Geographical Profiling

Geographical profiling draws on two foundational ideas from environmental criminology.

Distance-decay

Most offenders commit most offences close to home. Frequency falls as distance from the anchor grows. The fall isn't linear, it's closer to exponential, and there's a small dip in the very first hundred metres or so because nobody wants to be spotted on their own street. That little dip is the buffer zone, and the function as a whole is the distance-decay function.

Routine activity theory

Cohen and Felson published this in 1979 in American Sociological Review. The claim is plain. A predatory offence happens when three things meet in space and time: a motivated offender, a suitable target, and the absence of a capable guardian. Take any of the three away and the offence doesn't occur. Geographical profiling cares about this because offenders meet targets along their own routine paths, the route to work, the cafe near college, the shortcut through the park. Anchor points generate routine paths, routine paths generate offence opportunities, and offence locations therefore leak information about anchor points.

The technique developed across three research groups over roughly twenty years.

• Paul and Patricia Brantingham (1981, Simon Fraser). Environmental Criminology. The Brantinghams modelled the geometry of crime sites and gave us the mental-map idea: offenders carry awareness spaces of nodes (home, work, leisure) connected by routine paths, and offences cluster in the awareness space.
• David Canter (1989 onward, University of Liverpool). The circle hypothesis. For a series of linked offences, in a high proportion of cases the offender's home lies inside a circle drawn through the two most-distant offences. Canter called these marauders, the offenders who orbit a fixed base. The opposite type, commuters, travel out from a base to a hunting ground and come back.
• Kim Rossmo (1995, Simon Fraser, then Vancouver Police). Took the Brantingham geometry, added a distance-decay function with a buffer zone, and produced a computable probability surface. This became Rossmo's formula, packaged in the software Rigel.

Rossmo is the primary citation for the formula; Canter for the marauder-commuter distinction; the Brantinghams for the environmental-criminology foundation underlying the entire method.

Rossmo's 1995 formula computes, for every cell in a grid covering the search area, a probability score that the cell contains the offender's anchor. The score combines two terms. One term grows as you move away from the buffer zone, because frequency is rising in the working range. The other term shrinks as you move far away from the offence locations, because frequency falls in the long-tail decay. The cell with the highest combined score is the best guess.

The formula appears in textbooks as a double-summation with a switching function for inside and outside the buffer zone. Understanding what it computes and what it assumes is more important than deriving it.

What it assumes:

• The offences are linked. Same offender. If linkage is wrong, the profile is rubbish.
• The offender has one anchor. Multi-anchor offenders (one home, one workplace, one girlfriend's flat) confuse the model.
• The distance metric matches the offender's mental geography. Straight-line distance works in some cities. In Mumbai, where the road network forces detours, network-distance metrics work better.
• Five or more offences. Below that, the surface is too flat to be useful. Rossmo's own validation work assumed at least five sites and worked best with ten or more.

The Canter circle hypothesis works on marauders. It fails on commuters. A common misapplication is to use the circle hypothesis on a commuter series and produce a confidently wrong inference.

Geographical profiling lives inside GIS software. Two tools dominate the academic literature.

• Rigel (Environmental Criminology Research Inc., Vancouver). Rossmo's commercial implementation. Sold to law-enforcement agencies. Closed source.
• CrimeStat (Ned Levine for the US National Institute of Justice). Open and free. Includes a Bayesian journey-to-crime module and several distance-decay options. Used widely in research and in police forces that cannot afford Rigel.

The analyst's workflow is short but disciplined.

1. Confirm linkage
   Establish that the offences in the series share a single offender. Linkage usually rests on MO, signature, victim profile and forensic ties. If linkage is shaky, geoprofiling cannot save you.
2. Geocode the sites
   Convert each offence address into latitude-longitude coordinates with enough accuracy that the cell-level grid is meaningful. In Indian practice this is the step that fails most often.
3. Choose a distance metric
   Straight-line, network distance, or travel time. Network distance is the better default in dense Indian cities.
4. Run the surface
   Feed the points and metric into Rigel or CrimeStat. Output is a heat-map probability surface across the search area.
5. Interpret with the IO
   Hand the surface to the investigating officer along with a hit-rate estimate. Highest-probability cells get prioritised for door-to-door, CCTV pulls, and informant tasking.

A good profile cuts the search area for the next investigative step by 70 to 90 percent. It does not name a suspect. It tells the IO where to look first.

Geographical profiling in India has lived mostly in research papers and Indian forensic-science classrooms. The operational footprint is small but growing.

• Delhi Police Crime Mapping Cell. Operational since 2015 in some form. Uses ArcGIS-based hotspot mapping more than full Rossmo-style profiles, but the data infrastructure for proper geoprofiling is in place.
• Mumbai Crime Branch CID GIS unit. Crime-mapping work since the early 2010s. Has produced offender-anchor inferences in a handful of high-profile serial cases.
• Bengaluru City Police. Smart-City-funded GIS deployments and pilots with academic partners at IIIT-Bangalore on journey-to-crime modelling.
• Hyderabad and Chennai. Smaller GIS units, mostly hotspot work, occasional offender-anchor inference in dowry-death and chain-snatching series.

The bigger structural shift is NCRB CCTNS. The Crime and Criminal Tracking Network and Systems, rolling out since 2009 and now interlinking most of the country's police stations, holds geocoded FIR data at scale. The 2023 CCTNS 2.0 upgrade and the BNSS 2023 push toward digital case records have meant that, for the first time, an analyst at NCRB or a state crime branch can pull a coherent series across districts. The pre-condition for geographical profiling, a linked series with accurate locations, is finally becoming achievable.

The most common over-claim, in textbooks and in oral summaries alike, is that a geoprofile identifies an offender. It does not. It produces a probability surface. The IO still has to do the door-to-door, pull CCTV from the high-probability cells, run informants, match against the offender database. The profile narrows the haystack. The needle still has to be lifted out by police work.

Two related limits.

• No psychological inference. Geographical profiling tells you about where the offender lives, not who they are. The psychological side belongs to behavioural evidence analysis and criminal motivation and typologies.
• Sensitive to linkage error. If two of your five offences were actually committed by a different person, the profile is contaminated and the surface points at no one's anchor. Linkage discipline matters more than algorithm choice.

For the upstream investigative spine that geographical profiling slots into, see history of criminal profiling and victim profiling and victimology.