You are on page 1of 26

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

GEOGRAPHICALLY
DISTRIBUTED DATA
(SPATIAL ANALYSIS)

Geographically Distributed Data

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Geologic data

Presented in a spatial form as maps, crosssection or block diagrams,


The geology is portrayed as filling a two or three
dimensional space
The raw data required for analysis of point
distribution are simply the coordinates, often grid
references, of each point.
The purpose of analysis is to identify and
distinguish homogeneity vs. heterogeneity,
isotropy vs. anisotropy, and randomness
Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

REGULAR

UNIFORM

RANDOM

Types of point
distribution

CLUSTERED

ANISOTROPIC
Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Geologic data

Randomness in a point distribution means that the


process which influenced the position of a point has
operated independently of the position of other points.
It has involved equal probability of occurrence in each
equal subdivision of the total area
Non-randomness can take many forms. The degree of
homogeneity can be greater or less than that of a
random distribution;
Although probability is homogeneous in random
distribution, the result has a degree of heterogeneity (or,
in extreme case, regular); with more heterogeneity they
become clustered.
Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Geologic data
Random

distribution are isotropic,


having no directional relationship
between points no fabric. Other
distribution may be anisotropic by
virtue of points forming lineation or
tending to be closer in some
directions than others
Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity

Testing for uniformity


Uniform distribution have points distributed with fairly
constant density over the area, they therefore how
homogeneity.
The recommended test for uniformity is highly intuitive
and straightforward
The area is divided into a number of equal, usually
square or rectangular, sub-areas, and the number of
points in each sub-areas is counted. The test involves
X2

Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity


X2 test for uniformity of point distribution
Ho : point distribution uniform
H1 : point distribution heterogeneous.
2
1 k
O j E
E j 1
2

Where k = the number of sub area , and E = (total number of points) / k

Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity

Example 1 : The thin section of limestone with burrow structures,


the position of burrows have been interpreted as points. We used
fifteen squares (dibagi 15 kotakan untuk analisa), dividing part of
the area into (5 x 3) sub-areas. The frequencies in these are:
21 18 19 16
14
20 16 18 17
16
16 18 16 18
13
Ho : uniform point distribution
H1 : non uniform point distribution
There are 256 points in 15 sub-areas, so the null model for the
data has 256 /15 = 17.066 point in each sub-area.
Using X2 = (Oj Ej)2 / Ej , the calculation is as follows:

Geographically Distributed

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity


21

20

18

16

19

18

16

17

14

<<< non-uniform points distribution

16
17.066 17.066

16

18

16

18

17.066

17.066

17.066

13
17.066
17.066 17.066

Uniform points distribution >>>

17.066 17.066
17.066

Geographically Distributed

17.066

17.066
9

17.066

17.066

17.066

17.066

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Oj

Ej

(Oj Ej )2 / Ej

21

17.066

0.906

18

17.066

0.051

19

17.066

0.219

16

17.066

0.066

14

17.066

0.551

20

17.066

0.504

16

17.066

0.066

18

17.066

0.051

17

17.066

0.000

16

17.066

0.066

16

17.066

0.066

18

17.066

0.051

16

17.066

0.066

18

17.066

0.051

13

17.066

0.969
X2 = 3.687

d.f v= k 1 = 14 = 0.05 critical X2 = 23.68 , fail to reject Ho ; data


are compatible with uniform
distribution
Geographically
Distributed
10

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity

Example 2:
The distribution of locations in a map, we divided by using in 3
x 3 array, the frequencies are:
3
5
3
6
7
9
7
5
7
Using Ho and X2 test as in example 1, there are 52 points in
nine areas, so we expect 52 / 9 = 5.777 points in each subarea according to the null hypothesis

Geographically Distributed

11

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for uniformity


Real points

Expected points

Geographically Distributed

12

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Oj

Ej

( Oj Ej )2 / Ej

5.777

0.105

5.777

1.335

5.777

1.335

5.777

0.009

5.777

0.259

5.777

1.797

5.777

0.259

5.777

0.105

5.777

0.259
X2 = 5.463

v=k1=8
= 0.05 critical X2 = 15.51
are compatible with uniform distribution.

fail to reject Ho ; data

Geographically Distributed

13

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for Randomness

Randomness in geological point phenomena, like uniformity, implies


effective homogeneity of the geologic medium, but additionally,
implies independence of points from each other. Truly random
natural point phenomena are probably rare in geology, but
randomness in geochemical survey sample site distribution is often
advocated.
The distribution of random points on a plane has much in common
with random points along a line, and hence with point events
through time, the characteristics of the distribution can be fitted to a
Poisson model.
Significant departure from the model implies non-randomness; this
can be direction of greater homogeneity (uniformity) or
heterogeneity (clustering)

Geographically Distributed

14

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for Randomness

X2 test for randomness in point distributions


Ho : point distribution random
H1 : point distribution tending towards uniformity or clustering
k

(O j E j ) 2

j 1

Ej

Where Oj is the number of sub-areas containing j points, and Ej is


the number expected according to the Poisson model.
There is normally one class for each number of points per sub-area,
but the number of classes k is determined so that each class has an
expected frequency of five or more.

