Newsgroups: sci.environment,comp.ai.neural-nets
Subject: Re: Neural Nets for Climate Change Prediction?
Message-ID: <ulall-0307962301170001@asy06.xy1.usu.edu>
From: ulall@kernel.uwrl.usu.edu (Upmanu Lall)
Date: 3 Jul 96 23:01:19 MDT
References: <31DA5682.5369@webdev.co.uk> <31daaa64.3854044@news>
Organization: Utah State University
Nntp-Posting-Host: asy06.xy1.usu.edu
Lines: 70
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!newsgate.duke.edu!news.mathworks.com!newsfeed.internetmci.com!news.internetMCI.com!pull-feed.internetmci.com!news.provo.novell.com!news.cs.utah.edu!cc.usu.edu!asy06.xy1.usu.edu!user
Xref: glinda.oz.cs.cmu.edu sci.environment:108085 comp.ai.neural-nets:32364

I believe that there is a paper by Latif, where they use a Neural net to
predict the Indian Monsoon
I can dig it up if you would like. There is also a book by Weigend that
may have some interesting applications.

 Regarding adaptive (i.e. nonparametric methods) you may find the
following reference of interest:
 Lall, Upmanu, Sangoyomi, Taiye and Abarbanel, Henry D.I., Nonlinear
dynamics of the Great Salt Lake: Nonparametric short-term forecasting.,
Water resources research., APR 01 1996 v 32 n 4, Page:   975   -

A multivariate adaptive regression spline is used to estimate the markov
transition function or in nonlinear dynamics terminology - a map to
predict the Great Salt Lake for up to 4 years in the future and various
chaotic and other models for various periods of time.

In general transitions from one regime to another, that have occurred
infrequently in the past or occur with very large derivatives can't be
approximated. However, once the transition has occurred predictability
returns.
I'll be happy to discuss this issue further.



In article <31daaa64.3854044@news>, steve@tropheus.demon.co.uk (Stephen
Wolstenholme) wrote:

> On Wed, 03 Jul 1996 11:16:18 +0000, Steve Crossan
> <stevec@webdev.co.uk> wrote:
> 
> >Hullo all,
> >
> >Is anyone aware of any work in the use of Neural Networks and other 
> >adaptive techniques for climate change prediction? 
> >
> Steve,
> I did a little work in short term prediction of local weather
> condition using a neural network. After training the net on a years
> worth of measures (four days at a time, pressure, rainfall,
> temperature etc) the net could predict the following days measures
> quite accurately. However it never predicted sudden changes. It was
> pointed out to me a number of times that if you predict tomorrows
> weather to be the same as todays then you will be correct far more
> often than incorrect. The neural net was no more accurate than that
> method. I've never looked into climate (as in long term, global
> weather). Is the data available?
> 
> 
> Steve
> 
> --------------------------------------------------------------------------
> Stephen Wolstenholme, Cheadle Hulme, Cheshire, UK   
> steve@tropheus.demon.co.uk
> 
> Author of Neural Planner, Windows neural network system
> 
> ftp.demon.co.uk/pub/ibmpc/win3/apps/ai/np410.zip
> http://www.winsite.com/pc/win3/programr/np410.zip
> http://www.simtel.net/pub/simtelnet/win3/neural/np410.zip
> ftp://ftp.simtel.net/pub/simtelnet/win3/neural/np410.zip
> --------------------------------------------------------------------------

-- 
Upmanu Lall
Professor
Utah Water Research Laboratory  & Civil Env. Eng.
Utah State University 
Logan UT 84322-8200
(801)-797-3184
FAX: 801-797-3663
