ANSER -- 
ARTIFICIAL NEURAL NETWORK EXPERT SYSTEM 
FOR SATELLITE-DERIVED ESTIMATION OF 
RAINFALL 
Ming Zhang* and Roderick A. Scofield 
NOAA/NESDIS 
Satellite Application Laboratory 
5200 Auth Road, Room 601 
Washington, D.C. 20233 
ABSTRACT: 
This research presents an artificial neural network expert 
system technique for rainfall estimation from satellite 
data. An ANSER-Artificial Neural network expert system 
for Satellite-derived Estimation of Rainfall is being 
developed in the NOAA/NESDIS/ SAL. Using the 
ANSER technique, estimation or computation of rainfall 
amounts will be 10 times faster. The average error of 
rainfall estimates for the total precipitation event will be 
reduced to less than 30%, the currently achievable 
accuracy. This research work will be a big step toward 
creating an rainfall estimation expert system using an 
artificial neural network from the remotely-sensed data. 
KEY WORDS: 
Artificial Neural Network, Expert System, Rainfall, 
Satellite-derived, Estimation 
* This work was done while the author held a USA 
National Research Council Postdoctoral Research 
Associateship at the National Oceanic and Atmospheric 
Administration (NOAA), National Environmental 
Satellite, Data, and Information Service (NESDIS), 
Satellite Application Laboratory. 
1. INTRODUCTION 
Assessment of global climate change is a very important 
research area for the future of man and his environment. 
Rainfall estimation is a key parameter in this research. 
During the past 20 years, there has been a great increase 
in our understanding of how satellite data can be used to 
estimate rainfall. But, even with the use of interactive 
computer systems, the time needed to prepare estimates 
of rainfall is about a half hour. Verification results show 
that the average error for an event is about 30%. 
Some Artificial Intelligence (AI) Systems for weather 
forecasting have been designed to be objective and 
automated (Zhang and Scofield, 1992); others are 
designed to augment human skill. In the Knowledge 
Augmented Severe Storms Predictor (KASSPr) system 
(Bullas, 1990), knowledge was elicited in a series of 
interviews and exchanges of documentation between the 
developer and an expert in severe weather forecasting. 
The Convex system (Weaver, 1987) first uses as automated 
analysis of the Denver morning sounding, combined with 
estimates of expected afternoon temperature and 
dewpoint, to determine the relative instability of the host 
785 
air mass and its likelihood of initiating convection later in 
the day over the region of interest. The knowledge base 
for the Willard system (Zubrick, 1985) is a structured 
hierarchy of 30 rules. Most of the rules were developed 
using the inductive generalization feature of Rule Master, 
an expert system shell. Several investigators showed that 
estimating rainfall from both geosynchronous and polar- 
orbiting satellite was feasible (Woodley, 1972). A 
complete review of rainfall schemes that use visible, 
infrared, or microwave satellite data is presented by 
Barrett and Martin (1981). Most of the important facts 
about rain clouds have been extracted for use in the 
present estimation scheme. Currently, satellite-derived 
precipitation estimates (Scofield, 1987) and 3-hour 
precipitation outlooks for convective systems, Extratropical 
cyclones, and tropical cyclones are computed on the 
NOAA/NESDIS Interactive Flash Flood Analyzer (IFFA) 
system and transmitted to National Weather Service 
Forecast Offices, and River Forecast Centers. However, 
this system permits the computation of rainfall estimates 
for only one convective system at a time. This is due to 
the considerable time needed for image processing, 
interpretation, and the computation involved in the 
estimation of rainfall If there are several storms 
occurring, an automatic estimation technique would be 
useful in providing rainfall estimates for the entire 
country. Digital satellite data is used in the estimation 
process. Artificial neural network (ANN) techniques are 
explored as a possible improvement to current techniques. 
This research applies ANN techniques to the 
enhancement of knowledge for automatic rainfall 
estimation from satellite data. 
ANN computing is an area that is receiving increased 
research interest. Since ANNs are massively parallel 
systems, ANN computers have tremendous speed and 
nonlinear advantages over traditional digital machines. 
Hopfield developed the first architecture of a ANN, the 
Hopfield network, (Hopfield, 1982). Carpenter (1989) had 
researched adaptive resonance theory (ART) 
architectures. A conventional ANN architecture is the 
Back Propagation (BP) ANN. Matsuoka (1989) 
introduced a new training model, the Integrated Neural 
Network (INN). INN can reduce training time for syllable 
signal processing.  Linsker (1988) described a Self- 
Organized architecture in a perception network; this 
model can recognize special features of its environment, 
without being told which feature it should analyze. 
In a study by Xie and Scofield (1989), where the 
Scofield/Oliver Technique was used, there were significant 
differences between the rainfall observations and the 
satellite-derived rainfall estimates. In this paper, ANN 
technique will be used for estimates of rainfall. The main 
research efforts developed in this paper are as follows: 
(1) The architecture of the ANSER system for estimates 
of rainfall; 
(2) The parallel and nonlinear reasoning networks for 
estimation of rainfall; 
(3) 1/2 hour training algorithm of a reasoning network for 
estimates of rainfall; 
(4) Several experimental results of estimating rainfall 
using the ANSER system.