Re: r.mdig.kernel
Joel Pitt
> Also just wondering if there is some specific documentation for usage of
> r.mdig.kernel? I tried looking at the man and -h, but they aren't terribly
> specific - I was thinking about starting a page on the wiki, but I thought
> I'd check with you -
Go for it - I didn't go into much detail with the documentation
because I changed this module a lot during my research. You are right
that the documentation should be much better. Please do create a wiki
page :-)
> for example, I fit a neg exp model to some dispersal data for H. lepidulum -
> so I have some parameter estimates, but do they need to be normalized into a
> pdf, or even a cdf form?
Yes, the parameter estimates should be normalised. The in-built
negative exponential has a PDF form of:
k(x) = 1/B* exp^(-x/B)
where B is the scale parameter. I am reasonably sure that this form is
already normalised... and as such, the parameters are the same for the
CDF.
> Would the frequency argument be the proportion of individuals in a cell to
> be dispersed?
The frequency argument is the mean of the Poisson distribution
generating dispersal events. It is currently independent of the number
of individuals, but could potentially be shaped by population/density.
The problem is working out how exactly this should work...
> Is the -z flag just to cut down on processing time?
> Sorry to bombard you...
The -z flag increases processing time, but allows you to delete the
null bitmask of the maps and have the module ignore processing cells
with a zero value (which is different from having a NULL value). This
is a GRASS specific thing I added to reduce the size of the
distribution maps output by MDiG. You shouldn't need to worry about
it.
J