Kernel Density Crime Mapping
Also known as: KDE Crime Mapping, Crime Density Surface Mapping, Hot Spot Density Mapping, Kernel Smoothing of Crime Events
Kernel density crime mapping turns a scatter of geocoded crime points into a smooth, continuous surface that shows where incidents concentrate. Each event is spread out over a small neighborhood by a kernel function, and the overlapping contributions are summed across a fine grid so that areas with many nearby crimes glow as peaks. Chainey, Tompson, and Uhlig (2008) showed that, among common hot-spot mapping techniques, kernel density estimation is one of the most accurate at predicting where future crime will occur, which is why it became the default crime-mapping surface in policing.
Key highlights
- Transforms cluttered dot maps into an intuitive, continuous surface that communicates concentration at a glance.
- Empirically among the most accurate simple techniques for predicting where future crime will occur.
- Produces a smooth prediction surface that can be thresholded into hot spots and validated against later crime.
- Requires only point locations, making it applicable to nearly any geocoded crime dataset.
- Implemented in essentially all GIS and spatial-statistics software, ensuring accessibility and reproducibility.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use kernel density crime mapping when you have geocoded point events and want a clear, continuous picture of where crime concentrates — for briefing, tasking, and short-term forecasting of likely future locations. It excels at visualization and at producing prediction surfaces that can be evaluated and is the standard for operational crime mapping. It is less appropriate when you need a formal statistical test that a cluster exceeds chance (use Getis-Ord Gi*), when you want to explain why places are risky (use risk terrain modeling), or when events are too sparse to form meaningful density. Because results hinge on the bandwidth, it should be used with transparency about that choice rather than as an objective measure of concentration.
Strengths & limitations
- Transforms cluttered dot maps into an intuitive, continuous surface that communicates concentration at a glance.
- Empirically among the most accurate simple techniques for predicting where future crime will occur.
- Produces a smooth prediction surface that can be thresholded into hot spots and validated against later crime.
- Requires only point locations, making it applicable to nearly any geocoded crime dataset.
- Implemented in essentially all GIS and spatial-statistics software, ensuring accessibility and reproducibility.
- The surface depends strongly on the bandwidth, so the same data can show many small or a few large hot spots.
- It is descriptive and does not test whether any concentration exceeds what spatial randomness would produce.
- Raw density reflects crime counts, not risk per capita or per opportunity, unless normalized by an at-risk population.
- Edge effects can distort estimates near study-area boundaries where part of each kernel falls outside the data.
- It explains neither the causes of concentration nor any temporal dynamics in a single static surface.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How should I choose the bandwidth?
The bandwidth is the single most influential setting, so choose it deliberately rather than by default. Options include fixing it to a meaningful spatial scale (a block, a beat), using rules of thumb such as those discussed by Silverman, or applying data-driven selectors. Good practice is to set the bandwidth a priori on substantive grounds, report the value used, and show how the surface changes under alternative bandwidths so readers can judge sensitivity.
Is a kernel density hot spot the same as a statistically significant cluster?
No. Kernel density estimation shows where incidents are dense but does not test whether that density exceeds chance. A region can look hot simply because more people and activity are there. To claim a cluster is statistically significant you need a test such as the Getis-Ord Gi* statistic or a point-pattern method; KDE is best treated as a visualization and prediction surface rather than an inferential test.
Should I map raw crime counts or rates?
Raw kernel density reflects the number of events, so dense areas often coincide with dense populations or activity. If the question is about risk to people or targets, you should normalize — for example by estimating a density of crime relative to a density of population or opportunities (a dual-kernel or risk-surface approach). Whether to map counts or rates depends on whether you care about the volume of crime or the relative risk.
Sources
- 1.Chainey, S., Tompson, L., & Uhlig, S. (2008). The utility of hotspot mapping for predicting spatial patterns of crime. Security Journal, 21(1–2), 4–28.
- 2.Silverman, B. W. (1986). Density Estimation for Statistics and Data Analysis. Chapman and Hall.ISBN 9780412246203
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Kernel Density Crime Mapping. ScholarGate. https://scholargate.app/criminology/kernel-density-crime-mapping