IMAGINiT Technologies White Paper 1

Surface Smoothing Using Kriging

Surface Smoothing Using Kriging

AutoCAD® Civil 3D® has the ability to add data points to a surface based on mathematical criteria. This
gives you the ability to strengthen the surface definition in areas where data may be sparse or where
you would like to generate data extending beyond the current TIN. Civil 3D accomplishes this using
Natural Neighbor Interpolation (NNI), a geometric based procedure, or Kriging, a statistical method. NNI
is straightforward, operates within the current TIN boundary and requires no knowledge of the surface
distribution. Kriging allows you to fill in areas within the TIN or extrapolate beyond the existing TIN
boundary. Using Kriging effectively requires knowledge of the surface distribution and an understanding
of the statistical process involved. Although more complex, Kriging is more adaptable and can yield
more accurate results.

Surface Distributions
The following table shows the basic statistics for two surface populations represented by point elevation data. You can see
that they look quite similar. Even the histograms drawn below the table give the impression that the surfaces are similar.
The histograms plot the number of points (Y axis) versus the elevation range (X axis).

Population A B
Elevation Point Count 16137 16137
Average 100 100
Standard Deviation 20.2 20.2
Median 100.3 100.8
10 Percentile 73.6 73.8
90 Percentile 125.73 124.31

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 2
Surface Smoothing Using Kriging

However, when we look at the surface plots, we see that they are dissimilar. The surface on the left representing population
A is “rougher”, having more abrupt gradations between elevations and spatial variations between the high and low areas,
while the surface on the right representing population B is more homogenous. Since both surfaces have the same standard
deviation, we cannot say that one surface is more variable than the other. The difference is in the texture; the way the
variations relate to each other in space. Standard statistical methods do not take into account this spatial relationship.

The Variogram
Kriging relies on a statistic called the variogram. The variogram graphically represents the spatial continuity of the surface
population you are working with. The variogram plots shown below illustrate the different nature of the two surfaces. Each
variogram shows the experimental data as connected dots and the theoretical model as a solid line. You can see the marked
difference in the initial slopes. The variogram on the left (population A) has a steep slope and then levels off sharply, while
the one on the right (population B) shows a more gradual slope and leveling. The point where the variogram levels off is
called the sill. What the variogram for population A shows is that there is only a short distance where continuity exists
between data points. In other words, the surface is rough or discontinuous.

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 3
Surface Smoothing Using Kriging

The mathematics of the derivation of the variogram are complex and rather dry, so we’ll skip most of that and just look at the
equation in its final form with a brief explanation of what it represents. The equation for the variogram is:

This form of the equation often is called the semi-variogram, since the usual derivation ends up with the left side of the
equation being written as . Regardless of the form, the result is the same.

So, what’s this all mean for our surface? Well, the first thing to note is that the variogram is a function of separation of the
points (the ∆x and ∆y) rather than the actual location (x, y) of the point. The right hand side of the equation represents the
expectation (ε[ ] is called the expectation operator) of the inverse measure of the statistical dependency of the random
variables Z(x + ∆x, y + ∆y) and Z(x, y). This expectancy gives us the theoretical variogram. There are an infinite number of
models that can be derived for the theoretical variogram. In practice, only a few are used. Civil 3D 2011 makes use of five
variogram models; Linear, Monomial, Spherical, Exponential and Gaussian. When you undertake surface smoothing in Civil
3D 2011, you will compare one of these theoretical models to the experimental data which is gathered from actual surface
points. While the variogram equation shown above is great for the theoretical model, it isn’t much use for our experimental
surface data. For that, we need another equation to calculate our experimental variogram. That equation is:

The term (∆x, ∆y) as we mentioned before, represents the spatial difference, sometimes called the separation vector, of the
points included in the set of the points gathered from the surface population. The experimental variogram is calculated for a
particular separation vector by averaging one-half the difference squared of the z-values over all pairs of the observations in
the set whose separation vector is approximately equal to that particular vector.

Once the experimental variogram has been calculated, it can be compared to a theoretical variogram so that we can derive
the additional points we require to smooth the surface in the area of interest. One thing to note when using Kriging is that
this is a very computation intensive process, since processing even a moderate number of points selected from the surface
generates a significant number of iterations. In fact, if you select n points, there will be n(n – 1)/2 pairs processed.

Using Kriging in Civil 3D 2011

Let’s look at an example of surface smoothing using the Kriging procedure in Civil 3D 2011. Here’s a sample surface shown
using a TIN style.

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 4
Surface Smoothing Using Kriging

We’ll select the Smooth Surface command from the surface edits right click command popup menu.

This command launches the Smooth Surface dialog shown below. Until Kriging is selected as the smoothing method, the
dialog will appear slightly different.

Variogram
Plot Window

Variogram
Parameters

Now, let’s look at each of the areas within this dialog. Clicking the left mouse button in a field will activate any value
selection methods available for that field (drop down lists, drawing object selection buttons, etc.)

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 5
Surface Smoothing Using Kriging

Select Kriging as the smoothing method.

The Kriging Method section allows us to select the variogram model, point selection method and the points themselves.

Variogram Model
Civil 3D contains five theoretical variogram models.

