SIR-C Applied to a Silk Road Archeological Survey

April, 1997

Pamela Logan, Director of Research

China Exploration & Research Society

Summary

Synthetic aperture radar images have been acquired of the southern Taklamakan Desert, an area where China's Silk Road once flourished and prosperous cities were built. Because of gradual climate change, many Silk Road communities were abandoned and subsequently lost among the dunes of the Taklamakan. The radar images of these areas reveal features--tracks and spots of unnaturally high reflectivity--that do not appear on visible-light photographs and may be signs of ancient inhabitation. Ground-truthing has not yet been conducted, and will be a crucial step to identifying these features and proving the value of SAR to archeological surveys of the Taklamakan.

Background

The Silk Road is an ancient caravan route that connected China to civilizations of the West. Along its dozens of branches, many ruins, some nearly two thousand years old, have already been identified, and have provided important clues about the ebb and flow of ideas across Eurasia.

The Taklamakan Desert was one of many obstacles facing Silk Road travelers, and caravan routes were obliged to skirt it along its northern or southern rim. Here, a string of oases are fed by water flowing from the surrounding mountains. Because the desert has been gradually drying over the last two millennia, these Silk Road caravansaries were gradually pushed outward toward the water sources at the desert's edge; the cities' original sites were abandoned to encroaching dunes. In these now-hostile areas, many ruins are thought to lie undiscovered.

SIR-C, the latest generation of NASA's Spaceborne Imaging Radar program, was flown on Space Shuttle Endeavor in April and October of 1994. It captured images of a portion of this caravan route in Xinjiang Province, east of the city of Khotan (Hetian) on Taklamakan's southern rim (Fig. 1). Synthetic aperture radar is a particularly promising tool for desert archeology because long-wavelength microwave radiation has been found to scatter from objects buried beneath one to two meters beneath fine, dry sand. Radar scattering is also highly sensitive to the size and orientation of hard objects lying on the sand's surface. In cases where the objects sought are the same color as the sandy background, SAR can be a far more effective detection tool than visible-light photography. Thus there is good reason to believe that use of SAR might result in discovery of previously unknown Silk Road ruins sites.

Prior and Concurrent Work

While satellite and aerial photography have been used by many archeologists to investigate sites of ancient habitation, the advent of SAR has greatly enhanced archeology's arsenal of remote sensing techniques. The availability of SIR-B data inspired researchers to seek the lost city of Ubar, center of frankincense trade in the Omani Desert. The much-publicized discovery of ruins thought to be Ubar was made, in the end, using LANDSAT Thematic Mapper images, but SAR was an important catalyst for the project.

More recently, SIR-C data is being used by archeologist Elizabeth Moore to investigate the ruins of Angkor, the great Khmer metropolis in Cambodia. On the Silk Road, SIR-C images have revealed a possible man-made canal near the settlement of Niya, although we are unaware of any ground-truthing conducted to confirm the discovery. It is expected that within the next few years many SIR-C based archeology projects will reach fruition as more researchers complete ground surveys to confirm sightings from space.

The project described here is companion to the work of Tom Farr, who has used SIR-C data to investigate alluvial fans and geological fault movement in the nearby Kunlun Mountains.

Terrain

The Taklamakan Desert is located within the Tarim Basin, an oval-shaped depression ringed by ranges of high mountains. The basin is 1500 km in length from east to west, and 600 km north to south, with a typical elevation of 1000 meters in the center. Sand dunes are the predominant feature within the desert, the largest of them 150 meters high.

Snowmelt from the encircling mountains flows into the Taklamakan through a system of rivers streaming from alluvial fans. These rivers support oases that are intensely cultivated by people of the Uighur nationality, who have built a string of communities at the desert's edge. Outside cultivated areas vegetation is scanty, consisting of grasses and bushes that feed on underground water, both of which peter out as one moves inward toward the desert's bone-dry center. Dead and dying poplars in the marginal areas testify to the dwindling water supply of the region.

While the Taklamakan is not the largest desert in the world, it is one of the most inhospitable owing to the fineness of its sand--which give poor traction to vehicles and works its way into even the most carefully protected equipment. High winds and extremes of temperature reduce the time when field work is possible to just three or four months a year. Few people except oil explorers and archeologists venture deeply into the desert.

