt would be informative. However, the dating of desert tracks and hollow ways in aerial photography and
satellite imagery is problematic. This is because an identified lineament may theoretically be of any age. It may be
possible to determine its relative age through the principle of superposition (cross cutting relationships). More precise
dating may be achieved through an examination of any surface ceramic remains and associated epigraphic materials39.
Desert road archaeology has raised awareness of the significance of caravan routes and desert oasis settlements where
evidence for some of the earliest writing in Egypts history has been found. The Theban Desert Road Survey is one such
project examining desert routes, and recording the earliest examples of epigraphic materials40.
Archaeological sites in Egypt are often found buried beneath or situated adjacent to modern developments. This can
make site detection challenging. An important step toward solving part of this problem is to use spectral analysis to
separate ancient and modern building materials. The two principal building materials used in ancient Egypt were sunbaked mudbrick (acquired from the flood plain and used for the construction of houses and town walls), and stone
(predominately used in the construction of temples and tombs). Amongst surviving monuments, sandstone is the most
prevalent, however limestone and granite is also found. Creating databases of the spectral signatures of these materials is
expected to substantially improve detection methods. An even greater problem often encountered is from local farmers
who harvest the mudbrick from tell sites (mudbricks are high in nitrate and phosphates), to be used as a crop fertilizer
(known as sebbakh). Tell Mudbricks are also being recycled for use in road and building construction. If recycled into
modern structures, ancient and modern spectral signatures will be identical and impossible to separate.
The nature of the archaeological landscape needs to be understood before a remote sensing survey can be carried out.
Egypt consists of two very diverse landscapes: the water-rich Nile Valley (Delta and floodplain) and oases; and the dry
sandy barren desert (Western and Eastern Deserts, and the Sinai Peninsula). Both environments require the use of very
different tools and techniques. It is essential to determine the spatial and spectral characteristics of the target. There also
needs to exist sufficient contrast between it and the surrounding matrix for it to be differentiated13. These will have an
impact on the choice of appropriate sensor. For example, radar does not penetrate into the subsurface in moisture-rich
regions due to increased reflectivity, and so it would make a poor choice for subsurface detection in the floodplain and
Delta regions. Finally, a remote sensing survey should never be used in total isolation. It is vital when possible to follow
up with a ground-based confirmation of imagery results through a surface survey, followed by, if necessary, geophysics,
coring, and ideally excavation.
Digital image processing enhances the imagery to make it more interpretable. There are many types of imagery
enhancement techniques, including radiometric (modifying values of individual pixels), spectral (transforming pixel
values on a multiband basis), and spatial (modifying pixel values based on the values of surrounding (neighborhood)
pixels). Examples of these techniques include the contrast stretch, principal components analysis, and convolution
filtering. In this section, I highlight some of the methods for displaying and enhancing imagery that have been
successfully used in previous site detection studies. These include simple band combinations for imagery visualization,
pan-sharpening to increase multispectral resolution, and principal components analysis for reducing spectral redundancy
due to inter-band correlation. For a more in depth analysis of digital image processing the reader is directed to Lillesand
et al.1 and Short41.
Image processing software can only display three bands of data at a time (where a single band is sent to each of the red,
green, and blue color guns of the display). However, multispectral satellite imagery consists of multiple bands of data,
where each band is sensitive to a different region of the EM spectrum (for example Landsat ETM+ has 8 bands, ASTER
has 14, and Hyperion has 220). Choosing to display a different band combination can highlight different attributes of the
landscape, such as vegetation patterns, geological features, and archaeological anomalies. A choice has to be made about
which three bands to use to best display the particular features of interest. Popular band combinations using Landsat
TM/ETM+ imagery include: 3, 2, 1 (R, G, B); 4, 3, 2 (NIR, R, G); and 7, 3, 2 (SWIR, R, G). The band combination of
3, 2, 1 is also known as true color as it displays the scene using the same red, green, and blue wavelengths of light that
our eyes are sensitive to, helping to create a more natural looking image. The combination 4, 3, 2 is often used in
archaeology for displaying crop marks and other surface vegetation patterns. Vegetation appears red in this combination
due to the near-infrared channel being sent to the red color gun. This is because healthy vegetation reflects strongly in
the near-infrared wavelengths. The combination 7, 3, 2 is often used in geological interpretation of rocks and mineral
types, which may also be useful for archaeological interpretation. Non-true color images are often referred to as falsecolor composites (FCC) because the band combinations give an artificial representation of the landscape. Other imagery
products will have similar band combinations depending on their band allocations.
