International Archives of Photogrammetry and Remote Sensing. Vol. XXXII Part 7C2, UNISPACE ILL Vienna, 1999
54
/S^f\
ISPRS
UNISPACE III - ISPRS Workshop on
“Resource Mapping from Space”
9:00 am -12:00 pm, 22 July 1999, VIC Room B
Vienna, Austria
ISPRS
data processing phases from the images.
The final step in preprocessing is the classification of the
atmospherically corrected images. The classes are the crops (com,
wheat etc.), forest, water, towns etc. In this phase it is necessary to
use ground reference data (i.e. ground truth data). The output of this
The detailed relations between the preprocessed (or "calibrated")
satellite remote sensed data (surface reflectances, vegetation indices
etc.) and the important agricultural data (canopy state, crop yields
etc.) are very complicated, not well known and in some cases the
agricultural data practically can not be derived from these relations
because the number of unknowns are too high, the relations cannot
be invert to the interesting unknown etc. Therefore in the present
state of knowledge the application of heuristic, quasi-empirical
relations is the real solution of our problem (Hamar et al, 1996). A
special aspect is besides these that □ in general □ using heuristic
models are simple and cheap.
During the development of a heuristic model, in first step we must
derive special agricultural index (or indices) using the primary
satellite RS data (surface spectral reflectances or a vegetation
index, e.g. greenness (GN)), which have real andconfinned relation
to the unknown agricultural data. In our yield estimation and
forecasting models the special index is in high resolution case the
"General Yield Unification Reference Index" (GYUR1) and in the
low resolution "robust" model □ i.e. in the basic model □ the
"General Yield Unification Robust Reference Index" (GYURR1). In
second step we derive an accurate mathematical relation between
the special agricultural index and the interesting agricultural data.-
e.g. the yield.
During the development of the method and in the practical
application it is necessary to make some essential manipulation-
with satellite remote sensed data. In a few cases the agricultural
In our yield forecasting and estimation methods after the
development phase we liave a highly reliable, heuristic relation
between the special agricultural index (in tliis cases tire GYURI and
GYURRI) and the yield of the different crops. At the beginning of
the whole project six crops (com. wheat, sunflower, sugar beet,-
barley and alfalfa), forests and the great meadow 'Hortobagv' were
investigated, at the latest phase ten crops (com, wheat, sunflower,
sugar beet, barley, alfalfa, maize for silage, pea, rye and red clover)
were investigated. □ At the end of a given agricultural year it is
possible and is suitable to control and enhance these relations
knowing the reference agricultural data of the given year.
The main steps of yield forecasting and estimation are:
At a given calendar day (DOY) a segment of unified temporal
profiles (UTP's) is known from the starting day to the given day.
The analysis of these UTP segments produces die stress-indications
as was demonstrated in Figs 1. and 2.
In forecasting let the given actual segment of UTP's be continued
supposing a good final (continuous) growing phase. The result is
the forecasted UTP of a given crop.
step is the input of the acreage determination.
It is clear that this preprocessing is not only important but in most
cases more complicated than the final agricultural application itself.
technology has inherent temporal and spacial inhomogeneities, see
e.g. the growing of alfalfa. Therefore an investigation on field level
in tire present state of our knowledge and technology' will detect a
very' inhomogeneous state of different fields with the same yield □
see the Fig. 3. □, but a robust model using a great spacial averaging
(filtering) can produce acceptable results. Besides tliis the temporal
sampling of canopy depends on territorial and temporal weadier
conditions (haziness, clouds), and this produce quasi-stochastic
sampling differences in the temporal profiles of different places and
of different years. So it is necessary to use mathematical
transformation to unify’ the whole data set and to guarantee the
compatibility of the temporal profiles of the different years. In the
followings this mathematically transformed data set will be the
unified temporal profile (UTP) of a given vegetation index for a
given plant. The surface relief also have a great effect on remote
sensed data □ e.g. Ferencz et al, 1987 □ and on agricultural
indices. This effect and the results of correction are demonstrated
on Fig. 4.
In the determination of the total crop yields we have two tasks: the
determination of acreage and the derivation of the yield. The main
steps to determine the land use and acreage of crops are:
Classification of multitemporal, high resolution RS data at the
beginning of and during the growth.
Ordering the pixel-level classified data set into a probable field-
structure.
Determination of crop acreage.
The forecasted GYURI’s or GYURRFs could be calculated using
the forecasted UTP's, and these special agricultural indices
determine the forecasted yields. These data mean the upper limits of
the possible yields in the given cases. However, unpredictable
stresses can decrease these probable yields.
If tliis whole process is carried out at the beginning of harvesting or
just before of it, then the actual UTP's are complete, the special
agricultural indices can be determined using these complete UTP's
and they define the estimated yields.
The estimated yield and acreage data define the total agricultural
production.
3. RESULTS AND CONCLUSIONS
The Fig. 5. and Table 1 present results of a quasi-high resolution
method applied at com. In tliis method Landsat TM and NOAA
AVHRR data are used together. The method is useful to determine
the county-averages of yield. However, it is successful in field level
investigations too.