In the past, most archeological sites were discovered by herdsmen or other local people traveling into marginal areas as part of their daily life. Thus, all the land suitable for grazing, or lying on travel routes, is considered to be previously surveyed for purposes of this research. This investigation is aimed not at the marginal areas but at "the deep desert"--places where man has not set foot for hundreds of years.

Types of Objects Sought

Some previously-discovered Silk Road ruins were surveyed by the authors during a reconnaissance mission in 1993. Two types of ancient structures were observed: wood and earth. Wooden ruins are found in Qing dynasty sites (less than 350 years old). They consist of rows of closely­spaced wooden sticks or branches stuck in the sand--the remains of wattle-and-daub walls. The sticks are typically 1­3 cm in diameter and protrude no more than half a meter above the surrounding sand.

The walls of homes are arranged in rectangles that are limited by the length of available timber to a maximum of 7 meters across--significantly less than SIR-C's effective resolution of 25 meters. However the contrast in size-scale and scattering properties between the smooth, hard sticks and soft sand may well create a distinct radar signature. Walls arranged in irregular or round shapes were also observed and must have been unroofed enclosures for livestock.

Solid earth structures are more durable than wooden ones; consequently older sites consist entirely of earth remains. Most of these are eroded walls­­typically very thick (probably up to 2 meters originally)­­with no smooth, vertical surfaces remaining. Examples of this type are shown in Figure 2, which depicts part of a site called Maligawat, near the present-day city of Khotan. (Unfortunately, no SIR-C data was collected at this site). Small settlements may have as few as two or three houses. Large settlements, for example the city of Niya, left ruins spread over an area as great as 10 to 20 square kilometers.

Other features indicative of former settlements are channels for directing water, and dry riverbeds. It is very unlikely that tracks on the Silk Road itself are detectable owing to the widely dispersed paths used by caravans and subsequent movement of sand. Many sites are littered with fragments of ceramic pottery that may serve to increase radar scattering from the desert floor.

SIR-C Images: Processing and Enhancement

As can be seen in Figure 1, a mosaic of overlapping swaths provide fairly good coverage of the area between Khotan and Qiemo, although there was one gap. Other parts of the Taklamakan received considerably less complete coverage.

In order to best distinguish size scales of surface objects and penetrate the fine-grained sand that forms the background of the images, the L-band was expected to be the most useful; experience subsequently bore this out. Most of the data acquired in the area of interest was dual-pole; therefore the choice of polarizations was limited to the HH and HV channels, which were incorporated into the working image as red and green respectively. C-Band HH was assigned to the blue channel. X-band data were not used, being very similar to the C-band.

Various methods of image filtering and enhancement were tried, but none were found particularly advantageous and so the working images were unfiltered. No averaging was performed so that the maximum resolution possible was retained in the working image.

Geo-referencing

Geo-referencing of the SIR-C images proved to be difficult, as the corner coordinates provided with the data are not sufficiently accurate for on-the-ground navigation in such a bewildering environment as the Taklamakan. Sand dunes in the region of interest are typically three kilometers wide by ten kilometers long, and 75 meters high. Neighboring dunes tend to be quite similar in size and shape, therefore the central problem is for the ground-truthing team to be able to distinguish one dune from another using a GPS receiver to fix position, and to relate the dunes to possible archeological features identified on the working radar image.

The georeferencing challenge is compounded by the difficulty of obtaining high-resolution topographic maps of the area. The PRC government generally regards such maps as classified, does not allow them to be transported or transmitted outside the country, and access to them by foreigners is strictly limited, especially in this sensitive area.

The best maps that we were able to obtain outside China were published in 1983 by the Soviet Union, with a scale of 1:100,000 and contour intervals of 40 meters. They do reveal the outlines of large dunes; however these outlines do not coincide very well with dunes shown on the radar images, probably because radar scattering properties are highly dependent on angle of incidence, and hardly at all on elevation. Therefore radar images cannot be expected to show dune "edges" at a uniform elevation in the same way as they appear on the topographic maps. In practice, aligning the two has been found to be virtually impossible.