Variogram Model Graph

Linear

Monomial

Spherical

Exponential

Gaussian

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 6
Surface Smoothing Using Kriging

Once you’ve selected the input points for the experimental variogram, Civil 3D 2011 will plot these points and the
theoretical variogram in the window at the bottom of the dialog. The Linear variogram is the default model for this
dialog.

Point Selection Method

This field is a dropdown that allows you to specify the points to use for the surface smoothing extrapolation. The
choices are:
 Select Points which selects all points inside a rectangle, polygon, surface, or parcel.
 Random Points which selects random points inside a rectangle, polygon, surface, or parcel.
 Select All Points which selects all points on a surface.

Select Points
This field allows you to select the actual points in the drawing. It is available only when either the Select Points or
Random Points parameter is selected. Click the button and follow the command line or dynamic cursor prompts
to select the points. Once the points are selected, the number of selected points will be visible in the field.

The Point Interpolation / Extrapolation Output section lets us choose our output parameters.

Output Locations
This field specifies the output location for the points.
 Grid Based outputs points on a grid defined within specified polygonal areas selected in the drawing.
 Centroids outputs points at the existing surface triangle centroids within specified polygonal areas selected
in the drawing.
 Random Points outputs a specified number of random points within polygonal areas selected in the
drawing.

Select Output Region

This field specifies the area where the points will be output. Click the button and follow the command line or
dynamic cursor prompts to select the area.

Grid X - Spacing
This field specifies the X distance between the grid lines. This is available only when the Grid Based option for point
output is selected. Click the button to digitize the spacing in the drawing area.

Grid Y - Spacing
This specifies the Y distance between the grid lines. This is available only when the Grid Based option for point
output is selected. Click the button to digitize the spacing in the drawing area.

Grid Orientation
This field lets you specify the orientation direction for the grid in the Y and X directions. It is available only when the
Grid Based option for point output has been selected. Click the button to pick two points in the drawing from
which to define the orientation direction.

Number of Output Points

This field displays the number of points that will be output to the surface.

Once the input parameters have been chosen and the original surface points selected, Civil 3D 2011 will draw the variogram
model (red line) and the experimental variogram (points). You can compare the points to various theoretical models by
selecting the model from the Variogram Model field dropdown. To the left of the variogram plot are three parameters
relating to the variogram. More specifically, these are:

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 7
Surface Smoothing Using Kriging

Parameter A
This is the distance along the X axis where the sill takes effect. This is often called the lag of the variogram. The
Linear variogram model has no sill, so if you choose this model, you will not see this field.

Parameter C
This is the distance along the Y axis where the sill takes effect. This is often called the scale of the variogram.

Nugget Effect
The nugget effect is measured by the distance above zero on the Y axis where the experimental variogram meets
that axis. The nugget effect represents the distance between points where there is no expectation of measurable
variance. As the name suggests, it represents a nugget or concentration of data. Originally, Kriging was developed
for use in the mining industry (Krige, for whom this analysis is named, is a South African mining engineer), to try to
estimate the size and grade of an orebody based on data from a number of drill holes. One of the considerations
was, for any ore intersection in a drill hole, did this intersection represent a larger volume of ore or was this just a
coincidental intersection with a small concentration of ore. The nugget effect represents this lower limit of
confidence in the separation vector between pairs of points.

Here’s a smoothing done on our sample area using a Gaussian model. I’ve used the TIN triangle centroids as
outputs, and have adjusted the scale and the lag of the variogram to try to fit the data better. Adjusting the scale
and lag is a bit of alchemy; you’ll have to experiment to find the combination that works best for the surface sample.
You also should look at different models to see how they fit the data. You can see that there are 270 output points
generated by this Kriging.

© IMAGINiT Technologies
www.rand.com/imaginit
 IMAGINiT Technologies White Paper 8
Surface Smoothing Using Kriging

Here is a before and after image of our surface TIN for the sample area.

Original Surface Smoothed Surface

About the Author: Jeff Morrow

Jeff, a senior civil and geospatial consultant with IMAGINIT, has over thirty years experience in the mining and
engineering sectors. He has been involved with AutoCAD and the Autodesk civil software applications since their earliest
days, and has consulted on the development of standards and procedures, CAD management, database and technical
applications, training and implementation, and project management. Jeff is recognized for his work in mineral
exploration, minesite reclamation and environmental issues. He is a graduate of McGill University, and is an AutoCAD
Civil 3D Implementation Certified Expert. You can reach him at [email protected].

About IMAGINiT Technologies:

IMAGINiT Technologies, a RAND Worldwide Company, is a leading provider of design and engineering solutions to the
building, manufacturing and infrastructure, media and entertainment industries. With over 40 offices in North America
and Asia-Pacific, IMAGINiT is a leading provider of Autodesk software solutions and the largest North American Autodesk
Authorized Training Center (ATC) partner. IMAGINiT delivers knowledge, expertise and design processes to clients
through proven technical support, training and consulting services. Our locations are supported by a vast pool of
engineering resources that are focused on developing real-life business solutions for their local clients. For more
information please visit www.rand.com/imaginit or email us at: [email protected].

All brand names, product names, or trademarks belong to their respective holders.
Rand IMAGINiT Technologies, Inc. is not responsible for typographical or graphical errors that may appear in this document.
© 2009 Rand IMAGINiT Technologies, Inc., All Rights Reserved.

© IMAGINiT Technologies
www.rand.com/imaginit