I don't know if this is specific enough to be helpful, but in addition to
conventional Bayes nets, this actually sounds very similar to the spatial
reasoning that was done in machine vision work by Geman and Geman. Their
paper on this, which is a classic, can be found in Shafer and Pearl's book
Readings in Uncertain Reasoning, published by Morgan Kaufmann. Here's a
citation:
@ARTICLE{Geman:84,
AUTHOR = {Stuart Geman and Donald Geman},
TITLE = {Stochastic relaxation, Gibbs distributions and the Bayesian
restoration of images},
JOURNAL = PAMI,
YEAR = {1984},
VOLUME = {6},
note = {Republished in \cite{UNCReadings}},
PAGES = {721-741}
}
PAMI is the IEEE Proceedings on Pattern Analysis and Machine Intelligence.
A great deal of vision work like this was done at the Brown University
Department of Applied Mathematics by Geman, Ulf Grenander and others. The
central theme was to take a grid of pixels and treat it as a Markov Field,
and then find interpretations as optimal labelings. Higher level processes
could also be represented (e.g. edges) as influences on the pixel random
variables. The geographical analogy should be clear.
I'm afraid I haven't tracked this work, so I can't provide more up-to-date
citations.
Best,
R
- -----Original Message-----
From: owner-uai@cs.orst.edu [mailto:owner-uai@cs.orst.edu]On Behalf Of
Scott Wooldridge
Sent: Monday, August 27, 2001 8:54 PM
To: uai@cs.orst.edu
Subject: [UAI] BBNs for GIS models
Hi,
I am looking to undertake a spatial risk assessment for the Great =
Barrier Reef Marine Park, which will take into considertaion the =
combined influences of cyclones, bleaching (elevated SST), flood plumes, =
and crown-of-thorn starfish (COTS). For example I would like to be able =
to provide a solution to the question, "What is the likelihood that a =
given coral reef will receive damage (of a certain degree) within a =
given time-frame".
Sufficient spatial predictive modelling has been done for all quantities =
(perhaps not the COTS), for me to assign the likelihood of occurrence =
for each hazard for different return periods. Obviously there is some =
correlation in the occurence of the different hazards (e.g. =
cyclone/flood plumes). The data sources can also be considered to be of =
different spatial resolution and accuracy.=20
I like the thought of being able to frame the combined risk assessment =
in terms of a spatial Bayesian Belief Network.=20
Is anyone aware of any software/code that has been written which could =
adequately address the spatial nature of this problem (i.e. BBN within a =
raster/GIS based structure)?
I have no practical experience with the implementation of BBNs and would =
enjoy hearing people's comments on the suitability of BBNs to the above =
problem.
Any better suggestions?
Many Thanks,
scott
- ----------------------------------------------------------------------------
Dr Scott Wooldridge
Spatial Analyst, Reef Futures Group
Australian Institute of Marine Science
PMB No. 3, Townsville MC Qld 4810
Telephone: (07) 47534334
Facsimile: (07) 47725852=20
E-mail: s.wooldridge@aims.gov.au
"the future belongs to those who see the possibilities before they =
become obvious" Theodore Leavitt
------- End of Forwarded Message
This archive was generated by hypermail 2b29 : Wed Aug 29 2001 - 09:45:38 PDT