Further challenges to geo-referencing and ground navigation include:

(1) Dune movement. In the Taklamakan, spring storms bring high winds that annually re-shape the landscape. For the largest dunes there is no great movement over several decades, however it is expected that the angle of slope--and therefore radar scattering properties--of the dunes will change due to deposit or erosion of sand.

(2) Lack of fixed, distinctive landmarks near the search area to serve as ground-control points. Towns and roads do exist but they are located 20 to 100 kilometers away; therefore georeferencing obtained from these features must be extrapolated over a significant distance, tending to compound errors. Moreover, towns have been growing and roads have been changing rapidly over the last few decades, rendering even the best maps inaccurate and incomplete.

Radar Image Features

The most prominent feature of the radar images is dunes; see the sample in Figure 3. It is not obvious, a priori, which parts of the image represent crests and which represent troughs, however Chinese archeologists familiar with the area were unanimous in the opinion that the dark green portions (A) are troughs, and the violet areas (B) are crests. This opinion is supported by the expectation that vegetated areas will tend to scatter more strongly in the cross-polarized (green) channel. The darkness of the troughs is probably due to underground moisture--known to be common in the area--limiting radar penetration. Ripples in the troughs, here running in a direction roughly perpendicular to the large dunes, are probably small- scale waves in the sand.

Superimposed on the background of dunes are a number of mysterious features which will hold the highest priority for the ground-truthing team in search of Silk Road ruins. Unusually bright areas such as C are suspect, although this particular one is perhaps too extensive to be anything but a propitiously tilted dune wall. Coursing through the dune marked by D is a faint, dark line that may be an ancient track or water channel--or it may be the result of visits by oil exploration vehicles. Areas of marked red color, for example the spot above F, indicate larger than average L-Band HH scattering, which may be a sign of large-size-scale objects buried or on the surface.

Future Work

Detailed examination of all the SIR-C images obtained of potential ruins sites is a mammoth task, let alone ground-truthing them. Many "suspicious" objects on the SIR-C images may well turn out to be natural formations; it will be essential to develop the "art" of reading SAR desert images to distinguish natural from man-made, thereby reducing the amount of arduous ground-truthing necessary to confirm (or disprove) discoveries. At present, it is difficult to project how long this process will take, and how soon it will be before finds of any importance can be made. However it is the conviction of archeologists such as Professor Wang Binghua of the Xinjiang Archeological Institute that the sheer density of the Taklamakan's ancient population and the richness of its culture, mean that many discoveries are waiting to be found.

The authors have selected the region around the present-day city of Qiemo as the starting point for this work. However Chinese archeologists believe that other areas, for example west of Lake Lop Nor and the neighborhood of known ruins sites such as Endere and Niya, also hold great promise. Fortunately, the climate of the Taklamakan is favorable to site preservation, so that there is no reason to hurry the investigation. Indeed, a strong argument can be made for delaying discoveries until resources are available to properly excavate, protect, and preserve the fragile artifacts (such as scrolls) that may be unearthed.

Conclusions

Remote sensing in general and SIR-C data in particular show great promise of being helpful to archeologists seeking desert ruins sites. However much remains to be done, especially to compare SIR-C image features with objects on the ground, a process that will enhance researchers' ability to discern man-made relics. Obstacles to this process are inaccurate geo-referencing, inadequate resolution, and incomplete coverage. If SIR-C data can be shown to have value in locating desert ruins, then the information acquired in a few days by Space Shuttle Endeavor may take decades to fully understand and exploit. However, the most important step--groundtruthing--is still ahead.

Figure Captions

Figure 1. Map showing region of interest.

File name: srmap2.jpg

Figure 2. Photograph of one well-known (and explored) ruins site, the city of Maligawat near Khotan. A local villager with a flock of sheep is crossing in front of two relics of the ancient city. In addition to the large earthen structures, the desert floor is littered with shards of pottery left behind by inhabitants some 2000 years ago.

File name: maligawat.bmp

Figure 3. Sample working image. LHH= red, LHV=green, CHH=blue. The size is 12 km x 10 km, the location is 28 kilometers northwest of Qiemo, and north is at the top.

File name: sample.tif

Note: This project has a home page, now located at http://www.alumni.caltech.edu/~pamlogan/silkroad/index.html. It will probably be moved some time in the next few months, but there will be a pointer left at this location.