Many remote sensing imagery products include a higher resolution panchromatic (grayscale) band in addition to lower
resolution multispectral (color) bands (e.g., Landsat ETM+, SPOT-5, IKONOS, and QuickBird). Pan-sharpening can be
used to increase the spatial resolution of the multispectral bands by merging their color information with the geometric
information from the panchromatic band, resulting in high-resolution multispectral imagery. For example, the 15 m
panchromatic band from Landsat ETM+ can be merged with the six 30 m multispectral bands, resulting in six
multispectral (color) bands each at 15 m spatial resolution. In addition to combining bands from a single satellite, it is
also possible to combine bands from two different satellites (e.g., the 5 m pan band from SPOT-5 with the 30 m
multispectral bands from Landsat ETM+). There are a number of algorithms available including intensity hue saturation
(IHS), Brovey transform, principal component analysis, and wavelet-based sharpening methods. A quantitative study
comparing different algorithms has shown that the wavelet based method offers better results most of the time42.
A problem often encountered when analyzing satellite imagery is deciding how to extract the most useful information
from a multidimensional dataset. This is because there is a tendency for multispectral data to be highly correlated when
the spectral bands are close together (if two bands are correlated, they share some of the same information). For example,
ETM+ bands 1 and 2 are highly correlated, whereas bands 1 and 4 are less so because they are spectrally further apart.
Principal components analysis (PCA) and similar techniques (such as the decorrelation stretch and canonical analysis)
may be used to reduce the redundancy and dimensionality of a dataset by transforming the bands into a smaller number
of non-correlated components, which when displayed are often more interpretable than the original source data. PCA is a
popular tool for analyzing data consisting of multiple bands (for example, ASTER or Hyperion data). Applying PCA
produces the same number of output components as bands in the original data. The first principal component (PC1)
contains the maximum amount of variation, with each successive component displaying less variation. For example, in
Landsat ETM+ data, usually the first two components together account for over 90% of the total variation in the original
data, while the first three account for almost all of the variation. If the remaining PCs contain no useful information, they
may be safely discarded. However, even higher order PCs can sometimes contain valuable information. Either the
components can be viewed individually in grayscale or a FCC can be generated by sending three components to each of
the red, green, and blue color guns of the display.
Other image processing tools often used in site detection include shape detection algorithms43,44, object-based
classification schemes, and various spatial filters (for example the Laplacian filter).
Figure 2. The Nile Delta with boxed study area. Tell sites are shown in yellow (EES) and red (UU). Imagery Google Earth.
A single Landsat 7 ETM+ scene (Path 177, Row 38), acquired on January 1, 2003, was downloaded from the Land
Processes Distributed Active Archive Center (LP DAAC, https://lpdaac.usgs.gov/), using the NASA Warehouse
Inventory Search Tool (WIST). December through February are the wettest months of the year in Egypt and provide
optimal conditions for tell site detection. An increased moisture content in the mudbrick has been shown to be a key
factor in locating archaeological sites19. Therefore, a January acquisition date was considered ideal. Although the chosen
scene had a relatively high cloud content (about 9%), the clouds were isolated to regions over open ocean to the north of
the study area, and did not impede the image processing. Landsat ETM+ imagery includes eight separate bands,
including four in the VNIR (bands 1 4), two in the SWIR (bands 5 and 7), one in the TIR (band 6), and one highresolution panchromatic band (band 8). All bands except for the thermal (which was discarded) were imported from
GeoTIFF (.tif) format into ERDAS IMAGINEs native file format (.img). The six bands (1, 2, 3, 4, 5, and 7) with a
spatial resolution of 30 m were layer stacked as a single multilayer .img file and each channel visually checked for
sensor errors prior to further processing.
