Geographic Profiling: Rossmo, Canter and the Jeopardy Surface

Kim Rossmo's CGT algorithm, derived from his 1995 Simon Fraser University doctoral thesis, is grounded in two empirical observations. Distance decay: the probability of an offence location decreases as distance from the offender's anchor point increases. Buffer zone: directly around the anchor point, offending probability is suppressed because the offender avoids committing crimes in recognisable proximity to home.

The CGT formula calculates a probability score for each cell of a geographic grid overlaid on the crime sites, summing inverse-distance functions from each scene modified by the buffer-zone dampening factor. The result, rendered visually, is the "jeopardy surface": a colour-coded topographic display showing high-probability zones for the anchor point in red, low-probability zones in blue or green.

The jeopardy surface is the primary investigative output. Investigators prioritise database queries (sex-offender registry, driver's licence data), surveillance, and canvassing by starting in the highest-probability zones. It does not produce a suspect; it tells investigators where to look first.

Rigel software (Environmental Criminology Research Inc., Vancouver) implements the algorithm and has been licensed to agencies in the US, Canada, UK, the Netherlands, Belgium, Australia, and New Zealand. The FBI's Behavioral Analysis Units have incorporated geographic profiling into their consultation practice.

Canter and Larkin published the circle hypothesis in Journal of Environmental Psychology in 1993, the same year Rossmo published CGT. For marauder offenders (those who operate radially from a fixed base), the anchor point falls within the circle defined by the two most geographically distant crime sites as its diameter. The empirical basis was 45 British serial rapists; approximately 87% compliance with the prediction was observed.

The marauder vs commuter typology follows directly. Marauders operate within a radius around a stable anchor point. Commuters travel from that point into a separate crime area and return. Lundrigan and Canter (2001) found the majority of serial murderers showed a marauder pattern across a combined sample of 126 US and 29 UK serial killers; the commuter pattern was more common where victims were transported and when the offender had changed residence during the series. The circle hypothesis performs poorly for commuters, whose crime sites and anchor point are in separate geographic clusters.

Dragnet software (University of Liverpool) implements the Canter approach via both the circle hypothesis and an independently derived decay function, and has been used by the NCA and other UK law-enforcement agencies alongside Rigel.

Brent Snook, Craig Bennell, and their collaborators published a series of papers between 2002 and 2005 that tested the practical performance of CGT and the circle hypothesis against simpler decision heuristics. The core papers are Snook, Canter, and Bennell (2002) in Behavioral Sciences and the Law and Snook, Zito, Bennell, and Taylor (2005) in Journal of Quantitative Criminology.

The 2002 study compared anchor-point-prediction accuracy across four methods in a sample of 54 British serial murderers: CGT (Rigel), Dragnet (circle hypothesis), the geographic centroid (average of all crime sites), and the spatial median. On most accuracy metrics, the simple geometric heuristics performed comparably to the full algorithms. The 2005 paper extended this to untrained participants and found they matched algorithmic accuracy by guessing the cluster centre.

Practitioners have responded that the probability-surface output enables ranked database searches that a single-point heuristic cannot support, because it shows the probability gradient across the whole search area rather than just the highest-probability point. This is a legitimate operational argument, though the Snook et al. papers specifically tested prioritisation accuracy and found the performance gap remained narrow.

Geographic profiling using CGT or Dragnet adds investigative value over unaided geographic intuition, but the magnitude of that advantage over well-implemented simple heuristics is more modest than early practitioner claims suggested. The method's documented contribution to the investigation of the Yorkshire Ripper (retrospective analysis confirming that CGT would have produced a high-probability zone near Sutcliffe's Bradford home), the BTK Killer (Wichita, Kansas, 2004-2005), and the Beltway Snipers (2002) supports its continued operational use, with appropriate acknowledgment of its limitations.

