Derek,

I have some experience using Spot and Landsat data
combined with DSSAT to project yields in maize.

My most accurate attempt was a yield estimate of 148.3
bu/acre on approximately 92,000 acres within a county
in Indiana using Spot and DSSAT.  The estimate was
based on one image acquired on Aug 15.  My estimate
would compare with 144.8 bu/acre on approximately
94,000 acres that National Ag Statistics published the
following March.  NAS was unable to give me a margin
of error for their data.

The tools I used were:
IDRISI GIS
DSSAT

Data sets that I used were:
A digital elevation model of the county
One Spot multispectral image
A digital soil map of the county
Detailed weather data as required by DSSAT
(I acquired 30 years of data and had data current thru
the date the image was acquired)

I used the genetic coefficient for Pioneer 3383 which
held a predominate market share (80%) that year in the
mid west.

Purdue University has a test farm there and I used it
extensively for ground truthing and obtaining
beginning soil water and nutrient levels.

I obtained plant dates for other areas from local crop
insurance agents and the extension service combined
with observation of rain events to give me the most
probable range of plant dates in a given area of the
county.

Seeding rates were obtained by polling seed dealers
throughout the county and from the folks at the Purdue
test farm.  I deducted 10% from what they told me to
arrive at a harvested population and used an estimate
of 1.2 bu/acre for harvest loss.

I used DSSAT to give me a range of yield values with
known weather, soils, seeding rates and plant dates.
I did not change any values within DSSAT.

I did a supervised classification of the Spot image
and matched reflectance values with corresponding
DSSAT values on the Purdue test farm where we had
detailed ground truthing data to include estimated
kernel counts and weight.  As it happened, the image
was taken near PM.  This was just lucky, that was the
only cloud free image we were able to get that summer.

I deviated the DSSAT values based on the deviation of
the reflectance values throughout the county and then
calculated a weighted average yield.

There wasn't any noticeable stress associated with
weeds or disease as I toured the county.

A sizeable portion of the county experienced a severe
hail storm... it was severe enough to kill some of the
fields entirely.  The pattern and intensity of the
storm was readily apparent on the multispectral image.
 The storm definitely gave me a worst case scenario
from which I was able to graduate to best case at the
Purdue test farm.  Luck intervened again as the test
farm was not affected by the storm.

I found then as I have since that there is a high
percentage of yield variation due to slope and
compaction which inhibits water infiltration.  This
can be measured with remote sensing quite well if you
have the underlying data.

I was pleased with my results for field corn, however,
my objective was to forecast the yield of inbreds for
Pioneer on their seed fields within the county.  I was
unable to get coefficients for them so I could not use
a model.  There was not any correlation between
reflectance values alone and the actual yields.  Maybe
it would have been possible using DSSAT with known
genetic coefficients or the model the Joe Ritchie sent
me that was designed for inbreds, however, Pioneer
understandably deemed the coefficients as a trade
secret.

I am still pursuing a reliable solution using these
techniques for row crop farmers.  My thanks to Gordon
Tsuji, Jim Jones and  Joe Ritchie for their help and
guidance during the experiment.  I conducted this
experiment in 1994.

Good luck in your quest.


Sincerely

Doug Miller






=====
Doug Miller
609 Miami
Hiawatha, KS 66434
(785) 742-2501
[log in to unmask]
[log in to unmask]

__________________________________________________
Do You Yahoo!?
Try FREE Yahoo! Mail - the world's greatest free email!
http://mail.yahoo.com/