Copy Link
Add to Bookmark
Report
AIList Digest Volume 2 Issue 018
AIList Digest Sunday, 12 Feb 1984 Volume 2 : Issue 18
Today's Topics:
AI and Meteorology - Summary of Responses
----------------------------------------------------------------------
Date: 11 Jan 84 16:07:00-PST (Wed)
From: ihnp4!fortune!rpw3 @ Ucb-Vax
Subject: Re: AI and Weather Forecasting - (nf)
Article-I.D.: fortune.2249
As far as the desirability to use AI on the weather, it seems a bit
out of place, when there is rumoured to be a fairly straightforward
(if INCREDIBLY cpu-hungry) thermodynamic relaxation calculation that
gives very good results for 24 hr prediction. It uses as input the
various temperature, wind, and pressure readings from all of the U.S.
weather stations, including the ones cleverly hidden away aboard most
domestic DC-10's and L-1011's. Starting with those values as boundary
conditions, an iterative relaxation is done to fill in the cells of
the continental atmospheric model.
The joke is of course (no joke!), it takes 26 hrs to run on a Illiac IV
(somebody from Ames or NOAS or somewhere correct me, please). The accuracy
goes up as the cell size in the model goes down, but the runtime goes up as
the cube! So you can look out the window, wait 2 hours, and say, "Yup,
the model was right."
My cynical prediction is that either (1) by the time we develop an
AI system that does as well, the deterministic systems will have
obsoleted it, or more likely (2) by the time we get an AI model with
the same accuracy, it will take 72 hours to run a 24 hour forecast!
Rob Warnock
UUCP: {sri-unix,amd70,hpda,harpo,ihnp4,allegra}!fortune!rpw3
DDD: (415)595-8444
USPS: Fortune Systems Corp, 101 Twin Dolphins Drive, Redwood City, CA 94065
------------------------------
Date: 19 Jan 84 21:52:42-EST (Thu)
From: ucbtopaz!finnca1 @ Ucb-Vax
Subject: Re: "You cant go home again"
Article-I.D.: ucbtopaz.370
It seems to me (a phrase that is always a copout for the ill-informed;
nonetheless, I proceed) that the real payoff in expert systems for weather
forecasting would be to capture the knowledge of those pre-computer experts who,
with limited data and even fewer dollars, managed to develop their
pattern-recognition facilities to the point that they could FEEL what was
happening and forecast accordingly.
I was privileged to take some meteorology courses from such an oldster many
years ago, and it was, alas, my short-sightedness about the computer revolution
in meteorology that prevented me from capturing some of his expertise, to
buzz a word or two.
Surely not ALL of these veterans have retired yet...what a service to science
someone would perform if only this experise could be captured before it dies
off.
...ucbvax!lbl-csam!ra!daven or
whatever is on the header THIS time.
------------------------------
Date: 15 Jan 84 5:06:29-PST (Sun)
From: hplabs!zehntel!tektronix!ucbcad!ucbesvax.turner @ Ucb-Vax
Subject: Re: Re: You cant go home again - (nf)
Article-I.D.: ucbcad.1315
Re: finnca1@topaz's comments on weather forecasting
Replacing expertise with raw computer power has its shortcomings--the
"joke" of predicting the weather 24 hours from now in 26 hours of cpu
time is a case in point. Less accurate but more timely forecasts used
to be made by people with slide-rules--and where are these people now?
It wouldn't surprise me if the 20th century had its share of "lost arts".
Archaelogists still dig up things that we don't know quite how to make,
and the technological historians of the next century might well be faced
with the same sorts of puzzles when reading about how people got by
without computers.
Michael Turner (ucbvax!ucbesvax.turner)
------------------------------
Date: Wed 8 Feb 84 15:29:01-PST
From: Ken Laws <Laws@SRI-AI.ARPA>
Subject: Summary of Responses
The following is a summary of the responses to my AIList request for
information on AI and meteorology, spatial and temporal reasoning, and
related matters. I have tried to summarize the net messages accurately,
but I may have made some unwarranted inferences about affiliations,
gender, or other matters that were not explicit in the messages.
The citations below should certainly not be considered comprehensive,
either for the scientific literature as a whole or for the AI literature.
There has been relevant work in pattern recognition and image understanding
(e.g., the work at SRI on tracking clouds in satellite images), mapping,
database systems, etc. I have not had time to scan even my own collection
of literature (PRIP, CVPR, PR, PAMI, IJCAI, AAAI, etc.) for relevant
articles, and I have not sought out bibliographies or done online searches
in the traditional meteorological literature. Still, I hope these
comments will be of use.
------------------
Bob Giansiracusa (Dept of Computer Science, Penn State Univ, 814-865-9507)
reports that he and Alistair Frazer (Penn State Meteo Dept.) are advising
two meteorology/CS students who want to do senior/masters theses in AI.
They have submitted a proposal and expect to hear from NSF in a few months.
Capt. Roslyn (Roz) J. Taylor, Applied AI Project Officer, USAF, @RADC,
has read two of the Gaffney/Racer papers entitled "A Learning Interpretive
Decision Algorithm for Severe Storm Forecasting." She found the algorithm
to be a "fuzzy math"-based fine-tuning algorithm in much the same spirit
as a Kalman filter. The algorithm might be useful as the numerical
predictor in an expert system.
Jay Glicksman of the Texas Instruments Computer Science Lab suggests
that we check out
Kawaguchi, E. et al. (1979)
An Understanding System of Natural Language and Pictorial Pattern in
the World of Weather Reports
IJCAI-6 Tokyo, pp. 469-474
It does not provide many details and he has not seen a follow up, but
the paper may give some leads. This paper is evidently related to the
Taniguchi et al. paper in the 6th Pat. Rec. proceedings that I mentioned
in my query.
