[GRASSLIST:4267] Re: Releasing GRASS modules

[Benjamin Ducke]

Sure, here are some details:

The modules are a by-product of my work in an archaeological
predictive modelling project in North-east Germany.
Over a period of four years, I tried different approaches,
like regression statistics, geostatistics, explorative statistics,
neural networks classification and probability theory.

I finally settled for Dempster-Shafer Theory (DST) as a mathematical
framework. DST is a very flexible sort of generalised probabilit
theory.
With DST, the user first sets up a frame of hypotheses to check.
Next, he/she has to supply evidences that support one or several
of these hypotheses. Evidence is quantified using a basic probability
assignment (BPA). It is up to the user to define just how exactly
this has to be done.
Finally, all evidences are combined and a final "belief" value
is calculated for each hypothesis. Other useful metrics can be
calculated as well.
In theory, DST can be used for much more than just predictive modelling
and so can my modules.
But the tools are most straight-forward to use in predictive models,
were BPA calculation is easy and the frame of reference consists
only of two hypotheses ('SITE PRESENT' and 'NO SITE PRESENT').

DST also handles uncertainty, represented by the hypotheses set
{'SITE PRESENT', 'NO SITE PRESENT')} were you cannot decide between
the two.
It makes no assumptions about probability distributions and is
memory efficient (ran it with several thousand sites and
many millions of raster cells on a machine with 256MB RAM).

The tools also include modules for random sampling and 
automatic generation of raster categories given interval width
or number (sth. that was asked for in the last GRASS newsletter,
if I remember right).

If you want to know more about belief-based modelling, read:

http://gis.esri.com/library/userconf/proc99/proceed/papers/pap295/p295.htm

and

http://websrv5.sdu.dk/ejstrud/forskning.html

On Fri, 27 Aug 2004 15:46:15 +0200
Martin Wegmann <wegmann at biozentrum.uni-wuerzburg.de> wrote:

> Hello Benjamin, 
> 
> could you explain briefly which methods you are using for your modules 
> (statistical methods like GAM, ANN)? Is it comparable to GRASP 
> (http://www.cscf.ch/grasp/grasp-r/index.html) - unfortunately not fully 
> implemented in GRASS (no easy export of raster to GRASP)  - it sounds very 
> promising and I am looking forward to give it a try. 
> 
> cheers Martin
> 
> 
> [....]
> > I guess so. In the meantime I might set up a homepage of my own to
> > distribute my code. In the meantim, if anyone here is really interested in
> > it, contact me and I will send it to you via email.
> >
> > Benjamin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: grassDST53-1.0.tar.gz
Type: application/x-gzip
Size: 119078 bytes
Desc: not available
Url : http://lists.osgeo.org/pipermail/grass-user/attachments/20040827/64c2b97d/grassDST53-1.0.tar.gz