January 21, 1997
What are scientists seeing over the rainbow? The CIA would like to know
It sounds like the opening to a spy novel: The Central Intelligence Agency awards a $70,000 contract to a university for the study of Western rainbows.
In reality, though, the CIA is recognizing more than two decades of pioneering work by a University of Washington computer image analysis group. The recently announced contract is the agency’s first with the Remote Sensing Laboratory, part of the Department of Geological Sciences.
The title of the award might sound bizarre, but it does not refer to looking at rainbows so much as looking through them. The UW researchers analyze views of the ground, shot from satellites and aircraft, in the form of a type of photograph called a multispectral image. The images are made in different parts of the spectrum, such as infrared and visible light, much as the wavelengths of sunlight are separated in the formation of a rainbow.
From these computer-enhanced pictures, ground features, such as minerals or vegetation, can be identified. In this case, the UW researchers will make a detailed analysis of aerial photographs taken by the intelligence agency over deserts in the Western U.S.
“The C.I.A. wants our insights,” says Alan Gillespie, a UW professor of geological sciences. “But this is a two-way thing, because we can also learn from their data.”
The CIA’s interest in remote-sensing research is for intelligence- gathering. “Our main objective is to be able to detect enemy targets in simulated military environments,” comments Bob Alexander, the project monitor for the agency. However, Gillespie stresses that none of the lab’s work is classified.
The spy agency gave birth to modern photoreconnaisance in 1960 when its first successful information-gathering satellite was launched, the first of 94 that made up the super-secret Corona program. But today’s researchers, using digital code instead of the human eye, are just as likely to be interested in differentiating between soil and vegetation, or picking out deciduous from coniferous trees over vast areas.
It is not hard to see why the intelligence agency would want the UW’s expertise. Using computer digital analysis, the remote sensing lab has developed the ability to identify materials such as leaves, wood, soil and rocks by calculating the amount of light that each reflects. The spectral data from the five Landsat satellites, launched by NASA since 1972 (the latest of which collects radiation in seven wavebands), has enabled the UW lab, among other achievements, to make geological maps, to study surface weathering of deserts, and to chart the deforestation of the Amazon Basin.
The aerial data received by UW researchers is in digital form, with each number representing the brightness of a scene at a certain point. The numbers are called pixels, or picture, elements, and their tones and shading can compose a photograph of the scene.
Researchers, however, are much more interested in studying the scene pixel-by-pixel to see if there is detail about trees or soil or shadow, or how much of each there is in one picture element. In this way the pixels can reveal much about vegetation, minerals and other surface features.
Early Landsat data arrived in a form that made it difficult to distinguish between certain features — for example, crops and roads running along the edges of fields. But today, researchers at the remote sensing lab can use computers to analyse images and break them down into their constitutent parts. “We can figure out what is 90% corn and what is 10% asphalt,” says Gillespie.
This is done by studying “thumbprints,” the characteristic spectrum displayed by each constituent of a scene. Vegetation, for example, has a unique signature in the visible and infrared regions of the spectrum, and cannot be confused with rocks or soils. Researchers can describe a scene on the ground in terms of a few simple common signatures, such as corn or gravel.
In this way, the UW researchers often can pick out over 99% of what they call “background” detail. Only a fraction–perhaps 1%– of “foreground” detail does not fit the usual signature patterns. It is that tiny amount of information– what the CIA calls “military targets”–that the intelligence agency is interested in.
“In this case, the CIA already knows the answers to what is in the foreground,” says Gillespie. “They just want to see if our research is up to finding it.”
If the remote sensing lab can do that, then the intelligence agency will probably apply the technique to future interpretations of its data. And the UW lab can probably look forward to more CIA contracts in the future.
Alan Gillespie is at (206)-685-8263
or at alan@oz.geology.washington.edu
### <!—at end of each paragraph insert
—>