Dr. John Tsotsos and his students at the Univ. of Toronto Laboratory for
Computational Medicine have been working for several years on the ALVEN
system to interpret heart images in X-ray films. Dr. Tsotsos feels that the
spatial and temporal reasoning capabilities of the system would be of use in
meteorology. The temporal reasoning includes intervals, points,
hierarchies, and temporal sampling considerations. He has sent me the
following reports:
R. Gershon, Y. Ali, and M. Jenkin, An Explanation System for Frame-based
Knowledge Organized Along Multiple Dimensions, LCM-TR83-2, Dec. 1983.
J.K. Tsotsos, Knowledge Organization: Its Role in Representation,
Decision-making and Explanation Schemes for Expert Systems, LCM-TR83-3,
Dec. 1983.
J.K. Tsotsos, Representational Axes and Temporal Cooperative Processes,
Preliminary Draft.
I regret that I have found time for only a cursory examination of these papers,
and so cannot say whether they will be useful in themselves for meteorology
or only as a source of further references in spatial and temporal reasoning.
Someone else in my group is now taking a look at them. Others papers from
Dr. Tsotsos group may be found in: IJACI77-79-81, PRIP81, ICPR82, PAMI Nov.80,
and IEEE Computer Oct. 83.
Stuart C. Shapiro at the Univ. of Buffalo (SUNY) CS Dept. added the
following reference on temporal reasoning:
Almeida, M. J., and Shapiro, S. C., Reasoning about the temporal
structure of narrative texts. Proceedings of the Fifth Annual Meeting
of the Cognitive Science Society, Rochester, NY, 1983.
Fanya S. Montalvo at MIT echoed my interest in
* knowledge representations for spatial/temporal reasoning;
* inference methods for estimating meteorological variables
from (spatially and temporally) sparse data;
* methods of interfacing symbolic knowledge and heuristic
reasoning with numerical simulation models;
* a bibliography or guide to relevant literature.
She reports that good research along these lines is very scarce, but
suggests the following:
As far as interfacing symbolic knowlege with heuristic reasoning with
numerical simulation, Weyhrauch's FOL system is the best formalism I've
seen/worked-with to do that. Unfortunately there are few references to it.
One is Filman, Lamping, & Montalvo in IJCAI'83. Unfortunately it was too
short. There's a reference to Weyhrauch's Prolegomena paper in there. Also
there is Wood's, Greenfeld's, and Zdybel's work at BBN with KLONE and a ship
location database; they're no longer there. There's also Mark Friedell's
Thesis from Case Western Reserve; see his SIGGRAPH'83 article, also
references to Greenfeld & Yonke there. Oh, yes, there's also Reid Simmons,
here at MIT, on a system connecting diagrams in geologic histories with
symbolic descriptions, AAAI'83. The work is really in bits and pieces and
hasn't really been put together as a whole working formalism yet. The
issues are hard.
Jim Hendler at Brown reports that Drew McDermott has recently written
several papers about temporal and spatial reasoning. The best one on
temporal reasoning was published in Cognitive Science about a year ago.
Also, one of Drew's students at Yale recently did a thesis on spatial
reasoning.
David M. Axler, MSCF Applications Manager at Univ. of Pennsylvania, suggests:
A great deal of info about weather already exists in a densely-encoded form,
namely proverbs and traditional maxims. Is there a way that this system can
be converted to an expert system, if for no other reason than potential
comparison between the analysis it provides with that gained from more
formal meteorological approaches?
If this is of interest, I can provide leads to collections of weather lore,
proverbs, and the like. If you're actually based at SRI, you're near
several of the major folklore libraries and should have relatively easy
access (California is the only state in the union with two grad programs in
the field, one at Berkeley (under the anthro dept.), and one at UCLA) to the
material, as both schools have decent collections.
I replied:
The use of folklore maxims is a good idea, and one fairly easy to build
into an expert system for prediction of weather at a single site. (The
user would have to enter observations such as "red sky at night" since
pattern recognition couldn't be used. Given that, I suspect that a
Prospector-style inference net could be built that would simultaneously
evaluate hypotheses of "rain", "fog", etc., for multiple time windows.)
Construction of the system and evaluation of the individual rules would
make an excellent thesis project.
Unfortunately, I doubt that the National Weather Service or other such
organization would be interested in having SRI build such a "toy"
system. They would be more interested in methods for tracking storm
fronts and either automating or improving on the map products they
currently produce.
As a compromise, one project we have been considering is to automate
a book of weather forecasting rules for professional forecasters.
Such rule books do exist, but the pressures of daily forecasting are
such that the books are rarely consulted. Perhaps some pattern
recognition combined with some man-machine dialog could trigger the
expert system rules that would remind the user of relevant passages.
Dave liked the project, and suggested that there may be additional unofficial
rule sources such as those used by the Farmer's Almanac publishers.
Philip Kahn at UCLA is interested in pattern recognition, and recommends
the book
REMOTE SENSING: Optics and Optical Systems by Philip N. Slater
Addison-Wesley Publ. Co., Reading, MA, 1980
for information on atmospherics, optics, films, testing/reliability, etc.
Alex Pang at UCLA is doing some non-AI image processing to aid weather
prediction. He is interested in hearing about AI and meteorology.
Bill Havens at the University of British Columbia expressed interest,
particularly in methods that could be implemented on a personal computer.
Mike Uschold at Edinburgh and Noel Kropf at Columbia University (Seismology
Lab?) have also expressed interest.
------------------
My thanks to all who replied.
-- Ken Laws
------------------------------
End of AIList Digest
********************