Yorkshire Ripper (Peter Sutcliffe, 1975-1980, UK). A retrospective CGT analysis published by Rossmo showed that the algorithm's jeopardy surface would have placed Bradford, Sutcliffe's home city, in the highest-probability zone based on the crime sites available by 1977. The actual investigation focused on Sunderland following a series of hoax recordings and letters purportedly from the killer. Sutcliffe was arrested by chance during a routine patrol in Sheffield in 1981. The case is routinely cited as the founding motivation for systematic geographic profiling: with a jeopardy surface, investigators might have narrowed the search to Bradford years earlier. The caveat is that this is retrospective validation, which is epistemically weaker than prospective prediction.

BTK Killer (Dennis Rader, Wichita, Kansas, 1974-1991). Rossmo performed a geographic profile consultation on the BTK case in the early 2000s when cold-case investigators reopened it. The jeopardy surface identified a residential area in Wichita's Park City suburb as high-probability. Rader lived in Park City. Rader was ultimately identified through DNA from his daughter's cervical smear, submitted under a court order after Rader communicated with investigators via a floppy disk that was traced to the Christ Lutheran Church where he was a deacon. The geographic profile had identified the correct zone; again, biological evidence provided the identification.

Beltway Snipers (Muhammad and Malvo, October 2002, US). The geographic profile in the Beltway Snipers case is cited as an operational near-miss. The jeopardy surface based on the ten Washington DC-area shooting sites identified a probability zone in the northern Virginia suburb area consistent with the area where a white box truck had been spotted. Profilers simultaneously predicted a lone white male with a small car. Muhammad and Malvo were arrested in their customised Chevrolet Caprice at a rest stop in Maryland after a member of the public called in a tip. The geographic profile came closer to the anchor point than the demographic profile did to the offenders' characteristics, but neither resulted in the identification.

Nithari killings (Noida, Uttar Pradesh, India, 2004-2006). The Nithari case is a partial illustration of geographic profiling's limits in an Indian urban context. The offending (multiple murders of children and women attributed to Surinder Koli) took place in and around a single residential compound in Nithari village, a working-class neighbourhood on the outskirts of Noida. The geographic distribution of victim disappearances was consistent with a marauder model: victims were taken from a tightly defined area and remains were found near the offender's place of employment. But the investigation was resolved not through geographic profiling but through the discovery of human remains in the compound drain and subsequent forensic biological identification. No formal CGT or Dragnet analysis was reported as part of the investigation. The case illustrates that in densely populated urban Indian environments, the spatial dynamics of offending may not produce the spread of crime sites needed to generate a meaningful jeopardy surface.

Geographic profiling functions best under five conditions: the series is genuinely linked (see serial-offending linkage analysis); the offender follows a marauder rather than commuter model; there are at least five crime sites; the city's geography is reflected in the distance-decay parameters; and the anchor point has remained stable during the offending. When these conditions are unmet, the jeopardy surface is substantially less reliable. A commuter offender's home may fall completely outside the surface if marauder-calibrated parameters are used. A single mislinked case can distort the surface dramatically, and when linkage analysis is itself uncertain, geographic profiling and linkage analysis feed circularly into each other.

The method's output is also frequently misread. The jeopardy surface identifies the most probable zone for the anchor point, not the most probable suspect location. The anchor point may be a home address, a workplace, a regular haunt, a vehicle home base, or any other regularly used location. An offender who drives a long distance to work, parks at a commuter lot that is not near home, and offends near the commuter lot will not be predicted by a jeopardy surface calibrated to residential anchor points. The investigator using geographic profiling output needs to think about anchor-point type, not just residential address.

In the UK, the National Crime Agency SCAS provides geographic profiling as part of its consultation service to major crime investigations, alongside behavioural linkage analysis and offender profiling. The SCAS position, consistent with the ENFSI Investigative Psychology Working Group guidelines, is that geographic profiling is one intelligence product among several and should be used alongside, not instead of, conventional investigative strategies. In Australia, the Victoria Police Major Crime Investigation Division incorporates geographic profiling within a broader geographic intelligence framework that includes GIS crime-mapping and spatial trend analysis from the Crime Statistics Agency. In Canada, the RCMP has geographic profiling analysts embedded in its major crime units. In the US, the FBI BAU includes geographic profiling as a component of its consultation capability.