The raw uncalibrated digital numbers (DNs) were converted to exoatmospheric reflectance (a dimensionless value),
representing the ratio of upwelling to downwelling radiation, measured on a scale between 0 and 1 (or 0 100 %).
Converting to reflectance has many advantages, including the ability to compare spectral information from one satellite
scene to another. Although not strictly necessary when working with a single scene, this pre-processing step is required
when quantitatively comparing values between scenes of a multi-scene, multi-temporal dataset. Converting DNs to
reflectance is a two-step calculation. The DNs are first converted to radiance at sensor values (spectral radiance), with
units of W m-2 sr-1 m-1. This requires the gain and bias parameters of the sensor, which can be found in the header of the
downloaded metadata file. The second step involves converting spectral radiance to exoatmospheric (apparent)
reflectance, which requires scene-specific information, such as the solar zenith angle, and the mean Earth-Sun distance in
astronomical units. Further information, including the conversion equations for the above calculations can be found in
the Landsat 7 Science Data Users Handbook49.
The multispectral imagery was pan-sharpened with the high-resolution panchromatic band to 15 m using a principal
component algorithm with nearest neighbor resampling. Improving the spatial resolution of a dataset may not improve
the ability to classify finer levels of spectral detail, but it does allow for the mapping of smaller features, thus helping to
address the mixed-pixel problem. The results of the pan-sharpening are shown in Figure 3, where the city of Fuwa
(311220 N, 303324 E) is used as an example (displayed in a band combination of 4, 3, 2). The image is visibly
sharper, containing more definition in both the urban and vegetated areas.
Figure 3. Landsat ETM+ FCC (4, 3, 2) at 30 m prior to multi-resolution merge (left); same scene post 15 m PC merge (right).
Color composites were constructed in an effort to determine the combination of bands that would best enhance the tell
sites in the imagery. In addition to displaying the usual band combinations, i.e., 3, 2, 1 (true color) and 4, 3, 2 (false color
emphasizing the NIR in red), the Optimum Index Factor (OIF) was also calculated. The OIF determines the band
combination that displays the maximum variability in a scene, and is ranked from lowest to highest, with the highest
score indicating the combination that displays the highest variability. Landsat ETM+ has a total of 20 three-band
combinations (there are 20 ways of combining a total of 6 bands in groups of 3). To calculate the OIF, the correlation
matrix and the standard deviations for each channel were entered into an Excel worksheet containing the OIF equation,
to produce a ranked list of OIF combinations. The results indicated that band combination 3, 4, 7 ranked highest for the
pan-sharpened imagery, while 1, 4, 7, was the highest for the unsharpened imagery. It should be noted that both
combinations include spectral information from all three regions of the visible, NIR, and the SWIR. It has been
suggested that using channels from each of these three regions should result in a more interpretable image50. The OIF
only indicates the choice of bands for display, leaving the analyst to decide the color to which each band can be assigned.