Geographically Distributed

15

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for Randomness


The expected number of sub-area containing j points is calculated from the
following, based on the Poisson equation:

E j Te n / T

(n / T ) j
j!

Where n is the total number of points and T is the number of sub-area. For
j = 0, j! ( j factorial) is taken to be 1.
Example : we use the same data as in previous slide:
21
20
16

18
16
18

19
18
16

16
17
18

14
16
13

Geographically Distributed

16

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for Randomness


Ho : points distributed randomly
H1 : points non-random
Using X2 to test for Poisson distribution, where the expected number Ej of subarea with r points is

E j Te n / T (n / t ) r / r!
Using grid squares in a 10 x 6 array, we have: T = 60 , and n = 256
Te (-n/T) = 0.8417
Note: as usual for the X2 test, the number expected in each class should be more than five, so some
classes have been amalgamated in this and the following examples to get closer to this rule of
thumb. The final class includes the tail of the Poisson distribution, and the expected frequency is
calculated simply by subtracting the sum of others from the total, in this case 60.

Geographically Distributed

17

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

r : no. in subarea

Oj : no. of subareas

Ej

0.842

3.591

4.433

7.665

5.792

10

10.896

0.074

29

11.623

25.982

13

9.918

0.957

7.053

0.157

4.299

2.532

2.293

1.824

4.117

60

60.00

X2 = 44.04

(O j E j)2 / E j

d.f. v = k 2 = 6 ; = 0.05 , critical X2 = 12.59


Calculated value exceeds critical value, so we reject Ho; points are not
randomly distributed.
Geographically Distributed

18

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Example 2, with the same data in previous slides:


3
6
7

5
7
5

3
9
7

Using X2 to test the same hypothesis as above, using a 6 x 6 array of grid square:
n = 52 ; Te (-n/T) = 8.4916
r

Oj

8.492

0.030

13

12.266

0.044

8.859

0.390

4.265

1.540

0.445

0.134

36

36.0

Ej

T = 36,

(Oj Ej)2/Ej

0.059

X2 = 0.5237

V = k 2 = 2; = 0.05 , critical X2 = 5.99 ; fail to reject Ho, data compatible with points being
randomly distributed.

Geographically Distributed

19

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for clustering and regularity

There is a spectrum of possible point distributions


ranging from clustered to random to uniform to regular.
The hypothetical end member at the clustered end has
all points coincident; at the regular extreme there is
perfect equilateral spacing.

It would be no surprise to find grains of a certain mineral


clustered in xenolith, fossil sessile organisms to be
. clustered on local hard substrates or hydrothermal ore
deposits clustered around intrusions; the probability of a
point forming is not constant over the whole area in
these circumstances

Geographically Distributed

20

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for clustering and regularity


The nearest neighbor test for clustering and regularity
Ho : point distribution is random
H1: point distribution tends towards regularity or clustering

Where:

d
Z
se

d is the observed mean nearest neighbor distance,


is the expected mean nearest neighbor distance and
se is the standard error of the mean nearest neighbor distance
Geographically Distributed

21

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for clustering and regularity


We calculate these by:

1 n
d di
n i 1
Where:
di = distance from point i to the closest other point its nearest neighbor and
n = number of points;

Geographically Distributed

22

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for clustering and regularity


0.26136
se
n2 / A
Where: A is the areas of the region in which point distribution data are available (usually
the map area); and:

A/ n
2

The Z statistics, as usual, has critical values of 1.96 and 1.96 at = 0.05. If Z <
1.96 , we reject Ho and accept an alternative hypothesis of clustering; if Z > 1.96, we
reject Ho and accept an alternative hypothesis of uniformity or regularity

Geographically Distributed

23

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

Testing for clustering and regularity

WARNING:
There is an bias in this statistics due to existence of the
edge of the area being investigated; will tend to be over
estimated if close neighborhoods are ignored because
there are off the edge of the map!
This is best overcome by establishing a buffer zone: the
edge of the area for which the test is to apply is
withdrawn from the limit of the area for which data are
available.
Nearest neighborhoods measurement are then permitted
to straddle the new edge.

Geographically Distributed

24

DJOKO WINTOLO TEKNIK GEOLOGI FT-UGM

THANKS YOU

Geographically Distributed Data

25

Geographically Distributed

26

You might also like