negative range from fit.variogram through autofitVariogram

Hi Mark,

Thanks for the reproducible example. The problem is that when I look at 
the sample variogram, the semivariance values start high and end low. 
This is best illustrated by:

plot(variogram(theta_percent~1, sparse))

You see that there are outliers in the data that cause high 
semi-variance at a short distance. I would say that is 'strange' ;). You 
can try and identify which point causes this by:

plot(variogram(theta_percent~1, sparse, cloud = TRUE), identify = TRUE)
# Click on the plot to identify the point pairs

There is not really one value I think that causes this. It might be also 
attributable to the fact that your dataset is somewhat sparse in the 
distance range from 20-40 m.

As a quick fix you can restrict the model selection to "Sph", this 
works. And I would not be enthusiastic about using Ste. This is because 
the main difference between different kappa values if the behavior at 
short distances, but you don't have a lot of data on the short distance 
to fit this value on in a meaningful way.

At this stage I don't see a (relatively quick) fix that could solve this 
problem in an automatic way and on a more fundamental level. Do you have 
any suggestions? From an implementation point of view I can let automap 
discard any model that has negative values in it, this would ensure that 
the user gets at least the Sph model back.

cheers,
Paul

negative range from fit.variogram through autofitVariogram

Thread (5 messages)