187 
CROP YIELD FORECASTING--STATE OF THE ART 
Galen F. Hart 
Agricultural Research Service 
U.S. Department of Agriculture 
Beltsville, Maryland 
ABSTRACT 
Forecasting grain crop yield or 
production has been routinely done by 
government and agri-business for 
decades, particularly in the United 
States. The United States commodity 
markets are quite sensitive to changes 
in yield and production prospects as 
reflected in sometimes dramatic shifts 
in future grain prices. Methods of 
forecasting vary from expert opinion to 
detailed mathematical models of plant 
growth processes. This full range of 
sophistication and approach is found to 
have value depending on the decisions 
to be made. Three broad categories of 
yield forecast modelling are 
described. The traditional method uses 
expert opinion, trends and comparative 
data to forecast the future. A second 
category uses plant measurements. 
Samples are located in selected 
producer fields and measurements, both 
destructive and non-destruetive, are 
made as the season progresses 
concluding with crop cutting just prior 
to producer harvest. A five year 
history of these relationships is 
regressed to create models for using 
current season measurements to project 
end - of-season yield. Lastly, there are 
models that attempt to mathematically 
represent the plant growth process 
including grain development. Extensive 
and sometimes complex mathematical 
structures are used in this modelling. 
Initial conditions, environmental 
information, current data (e.g. 
temperature and moisture) and state 
variables are combined to simulate 
daily plant growth. Then, by assuming 
a set of future conditions, these 
models are used to forecast yield. 
Refinements to improve performance are 
being made in all the yield modeling 
categories. Examples are given in the 
operational use of models. An 
intriguing area of research is the 
incorporation of spectral data in 
models and a review is provided of the 
current status of these efforts. 
KEY WORDS: Vegetation Indices, 
Estimation, Production 
Variability 
INTRODUCTION 
Over the past decade, there have been 
no break throughs in the ability to 
forecast crop grain yield either at the 
field level or for large areas. 
Clearly, however, progress has been 
made in understanding what causes 
change in yield potential. Operational 
large area yield forecasting procedures 
have changed little except for 
refinement. However, some notable 
events have occurred. The demand for 
more accurate forecasts and estimates 
of grain production worldwide has 
increased both from the economic and 
humanitarian points of view. Trade 
balance and human rights are matters of 
concern. 
Inexpensive personal computing has 
allowed more complex modelling to 
become a tool for individual decision 
making rather than a power reserved 
only for well financed agri-business. 
This paper represents what is believed 
to be state-of-the-art in the United 
States. Since most operational and 
proprietary systems are not documented 
in the literature, this paper cannot 
fully represent the breadth of the 
subj ect. 
OPERATIONAL STATE OF THE ART 
The National Agricultural Statistical 
Service, U.S. Department of Agriculture 
- Domestic Forecasting and Estimating 
Program 
The yield forecasting and estimating 
Program of the National Agricultural 
Statistics Service (NASS) is by far the 
largest program of its kind (USDA, 
1983). It was developed primarily to 
replace or supplement a very large 
opinion survey system that had been 
used by the Service for decades. 
Monthly growing season forecasts by 
this organization are used 
inconjunction with other information by 
the commodity trade business to 
establish future prices for grains. An 
objective quantitative approach would 
provide estimates and forecasts of 
known precision. Such a forecasting 
and estimating program started in the 
early 1960's and now includes corn, 
cotton, potatoes, rice, soybeans and 
three categories of wheat (winter, 
spring and durum). Twenty-seven states 
are included in the program and over 
9,000 crop samples are measured. The 
information collected is objective 
because trained field personnel 
randomly locate, in farmers' fields, 
units to be sampled. Field personnel 
then count and measure attributes of 
the growing crop. No survey opinions 
are used. 
In the states involved in the surveys, 
the total land area in a crop has an 
opportunity to be selected in the 
sample for measurement. This is