## 1352AI-GEOSTATS: basic theoretical question

Expand Messages
• Dec 16, 2003
I have had no courses in geostatistics what so ever but I'm trying to
calculate the semivariance of a dataset on my own. Calculating
everything from scratch.

But I have some basic questions conserning the calculations that have to
be done.

First of all, I have data that is stratified on one level but not so on
another. a sample plot would look like this:

x x x
x x
x x x
x x
x x x

a few of these sample sites were randomly distributed in a grid pattern
of 9x9 km with grid cells of 1 km, 5 sites per cell.

so I have about +- 500 points sampled. (is this enough??)

So goes my question, is semivariance a good idea? Given the fact that a
sampling plot is 20x20m and the whole site 9x9km? Does it make sense to
include all values or should I only consider the 1x1km grid cells as
maximum. In the end, correlations between points in 1 site will be long
lost if compared to the ones in the other plots. Any idea's on this.

Anyway, if I go with calculating the semivariance for the whole 9x9km
what should I do. I made a flowchart of what should be considered this
is what I came up with.

first, calculate the distances between the pairs of sample points, and
the square of the differences between the attributes assigned to these
points.

If I have these points, because the site has no random distribution and
in general there are not that many points that have the same distances
you will have to reclass these distances and associated attributes into
classes to have your lag distances. From what I have read this h
distance may not be more then half the maximum distance between two
points.

This would leave me with lags, h, from the reclassed distances the
squared difference of attribute values and a count of observations of
lags per class (N).

As far as I know this are all the necessary values needed to calculate
the semivariance with following equation:

gamma(h)=1/2N(h)*sum((z(x)-z(x+h))^2

Am I'm doing this right???
Any input on this matter would be greatly appreciated.