It was decided to use both the 3, 4, 7 (R, NIR, SWIR) band combination suggested by the OIF calculation, in addition to
a combination of 4, 3, 2 (NIR, R, G) for loading up scenes for display and classification purposes. Further details of the
calculation and specifics of the OIF equation can be found in Chavez et al.51,52
To successfully detect tell sites, the spectral characteristics of tell earth needs to be identified and differentiated from
other landscape materials. Difficulties in detection are compounded by several factors. Modern developments often arose
adjacent to tell sites, or were even built over the top of them. In addition, tell mudbricks were often recycled to be used
as modern building materials. In an effort to gain a better understanding of the spectral characteristics of tell sites, ten
spectral profiles were collected for each of four different ground cover types: water, modern urban centers, agricultural
crops, and tell sites (Figure 4). Healthy vegetated areas (red) show the typical profile of low reflectance in the visible
(bands 1-3) and SWIR (bands 5 and 6), and high reflectance in the NIR (band 4). Water (blue) exhibits high absorption
throughout the spectrum as would be expected, producing minimal reflectance values. Modern towns and cities (black)
and tell earth (yellow) share similar spectral characteristics, with a modest decrease in the visible, an increase in the NIR,
and a decrease towards the end of the SWIR.
Figure 4. Spectral profiles for several landscape features. Tell sites are displayed
in yellow and have consistently higher spectral values in the NIR and SWIR.
PCA was run on the 30 m Landsat reflectance data to decorrelate the bands. This was achieved by transforming the six
correlated bands into six uncorrelated grayscale components (PC1 PC6). The first three components, PC1 PC3,
accounted for approximately 99.4% of the total variance of the original dataset (79.5%, 18.8%, and 1.1% respectively).
However, they were unable to separate modern settlements from ancient tell sites. It was the fourth component, PC4
(accounting for only 0.35% of the total variance), that was able to best differentiate tell sites from the modern sites
(shown in Figure 6 after a two standard deviation stretch). PC4 clearly identifies nineteen of the twenty-four previously
known tell sites (circled in green in Figure 6), in addition to eight previously unknown sites (circled in red). The
remaining five undetected sites, were either too small to detect or were buried beneath urban sprawl. The remaining two
components, PC5 and PC6, accounted for the remaining 0.25% of the variance, but did not contain any useful
information. Band numbers 7 and 5 (SWIR) from the original six-band dataset were found to be contributing the most to
PC4 (called the factor loadings). The aforementioned variances and factor loadings were obtained from the eigenvector
matrix and eigenvalues, which were calculated during the PCA operation.
Figure 6. Landsat ETM+ PC4 grayscale image after a two standard deviation stretch showing tell sites as high brightness values.
Sites previously known from an earlier survey are circled in green, while newly discovered sites are circled in red.
To exploit the increased spectral resolution of ASTER in the SWIR, two scenes were downloaded from LP DAAC. The
SWIR was resampled to 15 m to match the resolution of the VNIR dataset before layer stacking (TIR was discarded).
The DNs were converted first to radiance and then to reflectance values prior to mosaicking, using a procedure reported
by Smith53 and Milder54. After PCA was applied to the mosaicked ASTER scenes, it was found that PC5 was the most
appropriate component for emphasizing tell sites (high brightness values), in marked contrast to the low brightness
values of the surrounding landscape. ASTER PC5 is shown in Figure 7 after a two standard deviation stretch.
Figure 7. ASTER PC5 grayscale image after a two standard deviation stretch. Tell sites appear as high brightness values.
The Landsat ETM+ PC4 image was subset to examine an area in closer detail. Limits were set for detecting pixel values
greater than a given threshold, which was performed using the ERDAS IMAGINE model maker tool using an input
raster (the ETM+ PC4 image), a process (function), and an output raster. The function consisted of the formula:
EITHER 0 IF ( $n1_PCAinputimage < -0.020 ) OR 1 OTHERWISE
This function replaces the input raster with a binary image, where values below the threshold are replaced with a 0, and
values above the threshold are replaced with a 1. Thresholding makes it possible to determine which pixels most likely
represent tell earth. A threshold of -0.020 was found optimal after a process of trial and error. The CLUMP and
ELIMINATE GIS functions were used to remove small groups of pixels within the binary output image. The CLUMP
function works by performing a contiguity analysis on the input raster (using 8 connecting neighbors), to identify
clumps, or contiguous groups of pixels, which are then recoded into a single class. Finally, the ELIMINATE function
was applied to the clump thematic, to remove clumps of pixels of a specified size. The resultant image is displayed in
Figure 8, where the output from the threshold model has been overlaid in red on the Landsat scene (displayed in 3, 2, 1).
