Lecture #08: Spatial Data

Analysis –Variograms_02
 Variogram
 Variogram is the most
traditional choice to summarize
spatial continuity, in comparison
to correlation function and
covariance.
 Evidence of spatial dependence
 The experimental variogram provides evidence
that there is local spatial dependence.

The variability between point-pairs is lower if they are
closer to each other; i.e. the separation is small.

There is some distance, the range where this effect is
noted; beyond the range there is no dependence.

The relative magnitude of the total sill and nugget give
the strength of the local spatial dependence; the
nugget represents completely unexplained variability.
 There are of course variables for which there is
no spatial dependence, in which case the
experimental variogram has the sill equal to the
nugget; this is called a pure nugget effect
Regionalized variables

blue dots represent the data

 Regionalized variables

The structural com ponent (e.g., a linear trend)

The spatially correlated com ponent

The random noise com ponent (non-fitted)

 Regionalized variables
 Regionalized variables are variables that fall
between random variables and completely
deterministic variables.
 Typical regionalized variables are functions
describing variables that have geographic
distributions (e.g. elevation of ground surface).
Unlike random variables, regionalized variables
exhibit spatial continuity; however, the change in
the variable is so complex that they cannot be
described by any deterministic function.
 The variogram is used to describe regionalized
variables
 The Semi-Variogram
 An estimator of the heterogeneity of spatial
processes.

How contaminant concentrations vary with location?

Do small changes in location imply small changes in
concentration?

Are there distinguishable consolidated “hot-spots”?

Are there large scale trends in concentration?
 The variogram provides a reconnaissance tool
to evaluate these questions.
 Variograms

The sem i-variogram is

based on m odelling the
(squared) differences in
the z-values as a function
of the distances betw een
all of the know n points.
 Use of Variogram
 An estimator of the heterogeneity of spatial
processes.

How contaminant concentrations vary with location?

Do small changes in location imply small changes in
concentration?

Are there distinguishable consolidated “hot-spots”?

Are there large scale trends in concentration?
 The variogram provides a reconnaissance tool
to evaluate these questions.
 Semivariance
 This is a mathematical measure of the difference
between the two points in a point-pair.
 It is expressed as squared difference so that the order of
the points doesn’t matter (i.e. subtraction in either
direction gives the same results).
 Each pair of observation points has a semivariance,
usually represented as the Greek letter (‘gamma’), and
defined as:
 (xi, xj) =12[z(xi) − z(xj)]2

where x is a geographic point and z(x) is its attribute value.

(Note: The ‘semi’ refers to the factor 1/2, because there are two
ways to compute for the same point pair.)

So, the semivariance between two points is half the squared
difference between their values. If the values are similar, the
semivariance will be small.
 The variogram cloud
 This is a graph showing semivariances between
all point-pairs:

X-axis The separation distance within the point-pair

Y-axis The semivariance
 Advantage: Shows the comparison between all
point-pairs as a function of their separation;
 Advantage: Shows which point-pairs do not fit
the general pattern
 Disadvantage: too many graph points, hard to
interpret
 Variograms
In graphical terms:

Why ‘semi’-variogram? As the lag distance approaches infinity,

the expression converges to twice the variance. Therefore, dividing by 2
means that the sill approximates the variance.
Some Terminologies of Variogram
 Range is the distance (h) at which the
variogram reaches its plateau.
 Sill is the plateau that the variogram reaches
at the range.
 Nugget effect is the vertical jump from the
value of zero at the origin to the value of
variogram at extremely small separation
distances due to sampling error or short scale
variability.
 Features of the experimental
variogram
 Sill ~ Total variance among distant
samples. For completely independent data
this simplifies to sample variance S2
 Nugget Effect ~ Measure of small scale
spatial heterogeneity.
 Range of Influence ~ Distance at which
samples are uncorrelated.
 The semi-variogram is a measure of
dissimilarity
 Variogram components
 Nugget variance: a non-zero value for  when h =
0. Produced by various sources of unexplained
error (e.g. measurement error).
 Sill: for large values of h the variogram levels out,
indicating that there no longer is any correlation
between data points. The sill should be equal to the
variance of the data set.
 Range: is the value of h where the sill occurs (or
95% of the value of the sill).
 In general, 30 or more pairs per point are needed
to generate a reasonable sample variogram.
 The most important part of a variogram is its shape
near the origin, as the closest points are given
more weight in the interpolation process.
 Variograms

This is an exam ple of

a variogram produced
using A rcG IS 's
G eostatistical A nalyst.
 Variograms
 The technique can provide a quantification of the
scale of variability exhibited by natural patterns
of resource distributions and an identification of
the spatial scale at which the sampled variable
exhibits maximum variance.
 At larger lag distances (beyond the natural
‘scale’ of the phenomenon) harmonic effects can
be noted, in which the variogram peaks or dips
at lag distances that are multiples of the natural
scale.
 Given the noise present in natural environmental
data sets, it is unlikely that you will be able
clearly to identify multiple scales.
 Defining the bins
 There are some practical considerations, just like
defining bins for a histogram:

Each bin should have enough points to give a robust estimate of
the representative semi-variance; otherwise the variogram is
erratic;

If a bin is too wide, the theoretical variogram model will be hard
to estimate and fit; note we haven’t seen this yet, it is in the next
lecture;

The largest separation should not exceed half the longest
separation in the dataset; in general it should be somewhat
shorter, since it is the local spatial dependence which is most
interesting.
 All computer programs that compute variograms use
some defaults for the largest separation and number of
bins; gstat uses 1/3 of the longest separation, and
divides this into 15 equal-width bins.
 Evidence of spatial dependence
 The experimental variogram provides evidence
that there is local spatial dependence.

