Even though these LACIE yield esti-
mates were significantly below those of
the SRS, a test using 10 years of historic
data indicates that such differences are
to be expected occasionally. Table I
lists the results of a comparison of the
Phase III yield models with historic data
for the years 1967 to 1976. The models
were developed with data for the 45 years
prior to each of the test years. It can
be shown that the models were supportive
of the 90/90 accuracy goal. The model
errors exceeded the 90/90 tolerance
bounds only in the years 1971 and 1974.
In the Phase III crop year, the models
tended to perform about the same as in
1971 and significantly better than in
1974. An analysis of yield model be-
havior indicates that they perform
adequately if no unpredicted signifi-
cant changes in trend occur and if the
average weather conditions for a region
are not drastically different from the
historic data used in their development.
Where such weather departures occur, the
models tend to respond in the right
direction but not to a sufficient extent.
However, as could be seen in the Phase
III U.S.S.R. spring wheat regions, these
models did perform adequately in a de-
parture from normal which, while not
extreme, was of great economic importance
to the U.S. and other countries.
TABLE I.— 10-YEAR BOOTSTRAP TEST FOR U.S.
PHASE III YIELD MODELS WITH CONTINUED
TREND AGGREGATED TO THE USGP BY YEAR
WITH 90/90 CRITERION TEST.
Year AS Cig sims Error Joue
1967 | 21.6 225 +0.9 Yes
1968 | 26.0 24.6 -1.4 Yes
1969 | 28.4 29.4 +1.0 Yes
1970 | 28.2 26.6 -1.6 Yes
1971 | 30.8 27.9 -2.9 No
1972 | 29.3 29.1 20.2 Yes
1973 | 30.8 30.6 -0.2 Yes
1974 | 23.8 28.4 +4.6 No
1975 | 26.8 27.3 *0.5 Yes
1976 | 26.4 27.1 +0.7 Yes
19773] 27.5 24.9 -2.6
Mean error = -0.1 bu/acre
R.M.S.E. = 1.90 bu/acre
Accept 90/90
For comparison only, LACIE 1977 estimates
The Phase III results for production,
area, and yield in the "yardstick" hard
red winter wheat region with its larger
fields generally supported the results
achieved in the U.S.S.R. The coeffi-
cients of variation are well within
tolerance, and only a modest tendency
to underestimate can be detected. These
results are somewhat better than the good
results achieved in Phase II for the
region.
The results in the strip fallow (smal
fields) area of the hard red spring wheat
region of the U.S. showed significant
improvement but still exhibited a ten-
dency toward underestimation of spring
small grains. Figure 20 displays the
experimental estimates as compared to the
SRS report for the region. Econometric
ratio models developed in Phase II and
used to estimate the spring wheat from
the Landsat-derived spring grain esti-
mates worked well for the region. Also,
later results are indicating that spring
wheat can be differentiated from spring
barley directly from Landsat images
AREA
Millions of acres
ze 7,23 723, 16.97
14651602 1512 1552 1564
ceases ESCS
— | ACF
DEC | JAN | FEB | MAR APR [MAY [JUN T JUL | AUG | SEP J OCT | NOV | DFT
YIELD
a0
so f- 27.0 277 269 267 271
— M—————
E 2487234 236 234 234
Bushels/acre
~
0 ares ESCS
LACIE
DEC | JAN | FEB [MAR | APR [MAY] JUN ] JUL TAuGT see T oct [ NOV J OFC
PRODUCTION
1000
750 -
500 |= 465.8 4774, 4637 4613 M04
Millions of bushels
= 7374 366.4
zo en ESCS 363 5 356.7 3629
LACIE
DEC | JAN | FEB [MAR] APR | MAY | JUN | JUL | AUG] SEP J OCT Trov [5€]
LACIE — real-time reporting
4—state total spring wheat (includes durum wheat)
Minnesota, Montana, North Dakota, and South Dakota
Figure 20.— Phase III U.S. spring
wheat estimates.
acquir
wheat.
1% und
age ex
undera
differ
gions
U.S.S.
the ya
what w
in Pha
are op
sidera
the re
confus
diffic
F
wheat
truth |
Landsa
indica
in the
Phase
suppor:
gated |
alread
time ri
ment, i
compute
proces:
LACIE wheat percent
