In the
article, A Better Method to Smooth Crime Incident Data, crime data from the New
York City Police Department On-Line Complaint System is used to compare the
differences between methods of kernel estimation. Kernel estimation is the
spatial statistical method used to generate density maps. Police departments
use crime density maps to identify crime hotspots and plan patrol schedules, so
it is important that the map is as accurate as possible. In this week's lab we
used crime data from Washington DC to generate crime density maps and propose a
location for a new police station. The kernel estimation method used in the lab
was based on the areal extent of Washington DC. However, as this article points
out, there is a more accurate method for kernel estimating.
Kernel
smoothing estimates the variation in density of events based on a point
pattern, resulting in a map of smooth density values. Kernel estimation is
successful in making sense of complex point patterns. The most important step
in kernel estimation is selecting the bandwidth. Maps generated with a small
bandwidth are spiky in appearance, and those maps generated with a large bandwidth
appear smooth and generalized. As with either method, density maps can then be
used to create other datasets for further analysis.
Most GIS
programs base its kernel estimation on the areal extent of the event data
without considering the spatial distribution of the points. The result is large
bandwidths being selected for small sample sizes and small bandwidths for the large.
This article proposes selecting the bandwidth based on a predetermined number
of points, or neighbors, represented by the variable 'k' called the k-nearest neighbor
method. This method bases the bandwidth on the average distances between the
event data. Varying the value of 'k' allows the GIS analyst to specify the
degree of smoothing which reveals the previously unrealized variation in
density across the study area. This added capability must be used with caution
though, as the user defined input can still result in a misleading or
inaccurate map.
http://www.esri.com/news/arcuser/0199/crimedata.html
No comments:
Post a Comment