[R-sig-Geo] Help with the Kmeasure function from the Spatstat package

Adrian Baddeley adrian at maths.uwa.edu.au
Thu Dec 7 02:41:02 CET 2006

Debarchana Ghosh writes:

>I'm doing exploratory point pattern data analysis with 285 points in a
>irregular polygon. I'm really exploring and hence trying out all possible
>clustering functions on a single data set. In the 'Kmeasure' function, one
>of the arguments that the user needs to be supplied is the 'sigma', which is
>standard deviation of the isotropic Gaussian smoothing kernel. I don't know
>how to compute that or which function to use to compute that value.

The smoothing parameter is a matter of choice. Although there are 
some rules for automatically selecting a value of the smoothing parameter
based on the data, even these automatic rules need human supervision. 

The best advice is simply to try different values of sigma,
ranging from about 5% to 25% of the diameter of the observation window.
A very blotchy image probably means that sigma was too small, while a
completely flat image means that sigma was too large.

The value of sigma reflects an implicit assumption about the `scale'
of interesting features in the pattern. If you don't know what features
you are looking for, try different scales.

Adrian Baddeley

More information about the R-sig-Geo mailing list