Where do suspicious characters hang out? (Image: Alexandr Tovstenko/Getty Images)
The geographical profiling that catches serial killers can track bats to their roosts or sharks to their lairs – and could close in on deadly diseases too
IT WAS while finishing his PhD on the mating behaviour of sticklebacks that found himself drawn to a life of crime. Le Comber was reading a 2003 interview in Â鶹´«Ă˝ with Kim Rossmo, a Canadian cop-turned-researcher whose mathematical methods for tracking down serial killers were earning him a reputation as a latter-day Sherlock Holmes.
Advertisement
Rossmo’s insight was that felons, by and large, aren’t complicated characters. “Most criminals aren’t Hannibal Lecters,” he says. Often they act instinctively, making it possible to spot patterns in their actions. By studying hundreds of cases, Rossmo found that serial killers rarely target victims too close to home. But they are unlikely to kill a long way from home either: beyond a certain distance, the likelihood of a crime decays steeply, like an inverse power curve.
This insight allowed Rossmo, , to construct algorithms that could be used to home in on the probable whereabouts of a killer. He has since shown that such geographic profiling, which is now in use by law-enforcement agencies worldwide, could help to locate the perpetrators of other crimes, including vandalism and credit-card theft.
The patterns reminded Le Comber, now a researcher at Queen Mary, University of London, of the way sticklebacks hunt for nest material. Stripping the area close to home might leave their nests exposed, but venturing far away would take too much time and energy, so sticklebacks stick to a happy medium distance. Le Comber got in touch with Rossmo, and the pair arranged to meet in London. He had a burning question: if criminals are a predictable species, might others be equally so?
The bat pack
The first test of that wasn’t sticklebacks, but bats. Using Rossmo’s geographic profiling software, he entered the foraging grounds of pipistrelle bats as “crime locations” to see whether he could track down the bats’ roosting sites, the “suspects’ homes”. “To be honest, I didn’t really think it would work,” Le Comber says.
But it did. Compared with the most common method of tracking down animal abodes – simply combing the entire region around where they usually forage – geographic profiling halved the search area. The same was true when Le Comber used the algorithm to track foraging bumblebees, and to find the ground zero of invasive plant and insect species. At around the same time, another group at the University of Miami in Florida also teamed up with Rossmo to show thatgreat white sharks follow a similar hunting strategy to human serial killers.
“Great white sharks follow a similar hunting strategy to human serial killers”
The closeness of the matches came as a surprise to Rossmo. “It turned out to be an algorithm more of general movement patterns,” he says. “In all these situations, it’s sort of the same. It’s movement out of some centre point, which you don’t know.”
Following these first successes, Le Comber wotked with , a mathematical epidemiologist at Imperial College London, and Mark Stevenson, an analyst for a private health research firm, to pick apart Rossmo’s algorithm. Geographic profiling rests on fairly simple mathematics. Around each individual crime location, you draw a probability curve showing the likelihood that the killer lives nearby. You then add up the curves from many crime locations to produce a contour map of the search area in which “peaks” are hotspots where there is a good chance of finding the killer. For a serial killer, the model ideally produces a single, high-probability peak. Where multiple bat roosting sites or shark lairs are concerned, there will be more than one peak.
It soon became clear that, with some tinkering, geographic profiling could be better adapted to the biological world. Le Comber’s team developed a new model that was more suited to figuring out how many “suspect” locations there might be, and which “crimes” belonged to which suspect: for instance, which bat came from which roosting site. The model has proved to be robust. “The only time you get into real problems is when your search area is completely full of crimes,” says Le Comber. It also fails when animals travel out of their way to a special food source.
In work published last month, the team tested the improved model on a spate of malaria cases in Cairo, Egypt, between 2001 and 2004. The usual approach would have been to search for mosquito breeding sites in the area around the largest number of outbreaks – which amounted to roughly 300 square kilometres. The map produced using geographic profiling dramatically shrank the search area, pinpointing some sites to within just kilometres, and even picking out sites that had previously been overlooked ().
Other ways of locating probable mosquito breeding sites exist, such as satellite imaging of waterlogged areas, and Le Comber’s approach could be just as useful in the initial stages, says , who researches disease control at the London School of Hygiene and Tropical Medicine. But mosquito breeding sites may be scattered over thousands of tiny locations – in small ponds and people’s backyards, for instance – so profiling and remote imaging will never be a substitute for thorough on-the-ground surveys of mosquito larvae, she says.
Another problem is that people with malaria can move out of an infectious area, so it becomes difficult to locate exactly where someone became infected, says , a medical geographer at the University of British Columbia in Vancouver, Canada, and author of a book on medical cartography, Disease Maps. This is the same issue faced by those trying to find the source of the H5N1 bird flu strain.
Deadly species
Nevertheless, Le Comber believes that geographic profiling could be applied to many other diseases transmitted by both insects and humans. He and his team are now developing a version of their software that takes advantage of additional data, such as wind direction (which influences where mosquitoes travel, for instance), disease strains and local ecology.
They are also in early discussions with , an agency of the UK government’s health department, to see whether geographic profiling could be used to tackle the spread of tuberculosis and Legionnaires’ disease. “We don’t know whether this technique could be useful for public health at this stage, but it certainly seems sensible to test it,” says , a disease ecologist at Durham University in the UK who specialises in malaria control.
Meanwhile, the wheel may turn full circle. Following the success of geographic profiling in biology, Verity and Stevenson are starting to look into whether the new algorithm might provide fresh avenues for tracking down the deadly species that was the genesis of the technique: serial killers.
This article appeared in print under the headline “Killer app”