Modern urban regions appear unchanged (light gray), whereas regions likely to be tell sites are indicated in red.
Figure 8. Landsat ETM+ true color composite (3, 2, 1) with PC4 overlay displayed in red.
Figure 9A shows an enlarged view of the above Landsat image and is provided with several additional annotations.
Circled locations represent regions of brightness in PC4 (Figure 6), and correspond well with the locations of tell sites.
Previously mapped sites from both the EES and UU are circled in green, whereas newly detected sites are circled in red.
In addition to detecting nineteen known tell sites, eight new sites were identified, including one of circular shape with
approximate radius of 110 m, located at 311831 N, 304858 E (Figure 9B). Tell sites adjacent to modern towns or
cities are well differentiated and are easily identifiable using the thresholding approach, such as kom el-Haddadi, located
at 311958 N, 304719 E (Figure 9C), and kom el-Khariba, located at 311942 N, 305055 E (Figure 9D). Tell
sites (both known and unknown) were confirmed using high resolution QuickBird imagery from Google Earth. However,
five previously known sites were not detected using this approach. After examining their locations using Google Earth, it
became apparent that those tells had been leveled and buried beneath modern sites. Too little remained on the surface to
be detected using the spatial resolution of the satellite sensor. This technique created a small number of false positives, as
can be seen in the southeast corner of Figure 9A (indicated by uncircled red areas). Some of these regions may represent
tell sites, however they have not been confirmed with any degree of confidence. Future ground observations of these
regions will be able to clarify their origins. Adjustment of the threshold levels for the detection method will reduce false
positives, however this may also reduce the number of pixels representing real tell sites.
Downloaded from SPIE Digital Library on 11 Nov 2011 to 90.208.175.181. Terms of Use: http://spiedl.org/terms
Figure 9. Landsat ETM+ true color composite (3, 2, 1) subset with PC4 threshold overlay in red. Previously known and newly detected tell sites are
circled in green and red respectively (A); a newly detected tell site (B); the tell site of kom el-Haddadi (C), and tell site kom el-Khariba (D).
7. CONCLUSIONS
Satellite remote sensing is revolutionizing the way new archaeological sites are being detected and recorded. This is
made possible through advances in sensor design with the introduction of instruments offering higher spatial and spectral
resolutions. Remote sensing offers many advantages when compared to more traditional survey approaches, including
providing a safe, cost effective, and synoptic view of the landscape. Data obtained via satellite can be viewed and
analyzed from an image processing lab from anywhere in the world, making possible international and multidisciplinary
collaborations. Most importantly, remote sensing offers a non-intrusive method for archaeological reconnaissance, by
mapping and conserving sites for future generations of archaeologists. The practical component of this paper presented a
strategy for the detection of ancient settlement mounds, or tells, in Egypts northwestern Delta. PCA was applied to
both the VNIR and SWIR channels of Landsat ETM+ and ASTER imagery, and the results indicate that it can be a
useful technique for detecting tell sites in this region of the Nile Delta. It was the higher-order components that contained
the patterns of variability able to separate the spectral characteristics of tell sites. These features were not present in any
other lower-order component or individual ETM+ or ASTER channel. Furthermore, the increased spectral resolution in
the SWIR bands provided by ASTER yielded greater contrast in its principle components, when compared to using
Landsat imagery. This work constitutes an important first step for mapping the spatial distribution of these important
monuments, which in turn will help to provide a window into urban land distribution and settlement patterns at a time of
the first emerging communities. Many of these sites are at risk from urban, industrial, and agricultural expansion, in
addition to looting and vandalism. It is important to document these sites before they completely disappear from the
archaeological record. However, it should be made clear that the results from these investigations have not yet been
independently verified from the ground, and any newly detected sites should remain as anomalies until a ground
campaign can prove otherwise.