The variability between point-pairs is lower if they are
closer to each other; i.e. the separation is small.

There is some distance, the range where this effect is
noted; beyond the range there is no dependence.

The relative magnitude of the total sill and nugget give
the strength of the local spatial dependence; the
nugget represents completely unexplained variability.
 There are of course variables for which there is
no spatial dependence, in which case the
experimental variogram has the sill equal to the
nugget; this is called a pure nugget effect
 Omnidirectional Variogram
 “Omnidirectional” means all directions. It
combines all directional variograms into a
single semivariogram.
 Omnidirectinal variogram can be loosely
thought of as an average of the various
directional variograms.
 However, the calculation of omnidirectional
variogram does not imply that spatial
continuity is the same in all direction.
Purpose of Omnidirectional Variogram
 It serves as a useful starting point for
producing a clear structure of overall spatial
continuity.
 Since the omnidirectional variogram contains
more sample pairs than any directional
variogram, it is more likely to show a clearly
interpretable structure.
 It can help detect some erratic directional
variograms if an omnidirectional variogram is
messy.
 Choosing the Distance
Parameters
 Two distance parameters need to be
chosen:
 Lag spacing (lag increment)
- Assuming the samples are regularly
spaced over the study area, set the initial
spacing as:
sqrt (study area/total number of points)
 Lag tolerance = ½ lag spacing
Effect of lag size on variograms

Variogram with a lag size of 5m Variogram with a lag size of 10m

and a lag tolerance of 2.5m. and a lag tolerance of 5m.
 Choosing the Distance
Parameters…
 Distance parameters can be set differently
according to directions, if the sampling pattern
is noticeably anisotropic.
 For samples located close to each other, it is
possible to include an additional lag with small
separations and a small tolerance for this first
lag. E.g. 0.5m, 1m, 2m, 3m…
 Pattern of Anisotropy
 After having an acceptable omnidirectional
variogram, we can explore the pattern of
anisotropy with various directional variograms.
 Isotropy vs. anisotropy.
 The direction of maximum spatial continuity and
the direction of minimum spatial continuity .
 Pattern of Anisotropy…
 Choose a directional tolerance that is large
enough to allow sufficient pairs for a clear
variogram, yet small enough that the character
of the variograms for separate directions is not
blurred beyond recognition.
 Choosing the Directional
Tolerance
 After identifying directions, it is time to choose
directional tolerance.
 Directional tolerance should be large enough to
contain enough pairs and small enough to
avoid blurring the anisotropy resulted from
combining pairs of different directions.
 Try several tolerances and use the smallest
one that still produces good results.
 Sample Variograms for U
 Hole effect is a certain dip that occurs in a sample
variogram.
 It happens when there is a natural cyclicity or
repetition in a data set.
 Sample Variograms for U …
 There are many data sets for which the
sample variogram is simply inappropriate as a
descriptive tool.
 The variogram works well when there is no
proportional effect or no obvious clusters in a
data.
 Fluctuation in the lag means is a problem.
 Relative Variogram
 To account for varying means, relative
variograms scale the original variogram to
some local mean value.

Local Relative Variogram

General Relative Variogram
Pairwise Relative Variogram
 Local Relative Variogram

 Local relative variogram divides an area in

regions and treats data in each region a separate
population, Fig 7.16.
 A local variogram is scaled by the local mean.
Then the local variograms are summed to form
the overall variogram
i (h)
i 1 N i (h) m 2
n

LR (h)  i

i 1 N i (h)
n
 General Relative Variogram
 General relative variogram does not require
small populations which may cause local
variograms to be erratic.
 Instead, the semivariance for each h is
adjusted by the mean of all the data values that
are used to calculate the semivariance.
 ( h)
GR (h) 
m( h) 2
1 mh  m h
m( h)   vi  v j 
2 N (h) ( i , j )|hij h 2
 Pairwise Relative Variogram
 Pairwise relative variogram also adjusts the
variogram calculation by a squared mean.
 This adjustment, however, is done separately for
each pair of sample values, using the average of
the two values as the local mean:

2
1 (vi  v j )
PR (h)   vi v j 2
2 N (h) (i , j )|hij h ( 2 )
 Anisotropy
 There may be higher spatial autocorrelation in one
direction than in others, which is called anisotropy:

 The figure shows a case of geometric anisotropy,

which is incorporated in the variogram model by
means of a linear transformation.
 Semi-variogram tips
 We are assuming a normal distribution
 Gives us a picture of the relationship of
data values with distance.
 If you don’t have a good spatial structure
in the semi-variogram, don’t revert to IDW.
 Comparing Software for
Computing the Semi-Variogram

Practical Geostatistics 2000 ArcGIS Geostatistical Analyst

Assessing Fit of the Variogram
 Cressie Goodness
of Fit

For each point
used to create the
variogram, match
how well the
model actually fits
it
 Variogram characteristics – the
influence range

 The influence range is the minimum

distance between two independent random
variables.
 It is also the zone of influence of a random
variable.