Interpolation methods
Three of the most common
interpolation methods are Inverse Distance Weighted (IDW), Spline,
and Kriging.
IDW takes the concept of
spatial autocorrelation literally. It assumes that the nearer a sample point is
to the cell whose value is to be estimated, the more closely the cell’s
value will resemble the sample point’s value.
Spline virtually guarantees you a
smooth-looking surface. Imagine stretching a rubber sheet so that it passes through
all of your sample points.
Kriging is one of the most complex and
powerful interpolators. It applies sophisticated statistical methods that
consider the unique characteristics of your dataset. In order to use Kriging interpolation properly, you should have a solid
understanding of geostatistical concepts and methods.
How do I determine which interpolation
method to use?
The
type of interpolation method you use will depend on many factors. Rather than
assume one interpolation method is better than another, you should try
different interpolation methods and compare the results to determine the best
interpolation method for a given project.
Your
real-world knowledge of the subject matter will initially affect which
interpolation method you use. If you know that some of the features in your
surface exceed the z value, for example, and that IDW will result in a surface
that does not exceed the highest or lowest z value in the sample point set, you
might choose the Spline method.
If
you know that the splined surface might end up with
features that you know don’t exist because Spline
interpolation doesn't work well with sample points that are close together and
have extreme differences in value, you might decide to try IDW.
The quality of your sample point set can affect your choice of interpolation method as well. If the sample points are poorly distributed or there are few of them, the surface might not represent the actual terrain very well. If you have too few sample points, you might experiment with adding more sample points in areas where the terrain changes abruptly or frequently, then try using Kriging.
Regardless of the method
you use, you should thoroughly understand your data and the phenomenon
you’re trying to model before interpolating.