S. SUMMARY AND OUTLOOK
The use of ANSER for rainfall estimates is an important
tool for the enhancement of accuracy for automatic rainfall
estimation from satellite data. In this paper, the main
research efforts focused on the development of:
(1) The architecture for the ANSER system for rainfall
estimates;
(2 The parallel and nonlinear reasoning networks for
the estimation of rainfall;
(3) A 1/2 hour training algorithm for the reasoning
network to estimate rainfall;
(4) Several experimental tests of rainfall estimation
using the ANSER system.
Because the ANSER is a massive parallel processing
system, it makes estimating rainfall 10 times faster. In
these cases, after all the information was received, the
satellite-derived estimation of rainfall required about from
2 to 10 seconds of HDS 9000 CPU time to execute.
Because the ANSER is a nonlinear reasoning system, the
average error of rainfall estimates is reduced to less than
30%, the currently achievable accuracy. In this work, the
average errors of the rainfall estimates were always less
than 10.0%.
Decision trees are series process techniques and run step
by step and node by node. So decision tree techniques
require more running time than that of parallel techniques.
Decision trees are not able to use all rules, models,
knowledge and factors that are stored in the different
nodes at the same time. This is especially true when the
rules, models, knowledge and factors are very complicated
and nonlinear. So decision tree techniques always give
"rough" results.
ANN techniques have demonstrated superior performance
relative to classical methods for predicting the future
behavior of a pseudo-random time series. There are many
practical applications where, rainfall estimation and
forecasting, can be of great value. Forecasting natural
phenomena is a great area for using artificial neural
network expert system techniques.
This study only considered rainfall estimation using the
ANSER technique for some special cases. Further studies
will consider the satellite signatures that comprise the
convective rainfall estimation algorithm using ANN
techniques.
ACKNOWLEDGEMENTS
The authors thank Donald B. Miller, Frances Holt, Joanna
L. Newby, Tim Bellerby of Satellite Application
Laboratory of NESDIS/NOAA for their constructive
criticism, Juying Xie for providing satellite-derived input
data and Dick Pritchard of Planning Research Corporation
for providing satellite pictures, Paige Bridges for
propagation of the illustrations, and Lori Paschal for typing
the manuscript.
792
REFERENCES
Barrett, E.C., and D. W. Martin, 1981. The use of satellite
data in rainfall monitoring. Academic Press, New York, pp.
0-340.
Bullas, J., J.C. McLeod and B. de Lorenzis, 1990.
Knowledge Augmented Severe Storms Predictor(KASSPr)-
An operational test. In: 16th Conference on Severe Local
Storms, Boston, Amer. Meteor. Soc., pp.106-111.
Carpenter, Gail A., and Stephen Grossberg, 1989. Search
mechanisms for adaptive resonance theory (ART)
architectures. In: Proceedings of International Joint
Conference on Neural Networks, Washington D.C., Vol. 1,
pp.201-205.
Hopfield, J. J.,1982. Neural network and physical system
with emergent collective computational abilities. In:
Proceedings of National Academy of Sciences, USA,
Vol.79, pp.2554-2558.
Linsker, Ralph, 1988. Self organization in perceptual
network. Computer, 21:105-117.
Matsuoka, Tatsuo, Hiroshi Hamada, and Ryohei Nakatsu,
1989. Syllable recognition using integrated neural networks.
In: Proceedings of International Joint Conference on
Neural Networks, Washington D.C. Vol.1, pp.251-258.
Scofield, R.A., 1987. The NESDIS Operational Convective
Precipitation Estimation Technique. Monthly Weather
Review, 115(8):1773-1792.
Weaver, J.F, and R.S. Phillips, 1987. Mesoscale
Thunderstorm forecasting using RAOB data,surface
mesonet observations,and an expert system shell. In:
Symposium Mesoscale Analysis and Forecasting
Incorporating "Nowcasting." Boston, Amer. Meteor. Soc.,
pp. 327-331.
Woodley, W. L., 1972. Rainfall estimation from satellite
cloud photographs. NOAA Tech. Memo. ERL OD-11, 43
PP-
Xie, Juying, and R. A. Scofield,1989. Satellite-derived
rainfall estimates and propagation characteristics
associated with Mesoscale Convective Systems(MCSs).
NOAA Technical Memorandum NESDIS 25, 49 pp.
Zhang, Ming , and R. A. Scofield, 1992. Satellite-Derived
Estimation of Rainfall in Forward-Backward Thunderstorm
Propagation Model Using Neural Network Expert System
Techniques. In: Proceedings of 1992 IJCNN, Baltimore,
MD, USA.
Zubrick, S.M., and C.E. Riese, 1985. An Expert System to
aid in severe thunderstorm forecasting. In: 14th
Conference on Severe Local Storms, Boston, Amer.
Meteor. Soc., pp. 117-122.
pr
th
ste
cie
gio
we
reg
Sec