You are on page 1of 7

Susan Williams

GEOG 586
November 2013

Project 5: Interpolation Methods


Introduction
Where previous lessons discussed the importance of having knowledge of the
underlying process in order to accurately analyze the results, and the importance of
understanding the effects that varying parameters may have on a point pattern
analysis, this lesson similarly explored the various characteristics of surfaces
produced by different interpolation methods.
To explore Interpolation Methods, we experimented with elevation data from Central
Pennsylvania which enabled us to compare the different interpolated surface results
with the known answer in order to accurately analyze the differences (OSullivan,
2014).

Inverse Distance Weighted Method


Inverse Distance Weighted (IDW) Interpolation is a method that operates on the
principle of positive Spatial Autocorrelation in which data points located near one
another in space will likely have more in common than data points that are farther
apart. To do this, IDW calculates the local mean by weighting control points
according to their proximity, with nearer points being weighted more heavily
(OSullivan and Unwin, 2010).
First, a set of elevation points were derived from the existing Digital Elevation
Model. Then the ArcGIS Spatial Analyst tool was used to create an IDW interpolation
surface of the study area.
I experimented with altering the variables and found that a larger number of control
points does result in a smoother surface, even if it is not a drastically increased
number. For example, Figure 1 shows the map interpolated using 12 control points.
Figure 2 shows the map with all other variables the same but the number of control
points was increased to 20.
A Power greater than 1 weights points closer to the location more heavily.
Additional experimentation showed that much higher Power values resulted in a
choppy-looking surface, and a much lower Power value resulted in loss of detail and
a more uniform surface. A more moderate Power value such as the one shown in
Figures 1 and 2 produces the most moderate-looking surface. Many computer
1

programs use this Power setting by default (ibid). It reminded me of the quote,
With great power comes great responsibility!

After experimenting with the different settings for a bit, I decided on the map shown
in Figure 2.

Figure 1: IDW interpolation of Centre County Pennsylvania elevation data using 12 control points.
Compare to Figure 2 where all variables were the same with the exception of the control points,
which were set to 20.

The settings used for this map included 20 points, increased from the setting of 12,
which did smooth the surface a slight amount over Figure 1, a variable search

radius, and a Power of 2.


Figure 2: IDW interpolation of Centre County Pennsylvania elevation data using 20

IDW
haspoints.
advantages
and
ofvariables
course. were
It works
bestwith
on evenly-spaced
control
Compare
todisadvantages,
Figure 1 where all
the same
the
exceptionbut
of the
control
points,
which were
set the
to 12.
datasets
cannot
make
estimates
above
maximum or below the minimum
sample values. This may result in peaks or valleys that are more flattened than in
reality. Additionally, IDW does not factor in underlying trends in the data, such as
values that change more in a sloping direction (ibid). It is, however, arguably the
most straightforward and easily understood interpolation method and can be
calculated fairly quickly. It also produces reasonable results for many kinds of
datasets, although it is always desirable to have a DEM or some other source of
verification, which is not always realistic.

After using the Raster Calculator to subtract the actual elevation from the
interpolated surface, ArcMap produces a map displaying interpolation errors. I drew
contours at 50 m intervals and selected a red-to-blue gradient symbology to

represent the high and low extremes of the errors (see Figure 3 below).

The higher positive values (in darker red) indicate that the interpolated elevation is
much higher than the actual elevation, whereas a high negative value (in darker
blue) indicates that the interpolated elevation is much lower than the actual
elevation. The interpolated values at control points have little error, which is
reasonable since these control points are known values. The margin of error seems
to rise accordingly as distance from control points increases. It also appears that
the IDW interpolation results were especially high or low along peaks and river
valleys, where the interpolation efforts seem to have more or less averaged out the
elevations.
Kriging Using the Geostatistical Analyst
Like IDW, Kriging is a weighted average technique. Unlike IDW, Kriging measures
distances and direction between all possible pairs of control points to estimates the
spatial structure of the underlying surface and determine appropriate weights for
interpolation (ibid). An Isotropic Semivariogram assumes there is no trend in the
data.
Figure 3: IDW interpolation Errors of Centre County Pennsylvania elevation data. The higher
4
positive values (in darker red) indicate that the interpolated elevation is much higher than the
actual elevation, whereas a high negative value (in darker blue) indicates that the interpolated
elevation is much lower than the actual elevation. values

For the Map shown in Figure 4 below, I selected a Gaussian profile, although any
other would likely have been just as valid. Indeed, some experts feel that selecting
a semivariogram model is a highly controversial subject (ibid). Experimentation
with several types did not seem to alter the results very dramatically, if at all.
Figure 4: Ordinary Kriging with an Isotropic Gaussian Semivariogram of Centre County, PA
elevation data.

Where an Isotropic Semivariogram assumes there is no trend in the data, an


Anisotropic Semivariogram accounts for a directional dependency in the data.
While an isotropic model may be a good standard choice, we know from our DEM
that Centre County has some directional variation in the topography, so an
Anisotropic model will likely be a better choice for this particular situation.
For the Map shown in Figure 5 below, I selected a Gaussian profile again and left all
other variables the same as those used in Figure 4, but this time I selected an
Anisotropic model to account for the directional variation in the underlying data.

Figure 5: Ordinary Kriging with an Anisotropic Gaussian Semivariogram of Centre County, PA


elevation data.

The Anisotropic profile does appear to more closely fit the topography seen in the
DEM, and the contour lines are not as round but more elongated along the ridges.
Kriging has its own advantages and disadvantages. Kriging does give an estimation
of error, and the ability to account for anisotropy is also a great advantage. While
it seems to be more realistic than IDW, it is also more mathematically complex.
Conclusion
This project, much like the ones that preceded it, strongly reinforced the concept
that knowledge of underlying patterns and processes is critical to truly understand
the results of an analysis. In this particular instance, we developed an
understanding of variations in different interpolation methods that may affect the
surface outcome. In general, IDW displays a good general surface for normallydistributed data, but it has shortcomings where more extreme values are involved
(such as higher peaks or lower river valleys). Kriging does more accurately account
for such variations, but may still miss the mark if an underlying surface trend exists
that should be included as anisotropy in the kriging method.

Although there is no one best method, a successful analyst will understand the
advantages and disadvantages of each before selecting a method that is bestsuited for the calculation at hand. The best interpolation in the world cannot make
up for poor or incomplete data, nor can it account for all variables and situations.
Another valuable lesson to be gained from these exercises is that a good analyst
acknowledges that GIS is not made up of one-size-fits-all techniques.

References
O'Sullivan, David. (2014). GEOG 586: Geographic Information Analysis, Lesson 5:
Interpolation - From Simple to Advanced. The Pennsylvania State University
World Campus. Accessed November 2013 at https://www.eeducation.psu.edu/geog586/l6.html
O'Sullivan, D., & Unwin, D. J. (2010). Geographic Information Analysis. (2nd ed.).
Hoboken, New Jersey: John Wiley & Sons, Inc.

You might also like