Asthma HosPitalization Rates
Figure
9.21 Interpolation
bY
t(-o
6
Bronx CountY, New York CitY
inverse distance weighting (lDW), using census tract centroids
as
@o
I
the samPle Points'
t-
rn
N
(
A
I-rdu* Rot"" Pt
3 O l--l
t'Rl [Link]
Stahonary Polni Slrrrces
Ma-ior Highrvals,Tntck Rottlci 1.000 PoPtilcliolr -l - 0 Std Dev.
n n I
fl
t\.tan
[Link]'
?
Std. Dcv Std. Dev > 3 Std. Dev.
I -2 -3
nve rse d ista nce weig hted
points of known The inverse distance weighted (lDW) method of interpolation takes sample lt assigns values and interpolates a contrnuous surface from the values of those samples.
points of known values to unknown points based on the values of the nearest sample two alternative ways of selecting value. The inverse distance weighted method employs either a specithe sample points that are used in calculating the unknown values' IDW uses points that fied number of nearest neighbors (for instance, the six or eight known sample distance f rom the are nearest to tne unKnown point), or it uses a f ixed radius of a specified point' unknown point. The fixed radius method, in its calculations of value of the unknown This number will includes the number of sample points that are within that specified radius. of sample varv depending on the density and regularity of the sample points' The values point of unknown points used in the estimate are weighted according to distance from the values' The value. with closer points accorded more weight in determining the unknown values can influence of the known values, therefore, diminishes with distance. Estimated IDW interpolatlon' never be less than or greater than the range of the sample point values ln (discrete data), and ends up the GIS analyst begins with a layer of points of known values (regularwith a continuous data layer consisting either of isolines (contour lines) or a raster ized grid)of cellvalues (figure 9.21).
;Ait. o
Methods of Spatial Data Analysis
237
Spline The spline method of interpolation also uses sample points of known values to estimate the value of unknown points, but ratherthan creating a distance weighted average, as IDW does, it fits an imaginary surface through the sample points of known value, similar to a rubber sheet that has been draped over the tops of the sample points, assuming that the sample points have a vertical dimension (a z-value) in proportion to the attribute being estimated. Unknown points are estimated from the selected number of sample points nearby. The spline method generally produces a smoother interpolation, and the estimated points
may be given values that are outside the range of the known values of the sample points. As wlth the IDW method, spline interpolation requires a data layer of sample points, and returns an isoline or a grid Iayer of continuous data (figure 9.22).
L (
IDW- Rate/
1.000
\
I
:.
9.22 Comparison ol interpolation methods: Inverse
Figuse
(
.
-: sta o"u.
Distance Weighting versus Spline, using ZIP Code centroids
as the sample points. Compare
<-3 Std. Dev. ! -3 --2 Std. Dev. | -2 - -l Std. Dev. | i--l -l - 0 Std Dev. lM@ i l0- t StdDev. l-2stdDev. f !2-3StdDev.;
Splire-Rak/ l,0m
l
:
I -3 - -2 Std Dev. I -2 - -l Std t)ev. f--i-l-0stdDev. rMtu I i 0- l Std Dev. I -2stdDev. f
the difference between IDW
interpolation using census tract
centroids in figure 9.21 and IDW
interpolation using ZIP Code
centroids in this ligure.
r-) .r) I L'/ l{-
i
F
tt (
IDW
&nbm
of
tub/ I,m
Kriging
Ktigtng is an intetpolaIion ntetlrocl IlraI n; oltcn 1-rrclcrrcrl lry irrr;rlyr;l:; lrcr;ir(r:;r: rl ir(i(:()rr]l'. random variation in its estimating process. The nrethod takes its narne Irorr tlral ol ;r 1ir, rt' African mining engineer who developed a geostatistical method of [Link] rnincr;rl ,,',
bodies. Kriging is more complex mathematically than the others outlined above and, rr,l would argue, is a truer representation of reality, since it takes into account not only tltc 1rr, imity of the sample points, but also their directions. Kriging uses the values of the sample points in several ways: First it calculates a gerr 'l trend surface, or constant mean value, from all the sample points, referred to as the slrl tural component. Then it looks at localized variation, called the spatially correlated conrtr, nent. Then it takes into account a random noise or residual error component, and comb rr, the three components into a continuous surface of estimated values. Kriging uses a movirl window approach to selecting the sample points, but sample values along the direclr, ,' of the trend have more influence than values going against the trend. This method is bcr,' with regularly sampled data showing a strong trend. As with the spline method, kriging c;r estimate values beyond the range of the samples (figure 9.23).
,
Figure
9.23 Interpolation
by kriging, using ZIP
Code centroids as sample points. Compare to IDW
and spline interpolation methods using ZIP Code
.a-
centroids in fiqure 9.22.
q
Kriging -Rate
0
/ 1,000
-5.45
5.45 -9.61 9.61 -12.80
1
2.80 -1 5.25
ffi: I [Link] -ra.s< I [Link] I [Link] [Link] -[Link] f [Link] zs.u fl rooutu
[Link]
t
"b
t\
:.
a a
ZIP Code centroids
;;\tittt
Methods of Spalial Data Analysis
239
KEY . Crime Site
PBOBABLE
LOCATION
I I
I
f f
3?J3r.
B6-eo"/.
Figure9.24 GIS is increasingly
being used for law enforcement applications. The technique of geographic profiling uses
a
i)ZZ:;
72-760/o
I *:::;
4a-52%
67-71%
predictive model that interpolates from points of crime occurrences.
Copyright @ 2002
fl
ss-stu"
by The
New Y0rk Times
Co.
B-qzv"
39-42"/"
eprinted
bV
permission.
I I il
'txi.:; zq-zax
zo-23./" 15-19%
I
Geographic Profiling: One Example in Ganada
area - the science of predicting where criminals ltve based on where they commrt crimes. The theory behind geographic profiling is that most criminals will operar-e refatively btose t5 nome in i-comfort zone defined by wheie they live, *oif., pfuv
Investigators searching for the Washington
seraal shooter are using geographic profiling
to-u't"
Z:1:;
Dismiss
The above map was generated by geographic profilers working on a lggs case invotving a series of insuranceagent robberies in Vancouver British Columbia. The red dots represent the crimes and tEe red shading shaows the region where the offender will most likely base his operations. This is usuatty referred to as his anchor point.
[Link]. Psychologists call thisthe
pflncrpte.
Source: Envi@nmental Criminology Resealch lnc.
least-effort'
Geographic profilers use specially designed software that takes the locations of a reiated series of crimes along with the known and predictable pattern of a seiial criminal to produce a map ol the probable location of the criminal's anchor point (where the criminal leaves from or returns to after the crime). This can reduce an investigators' prime Search area by more than 90 percent.
Regardless of the method of interpolation used, the good analyst makes sure to review
lhe results very carefully and to note any anomalies. The software may estimate values irdmirably based on complex mathematical formulas, but the results may not necessarily tttake sense. ln other words, always understand your data and investigate results that don't rrtirke sense. At this point in technology, software is not usually able to pick up counterrrrltrrtrve resLrlts-that is still the province of humans.