asing
s. An
crops
Is not
erent
r the
ellite
Lal,
neral
plied
voids
vhich
Iracy
ined.
f the
"URI
, and
inear
.6%
yield
nces
apart
5.
ields
data
ples
unty
were
(see
each
ch. is
| the
S SO.
jpon
Was
the
nent
ient,
ous.
| the
the
f.
120
7 00000 Békes, 1991
AAAAA Hajdû, 1991
| Yo Szolnok, 1991
| smEuamm Békés, 1992
AAAAA Hajdû, 1992
*ooieiek Szolnok, 1992
| OOOO0 Békés, 19933
eeeee Szolnok, 1993 -
Yield (100 kg/ha)
S o © c
Ris oD
N
©
l
0 ' I T Ï T [ T T T T
0 1 2 3 4 0
T T T T T I T T T I T
6 7 8 9 10 11
GYURI
Figure 1. Data of 101 corn fields (little dots) with best fit line, and the calculated GYURI-values of 3 counties (symbols).
The y values are official statistical data.
Another effect for the year of 1993 is that the simple
cloud elimination algorithm found more images cloudy,
because of the generally low values in the remotely
sensed data, resulting fewer images for the index
calculations. The assumed linear relationship between the
yield and GYURI values could not be maintained for the
whole yield range with high accuracy, especially if the
lower range is considered. Therefore, in the future a
more complicated function is to be used to express the
relationship between yield and GYURI values. Further
uncertainties are introduced by the reliability factor of the
statistical publications, their errors are generally
unknown and unpredictable. Despite these difficulties the
results are promising and could be included in a future
operational yield estimation program.
The GYURIeyield relationship was provided for the
other 5 crops as well. With the exception of wheat,
summed acreage and also the size of the fields of these
crops are smaller thus resulting in fewer reference data.
Overall 98 wheat, 37 sunflower, 23 sugar beet and 25
barley reference fields were found and the correlations
were 84.7% for wheat, 69.8% for sunflower, 75.6% for
sugar beet and 68.1% for barley. In the case of alfalfa
the results couldn’t be well interpreted.
3. THE ROBUST METHOD
A new method was developed for estimating the national
and the regional average yields. It was partly an
improvement of the previously described method, but it
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 797
had to avoid some of its difficulties. Using the existing
database a modified method was developed resulting in a
new index called GYURRI (General Yield Unification
Robust Reference Index). It is important to emphasize
that cespite the name it is not like the ’robust’ methods
known from literature, and is based strongly on the
results and experiences of the previous method.
In this method we relied on low spatial resolution images
(1.1km x 1.1km, NOAA AVHRR) instead of the high
resolution ones (30m x 30m, LANDSAT TM). But the
NOAA AVHRR pixels over Hungary, even for wheat
and corn, the two crops with the greatest acreage, are
mostly mixed pixels. It means that the average
reflectance of a pixel is derived from a piece of land
which is a combination of a few, sometimes quite
different vegetation covers. The mathematical method
used for separating the components is called
‘decomposition’. This decomposition could be done in
principle, but in practice it is not that simple. The main
difficulty is that the accuracy of the geographical
correction of the AVHRR images is comparable to the
pixel size (a half pixel uncertainty even in the best case
is quite inherent). Therefore, even if we knew the
accurate crop distribution over the area (which requires
a very good classification) we would not be able to put
on a grid of AVHRR pixels with acceptable error limits
of plant area ratios. It is unnecessary to mention other
obstacles, this in itself was enough to make us decide to
tread another path avoiding decomposition. A method
wes reeced which doesn’t use uniform pixels and can
utilize "he mixed nixels, and can follow the vegetation