ACKNOWLEDGEMENTS
The author wishes to thank Keble College, Oxford, and the Keble Association, for a generous award in support of this
work. This work was also supported by an international society for optics and photonics (SPIE) conference travel award.
Thanks are also due to the Department of Earth Sciences, University of Hong Kong, for providing work facilities during
the writing of this paper. ASTER data was generously supplied by NASA through a program for approved educational
use. ASTER data are distributed by the Land Processes Distributed Active Archive Center (LP DAAC), located at the
U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center (https://lpdaac.usgs.gov/).
Finally, the author also wants to thank John Baines of the Oriental Institute, Oxford, and Gary Lock of the Institute of
Archaeology, Oxford, for their helpful comments and suggestions in the review of this paper.
REFERENCES
[1] Lillesand, T., Kiefer, R. and Chipman, J., [Remote sensing and image interpretation], John Wiley & Sons, Inc.,
Hoboken, NJ (2007).
[2] Lepsius, K., [Denkmler aus Aegypten und Aethiopien], Nicolai, Berlin (1849).
[3] Engelbach, R., The aeroplane and Egyptian archaeology, Antiquity, 3, 470-473 (1929).
[4] Etaya, M., Sakata, T., Yoshimura, S, Kondo, J., and Hasegawa, S., A study for the pyramid exploration utilizing
satellite data analysis and the discovery of ancient Egyptian remains at Dahshur North, Tokai University Research and
Information Center (1997).
[5] Etaya, M., Sudo, N., and Sakata, T., Detection of subsurface ancient Egyptian remains utilizing optical and
microwave satellite data, IEEE T Geosci Remote, 6, 2480-2482 (2000).
[6] Yoshimura, S., Kondo, J., Hasegawa, S., Sakata, T., Etaya, M., Nakagawa, T, and Nishimoto, S., A preliminary
report of the general survey at Dahshur North, Egypt, Annual Report of the Collegium Mediterranistarum
Mediterraneus, XX (1997).
[7] Kratovac, K., Egyptian archaeologists uncover missing pyramid, http://www.usatoday.com/tech/science/2008-0605-saqqara-missing-pyramid_N.htm, The Associated Press (2008).
[8] Alexanian, N., Schiestl, R., Seidlmayer, S.J., The necropolis of Dahshur. Third excavation report Spring 2006, Free
University of Berlin (2006).
[9] Jquier, G., Fouilles Saqqarah. Deux Pyramides du Moyen Empire, Le Caire (1933).
[10] The United States Geological Survey Spectroscopy Laboratory (USGS, http://speclab.cr.usgs.gov/) and the Jet
Propulsion Laboratory Spectral Library (JPL, http://speclib.jpl.nasa.gov/) (2011).
[11] Wilkinson, T.J., [Archaeological survey of the Tell Beydar region, Syria, 1997: A preliminary report], In Van
Lerberghe, K., Voet, G. (eds.), Tell Beydar Environmental and Technical Studies., Subartu 6, Turnhout, 1-37 (2002).
[12] Ur, J.A., CORONA satellite photography and ancient road networks: a northern Mesopotamian case study,
Antiquity, 77 (295), 102-115 (2003).
[13] Beck, A., Archaeological site detection: the importance of contrast, Proceedings of the Remote Sensing and
Photogrammetry Society (2007).
[14] Yamaguchi, Y., Kahle, A.B., Tsu, H., Kawakami, T., and Pniel, M., Overview of Advanced Spaceborne Thermal
Emission and Reflection Radiometer (ASTER), IEEE T Geosci Remote, 36, 1062-1071 (1998).
[15] Fujisada, H., Design and performance of ASTER instrument, P SPIE, 2583, 16-25 (1995).
