YIELD FORECASTING OF MAIN CROPS USING REMOTE SENSING DATA
Cs. Ferencz, D. Hamar, J. Lichtenberger, P. Bognár, I. Ferencz-Árkos, G. Molnár, Sz. Pásztor, P. Steinbach,
B. Székely, Gy. Tarcsait, G. Timár
Eótvós University, Space Research Group, Hungary
(H-1117, Budapest, Pázmány Péter sétány 2.)
Commission VII, Working Group 2
KEY WORDS: crop yield forecasting, remote sensing, NOAA AVHRR
ABSTRACT:
To estimate the yield of the main crops of Hungary and forecast their production using satellite data has increasing
importance. The paper outlines the problems, presents yield forecasting models and discusses the obtained results. An
operative yield forecasting and production estimating procedure can be developed based on the models for the main crops
for small and large regions, as well.
1. INTRODUCTION
Crop estimating using satellite images is worldwide
considered a matter of strategic importance. We aimed
the development of a yield estimating method based on
satellite images that are easily accessible, and densely
cover the growing season of the vegetation. These
criteria are best matched by the NOAA AVHRR images.
However, due to the poor spatial resolution of these
images, they must be combined with satellite images of
better resolution (LANDSAT TM) and with ground
reference data.
At first, we examined data of three consecutive years
(1991-1993) collected from 3 counties in Hungary
(Békés, Hajdü-Bihar, Jász-Nagykun-Szolnok) for 6 crops
(corn, wheat, sugar beet, sunflower, barley, alfalfa). For
all crops with the exception of alfalfa we succeeded to
find some relationships between the crop yield and the
calculated index called GYURI (General Yield
Unification Reference Index) for some groups of fields.
GYURI values are calculated uniformly for each year
from the remotely sensed data and characterize the
strength of plant growth.
For the data processing it is important to have good
spatial resolution satellite images (LANDSAT TM) of the
area suitably corrected to atmospherical and geometrical
effects. This condition couldn’t always be fulfilled
because of financial and technical obstacles, so we
developed a second, a so called robust method and a new
index (GYURRI, General Yield Unification Robust
Reference Index). This method is based on the experience
acquired by the previous method, but doesn’t use the
high resolution data only for checking the results. '
Surprisingly the new method gave good results even for
the alfalfa fields, where the previous method had failed.
By the help of a newly developed relief correction we
extended our examinations to the whole area of Hungary
and succeeded to give a good estimation for the yields of
the 6 different crops.
In the sections below we describe briefly the results of
the above mentioned methods and outline the potential
directions of future research.
2. RESULTS OF THE REGIONAL CROP YIELD
ESTIMATION METHOD
Since the remotely sensed data-yield relationship was not
possible to establish by pixel decomposition, a different
approach had to be found (Hamar et al., 1996). After the
geometrical and atmospherical correction of satellite
images (Ferencz et al., 1993, Lichtenberger et al.,
1995), efforts led to the development of GYURI (General
Yield Unification Reference Index) which can be applied
uniformly to each year and all crops. GYURI also avoids
the pixel-by-pixel decomposition of land cover which
never can be accomplished with the required accuracy
and certainty.
Of all the crops, corn was most thoroughly examined.
Farms supplied crop yield data for 101 corn fields of the
3 above mentioned counties. For all fields the GYURI
value was calculated from the remotely sensed data, and
plotted against the yield data. Assuming linear
relationship, lines were fitted and a correlation of 86.6%
was found between the data sets. The farm-provided yield
data certainly contains an inherent error which influences
the calculated parameters, but there was no way — apart
from a few case — to check or reduce these errors.
For verification of the method the county average yields
were calculated for the 3 counties. The real yield data
were known from official statistical publications. Samples
were chosen representing the given county
(representative samples’), their GYURI values were
obtained and county average yield were estimated (see
Figure 1.).
The results are containing all the three counties in each
year, except the 1993 result for Hajdü-Bihar, which is
still worked on. It is clearly seen that while in 199] the
estimation is very good, in years of draught it is less so.
It is partly because the ground truth information, upon
which the representative samples are based on, was
unreliable in these years, and partly because of the
draught, the vegetation characteristics is not so prominent
even in the most important periods of crop development,
making the classification procedure somewhat erroneous.
The other difficulty is that for fields of less yield the
GYURI values are less accurate and therefore the
deviation of the relating points on the plot is greater.
