yield data to determine what 
statistical relationship exists 
(Boatwright, 1988). A collection of 
AVHRR data was obtained over the 
midwest during the period 1983 through 
1987. The objective was to create a 
model based on the three crop years, 
1983 through 1985, then to estimate 
yield for 1987. In order to do this, 
data from several of the NOAA-n series 
satellites had to be combined. 
The data were first processed through 
the USDA, FAS, Metsat processor. The 
software has functions that screen for 
clouds, water and bare soil. A sun 
angle correction was applied to better 
equate data from different satellites. 
A vegetation index number (VIN) was 
calculated as a difference between 
channel 1 and 2. A further constraint 
was to have at least five satellite 
acquisitions during the growing period 
over the area of interest. AVHRR data 
from NOAA-6 through 10 satellites were 
used. An additive model adjusted for 
differences between satellite sensors. 
VIN values were interactively compared 
from different satellites for the same 
location within a two day period. 
After the appropriate data sets were 
obtained, coefficients of the model 
were computed to equate all values to 
those obtained from NOAA-6. 
A number of problems exist in this type 
of adjustment. There can be 
substantial atmospheric differences 
between consecutive days and even 
though near nadir passes were used, 
there are potentially some angular 
affect differences. Also, by using a 
simple channel difference the VIN 
values are contaminated by atmospheric 
scattered radiation. This effect is a 
function of the solar zenith angle and 
increases as the solar zenith angle 
increases. Data analysis from various 
morning and afternoon satellites, shows 
that the effect is larger for the 
morning satellites than afternoon. A 
better procedure to estimate a 
vegetation index would be to use a 
normalized difference rather than a 
simple difference. A normalized 
difference is rather insensitive to sun 
angle and atmospheric affects and also 
leads to a more stable "soil line" than 
simple difference values. In general 
better calibration data are needed from 
the satellite systems, however the 
process moved forward and data sets 
were created. 
For corn and soybean areas in the U.S. 
midwest, a 120 day crop season was 
selected as the appropriate period. 
This period was divided into six day 
intervals thus creating twenty 
intervals for the season. A curve fit 
procedure was employed to fit a VIN 
trajectory using a single component 
exponential model. After the best fit 
was obtained, the area under the curve 
of each of the twenty intervals is 
calculated. A linear function of 
values from the twenty areas is used to 
estimate yield. In general it was 
found that four interval areas are 
adequate to capture most of the 
correlative information of the system. 
Three of the values came from the time 
period up to peak greeness and one 
after. Also, there seems to be some 
forecast capability in this simple 
model at about sixty days into the 
growing season. 
Relationships from the single study 
using three years of data for model 
building and testing, and one for 
validation, gave a range of 
approximately +20 bushels per acre from 
official published data for corn using 
the full season model. Work continues 
to improve this general approach. 
Developing the trajectory using a 
normalized approach should improve 
performance. There is potential for 
using something as simple as this as an 
indicator approach for yield or 
production potential. 
Recent research (Wiegand and 
Richardson, 1990) extending earlier 
work, suggests a firmer scientific base 
for this approach. They specify a 
"spectral component analysis" approach 
and calculate évapotranspiration and 
provide an equation for yield 
estimation. The basic assumptions of 
spectral component analysis are that 
the crop canopy display the net 
assimiliate achieved in response to 
growing conditions that have occurred 
and that vegetation indices are a 
measure of the amount of 
photosynthetically active and 
transpiring tissue present. 
THE BLEND -- SPECTRAL AND PROCESS LEVEL 
MODELING 
Recently, (Maas, 1988 a and b) an 
effort has been undertaken to use 
satellite remotely sensed data to 
compliment the performance of crop 
growth models. One such effort for 
grain sorghum utilizes process based 
functional relationships in a very 
simplified form. The model operates 
with only three major "state" variables 
-- green (living) leaf area index, 
above ground drymass and stage of 
development. In this approach, field 
level remotely sensed data derived from 
Landsat multispectral inputs are used 
to calculate either an initial value of 
green leaf area index (GLAI) or to 
calibrate the trajectory of GLAI if the