[16] United States Geological Survey., Declassified satellite imagery1, http://eros.usgs.gov/Find_Data/Products
_and_Data_Available/Declassified_Satellite_Imagery_-_1 (2011).
[17] Lavrov, V.N., Mapping with the use of Russian space high resolution images, International Archives of
Photogrammetry and Remote sensing XXXIII, 760-767 (2000).
[18] Altaweel, M., The use of ASTER satellite imagery in archaeological contexts, Archaeol. Prospect., 12, 151-166
(2005).
[19] Parcak, S., Satellite remote sensing methods for monitoring archaeological tells in the Middle East, Journal of
Field Archaeology, 32, 65-81 (2007).
[20] Bassani, C., Cavalli, R.M., Fasulli, L., Palombo, A., Pascucci, S., Santini, F., and Pignatti, S., Integration of
airborne optical and thermal imagery for archaeological subsurface structures detection: the Arpi case study (Italy),
European Geosciences Union, 11, 7717 (2009).
[21] Cronin, F., Egyptian pyramids found by infrared satellite images, http://www.bbc.co.uk/news/world-13522957,
BBC News World (2011).
[22] Corrie, R.K., Ninomiya, Y., and Aitchison, J.C., Applying Advanced Spaceborne Thermal Emission and
Reflection Radiometer (ASTER) spectral indices for geological mapping on the Tibetan Plateau, International Archives
of the Photogrammetry, Remote sensing and Spatial Information Science, XXXVIII, 464-469 (2010).
[23] Robinson, C.A., Potential and applications of radar images in Egypt, Egypt. Jour. Remote Sensing, 1, 25-56
(1998).
[24] Parcak, S., Site survey in Egyptology, In Wilkinson, R.H. (ed.), [Egyptology Today], Cambridge University
Press, Cambridge, 57-76 (2008).
[25] Beck, Google Earth and World Wind: remote sensing for the masses?, Antiquity, 80 (308), (2006).
[26] Sherratt, A., Spotting tells from space, Antiquity, 78 (301), (2004).
[27] Menze, B.H., Ur, J.A., and Sherratt, A.G., Tell spotting Surveying Near Eastern settlement mounds from
space, Proceedings of the CIPA International Symposium, Torino, Italy (2005).
[28] Interview with Dr. Zahi Hawass, director of the pyramids, http://www.pbs.org/wgbh/nova/pyramid/excavation/
hawass.html. NOVA Online (1997).
[29] Butzer, K., [Early hydraulic civilization in Egypt], University of Chicago Press, Chicago (1976).
[30] Jeffreys, D., Archaeological implications of the moving Nile, Egyptian Archaeology, 32, 6-7 (2008).
[31] Lutley, K. and Bunbury, J.M., The Nile on the move, Egyptian Archaeology, 32, 3-5 (2008).
[32] Hillier, J.K., Bunbury, J.M., and Graham, A., Monuments on a migrating Nile, Journal of Archaeological Science,
34 (7), 1011-1015 (2007).
[33] McCauley, J.F., Schaber, G.G., Breed, C.S., Grolier, M.J., Haynes, C.V., Issawi, B., Elachi, C., and Blom, R.,
Subsurface valleys and geoarchaeology of the Eastern Sahara revealed by shuttle radar, Science, 218, 1004-1020
(1982).
[34] McCauley, J.F., Breed, C.S., Schaber, G.G., McHugh, W.P., Issawi, B., Haynes, C.V., Grolier, M.J., El Kilani, A.,
Paleodrainages of the eastern Sahara the radar rivers revisited (SIR-A/B implications for a mid-Tertiary trans-African
drainage system), IEEE Trans. Geosci. Rem. Sens., GE-24, 624-648 (1986).
[35] Wendorf, F., Close, A.E., and Schild, R., A survey of the Egyptian radar channels: An example of applied
archaeology, Journal of Field Archaeology, 14, 43-63 (1987).
