Image Analysis Update

Pamela Logan, Research Director
China Exploration & Research Society


Synthetic aperture radar images have been acquired of southern Xinjiang, an area traversed by China's ancient Silk Road. Analysis of the images is underway, and despite many obstacles, it shows good promise of producing a list of potential ruins sites. Purchase of visible-wavelength imagery to supplement the radar will clarify many ambiguous features, shortening the list of search ares to the most promising ones. These will be visited by a ground-truthing mission in 1997 when a team of CERS explorers and Chinese archeologists will venture into the Taklamakan Desert to seek evidence of the Silk Road's distant past.

Successful Acquisition of Radar Images

The year 1994 marked NASA's "Mission to Planet Earth," in which two flights of Space Shuttle Endeavor carried apparatus known as Space Radar Laboratory (SRL). This radar system is the most advanced ever flown, and represents a new generation of synthetic aperture radar technology. The apparatus consisted of two separate devices. The European-made X-SAR system operates at X-Band (3-cm wavelength), and transmits and receives vertically polarized signals. American-made SIR-C operates at two wavelengths, C-Band (6 cm) and L-Band (24 cm); it transmits and receives both horizontally and vertically polarized signals. The result is up to nine independent image channels that give a picture of the earth's surface never before possible.

Our area of focus is a strip of desert between the towns of Hotan and Ruoqiang, north of the modern-day highway that runs parallel to where ancient Silk Road caravans must have crossed the desert. In this region the April Endeavor mission successfully captured twelve swaths of the southern Taklamakan. The October mission captured three more, filling in gaps. These strips of imagery, which vary in width from 26 to 73 km and are typically 1000 km long, can be pieced together into a mosaic that covers most of the region of interest.

Image Analysis Procedure

Processing of radar images is a lengthy and computer-intensive procedure that has been underway since May 1994 and is still not yet complete. Low resolution, or "survey product" images are now available for most of the data-takes. These have a pixel size of 50 m x 50 m, and CERS scientist Pamela Logan has been using them as a guide for selection of higher resolution segments. High resolution images, whose pixel size is 12.5 meters square, and which require much more computer time to generate, are slowly being released to the dozens of researchers who await SIR-C/X-SAR data. CERS has so far been able to examine about a dozen high resolution Silk Road images; more are on order and will be issued over the next few months.

When the high resolution data is released, each individual researcher must still complete the processing according to the needs of his or her application. Using specialized image processing tools donated by PCI, a Canadian software company, Dr. Logan has been manipulating the data, searching for the optimum combination of wavelength, polarization, and enhancement that best distinguishes man-made objects from sand.

Radar Detective Work

The obstacles to image interpretation are many. Radar responds in a complex manner to terrain; scattering and absorption of the signals depends on such parameters as surface slope, moisture content, and vegetation coverage. Even in a desert as inhospitable as the Taklamakan (whose name means "go in and you don't come out!") there is water beneath the surface that affects radar scattering. Vegetation and sand can produce similar cross-polarization scattering, so it's difficult to assess surface conditions without first-hand knowledge of the area in question.

Deep in the Taklamakan, dunes as large as 8 km long, 2 km across, and 50 meters high are common, and they effectively obliterate any sign of whatever man-made features may lie beneath them. Rivers are a necessary pre-condition to human habitation and an obvious thing to look for in the radar images. But although it's possible in principle to locate dry riverbeds or underground streams using synthetic aperture radar, the constantly-changing meanders of a typical Taklamakan river make their signatures fuzzy at best. Moreover, the best resolution possible is 12.5 meters, which means that most man-made structures are no more than one pixel in size. Composed of crumbling clay, they are difficult to distinguish from piles of natural sand.

Extracting the Desert's Secrets

Despite the above-mentioned obstacles, the images offer tantalizing clues: areas of unusually high radar reflectivity, dark spots indicating higher-than-average moisture, and slender lines barely visible in the desert floor. Where or not these shadowy hints are the result of man's activities cannot be determined with certainty from the radar images alone. Yet it's impractical to send a ground-truthing expedition to all the dozens of suspicious areas that appear on the radar; a means of reducing the number of potential targets must be found.

The answer to this dilemma is to acquire more remote sensing information. This is readily done. Three nations have spaceborne imaging systems which have acquired visible and infrared wavelength data of the earth's surface: Landsat TM (U.S.A.), SPOT (France), and SOVINFORM-SPUTNIK (Russia/USSR). Choosing which image to purchase is a matter of weighing factors like coverage, spectral properties, resolution, and cost; this Dr. Logan's current task. With the acquisition of one or more additional images, many ambiguous features can be identified and excluded from consideration.

With the list of targets narrowed, she can plot a course among the dunes that will make best use of the limited time available to the ground-truthing team during its sojourn in the desert.

Return to Silk Road Project home page