Dead or Alive - Tom Clancy [130]
“That’s some nitty-gritty Cold War shit right there,” Janet Cummings said.
“It’s tried and true—goes back to ancient Rome.”
The fact that her colleagues seemed surprised by this turn of events told her that they—and perhaps the CIA at large—were still working with a perceptual deficit when it came to the URC’s intelligence capability. Providing the agents working the dead drops were careful, the system was an effective way to make secondhand exchanges.
“No way to know if they’re still active, though,” she said. “Not without boots on the ground.”
The phone at Ben Margolin’s elbow trilled. He picked up the handset, listened for thirty seconds, then hung up. “Nothing so far, but the computers are chewing away at it. The good news is, we’ve eliminated a sixty-mile radius around the cave.”
“Too many variables,” John Turnbull, head of Acre Station, said.
“Yep,” Janet Cummings, the NCTC’s Chief of Operations, replied.
Mary Pat Foley’s idea for solving the “Where in the world is this?” riddle surrounding the sand table Driscoll and his team had recovered from the Hindu Kush cave involved a CIA project code-named Collage.
The brainchild of some mathematician in the Langley science and technology directorate, Collage had been out of Acre Station’s frustration in answering a question to Mary Pat’s, in their case, “Where in the world is he?” The Emir and his lieutenants had long been fond of releasing photos and videos of themselves traipsing about the wilds of Pakistan and Afghanistan, giving the U.S. intelligence community plenty of hints about the weather and terrain of their locations but never enough to be of any help to UAVs or Special Forces teams in the area. Without larger context, points of reference, and reliable scale, a rock was a rock was a rock.
Collage hoped to solve that by collating every available piece of raw topographical data, from commercial and military Landsat images to radar imaging satellites such as Lacrosse and Onyx, to family photo albums on Facebook and travelogues on Flickr—as long as the image’s location could be solidly fixed and scaled to some point on earth, Collage put it into the hopper for digestion and spit it out as an overlay of the earth’s surface. Also into this mix went a dizzying array of variables: geological characteristics, current and past weather patterns, timber-use plans, seismic activity… . If it involved the earth’s surface and how it might look at any given time, it was fed into Collage.
Questions no one thought to ask, such as, “What does the granite in the Hindu Kush look like when it’s wet?” and “In what direction would a certain shadow lean with thirty percent cloud coverage and a dew point of x?” and “With ten days of twelve- to fourteen-mile-per-hour winds, how high is this sand dune in Sudan likely to get?” The permutations were daunting, as was the mathematical modeling system buried within Collage’s code structure, which ran far into the millions of lines. The problem was that the math wasn’t based solely on known variables but also imaginary ones, not to mention probability threads, as the program had to make assumptions about not only the raw data but also what it was seeing in an image or a piece of video. In, say, a thirty-second 640×480 video, Collage’s first pass would identify anywhere from 500,000 to 3,000,000 points of reference to which it had to assign a value—black or white or grayscale (of which there were sixteen thousand)—relative size and angle of the object; distance from its foreground, background, and lateral neighbors; intensity and angular direction of sunlight or thickness and air speed of cloud cover, and so on. Once these values were assigned, they were fed into Collage’s overlay matrix, and the hunt began for a match.
Collage had had some successes, but nothing of real-time tactical significance, and Mary Pat was beginning to suspect the system was going to come up short here,