[36] Wilkinson, T.J. and Tucker, D.J., [Settlement Development in the North Jazira, Iraq: A Study of the Archaeological
Landscape], Aris and Phillips: Warminster (1995).
[37] Ur, J.A., CORONA satellite photography and ancient road networks: a northern Mesopotamian case study,
Antiquity, 77 (295), 102-115 (2003).
[38] Wilkinson, T.J., French, C., Ur, J.A., and Semple, M., The geoarchaeology of route systems in northern Syria,
Geoarchaeology, 25 (6), 745-771 (2010).
[39] Darnell, J., 2010. Email conversation with the author, November 2, 2010.
[40] Darnell, D. and Darnell, J., Exploring the Narrow Doors of the Theban Desert, Egyptian Archaeology, 10, 24-26
(1997).
[41] Short, N., Remote sensing tutorial, NASA Reference Publication 1078, http://rst.gsfc.nasa.gov/ (2010).
[42] Vijayaraj, V., Younan, N., and OHara, C.G., Quantitative analysis of pansharpened images, Optical Engineering,
45 (4), 1-12 (2006).
[43] Di Iorio, A., Bridgwood, I., Carlucci, R., Bernardini, F., Rasmussen, M.S., Srensen, M.K., Osman, A., Automatic
detection of archaeological sites using a hybrid process of Remote Sensing, GIS techniques and a shape detection
algorithm, IEEE-ICTTA08, 3rd International Conference (2009).
[44] Di Iorio, A., Straccia N., Carlucci, R., Advancement in automatic monitoring and detection of archaeological sites
using a hybrid process of Remote Sensing, GIS techniques and a shape detection algorithm, Proceeding of the 30th
European Association Remote Sensing Laboratories (EARSeL) Symposium, Paris, France (2010).
[45] Menze, B.H., Ur, J.A., and Sherratt, A.G., Detection of ancient settlement mounds: archaeological survey based on
the SRTM terrain model, Photogrammetric Engineering and Remote Sensing, 72 (3), 321-327 (2006).
[46] Sherratt, A., Tellspotting, ArchAtlas,http://www.archatlas.dept.shef.ac.uk/Tellspotting/TellsMain.php (2006).
[47] Menze, B.H. and Ur, J.A., Classification of multispectral ASTER imagery in archaeological settlement survey in
the Near East, Proc 10th Intl. Symposium on Physical Measurements and Signatures in Remote Sensing, Davos,
Switzerland (2007).
[48] Menze, B.H., Muhl, S., and Sherratt, A.G., Virtual survey on north Mesopotamian tell sites by means of satellite
remote sensing, In B. Ooghe, G. Verhoeven (eds.) [Broadening horizons: Multidisciplinary approaches to landscape
study], Cambridge Scholars Publishing, Newcastle upon Tyne, UK (2007).
[49] NASA Goddard Space Flight Center, Landsat 7 Science Data Users Handbook, http://landsathandbook.
gsfc.nasa.gov/ (2011).
[50] Jensen, J.R., [Introductory digital image processing: A remote sensing perspective], Prentice Hall (2005).
[51] Chavez, P.S., Berlin, G.L., and Sowers, L.B., Statistical method for selecting Landsat MSS ratios, Journal of
Applied Photographic Engineering, 8, 23-30 (1982).
[52] Chavez, P.S., Guptill, S.C., and Bowell, J.A., Image processing techniques for thematic mapper data, Proc.
ASPRS-ACSM Tech. Paper, 2, 728-742 (1984).
[53] Smith, A.M.S., How to convert ASTER radiance values to reflectance: an online guide, http://www.cnrhome.
uidaho.edu/default.aspx?pid=85984, College of Natural Resources, University of Idaho (2007).
[54] Milder, J.C., [ASTER processing method], Department of Natural Resource, Cornell University (2008).