back
ADAPTATION OF CROP-WEATHER MODELS
IN AUSTRIA AND BULGARIA
Vesselin Alexandrov1,2 , Josef Eitzinger2 and Michael Oberforster3
1
- Bulgarian Academy of Sciences, National Institute of Meteorology and Hydrology,
66 Tzarigradsko shose Blvd., BG-1784 Sofia, Bulgaria, Phone: (+359-2) 975 39 86,
Fax: (+359-2) 988 44 94, E-mail: Vesselin.Alexandrov@meteo.bg
2
- University of Agricultural Sciences, Institute of Meteorology and Physics,
18 Tuerkenschanzstrasse, A-1180 Vienna, Austria, Phone: (+43-1) 470 58 20 34,
Fax: (+43-1) 470 58 20 60, E-mail: sepp@tornado.boku.ac.at
3
- Institute of Plant Production, Federal Office and Research Centre of Agriculture,
191 Spargelfeldstrasse, A-1226 Vienna, Austria, Phone: (+43-1) 732 16 4196,
Fax: (+43-1) 732 16 4211, E-mail: moberforster@bfl.at
1. Introduction
Year-to-year fluctuations in weather causes large variations in crop yields. Uncertainty
in weather creates a risky environment for agricultural production. During the last decades the
application of simulation and system analysis in agricultural research has increased
considerably. The simulation model is one of the most complex method among the
approaches used to describe the soil-plant-atmosphere system.
Numerous crop growth and yield models have been developed for a wide range of
purposes in recent years (e.g. Hoogenboom, 2000). These models range in complexity from
the most sophisticated simulators of plant growth, primarily intended for research into plant
physiological interactions, to multiple regression models using only a few monthly weather
variables to forecast regional crop yields. Generally, plant-process yield models have been
developed to predict yield at the level of an average plant in a specified field. Thus the input
data required by these models include plant parameters specific to the variety or hybrid
planted in some field and soils parameters describing the soil in that field. The prediction of
crop development is an important aspect of crop growth modelling. Crop models that use
daily weather, soil and plant data in simulating crop yields have the potential for being used to
assess the risk of producing a given crop in a particular soil-climate regime and for assisting
in management decisions that minimize the risk of crop production (e.g., Tsuji et al., 1998).
Current algorithmic models are frequently inaccurate when they are applied to
locations other than that where developed. According to Hoogenboom et al. (1999) before
using a crop model for a particular production region, it is important that a minimum amount
of crop phenological and yield data should be applied to allow estimation of the model
performance for that region’s cultivar types and for calibration of specific parameters. That is
why the objective of this study was to adapt (including calibration and verification) crop
simulation models for specific environmental conditions in Austria and Bulgaria.
Crop models, in general, integrate current knowledge from various disciplines,
including meteorology, soil physics, soil chemistry, crop physiology, plant breeding, and
agronomy, into a set of mathematical equations to predict growth, development and yield (e.g.
Hoogenboom, 2000). Baier (1979) provided some interesting background and terminology for
what he called “crop–weather models”. This term was assumed to be also used in this study.
back
2. Method and experimental material
The following crop-weather models were adapted for the environmental conditions in
selected regions in Austria and Bulgaria:
The CERES generic grain cereal and CROPGRO grain legume models of the Decision
Support System for Agrotechnology Transfer (DSSAT) (e.g. Tsuji et al., 1998) were used to
simulate crop growth, development and yield formation of important agricultural crops such
as winter wheat, maize, barley and soybean. These crop models are designed to use a
minimum set of soil, weather, genetic and management information. The models are daily
incrementing and require daily weather data, consisting of maximum and minimum
temperature, solar radiation and precipitation as input. They calculate crop phasic and
morphological development using temperature, daylength, genetic characteristics and
vernalization, where appropriate. Leaf expansion, leaf growth and plant population provide
information for determining the amount of light intercepted, which is assumed to be
proportional to biomass production. The biomass is partitioned into various growing organs in
the plant using a priority system. A water and nitrogen balance sub-model provides feedback
that influences various growth and development processes. A new soil water model was
completed, which contains improved infiltration, redistribution and root uptake calculations.
Restrictions to percolation are included in soil inputs so that perched water tables can be
simulated along with oxygen stress effects on root and crop growth processes. An option has
been added to compute potential evapotranspiration using the Penman equation, which uses
humidity and wind speed as input if these data are available (e.g. Hoogenboom et al., 1999;
Tsuji et al., 1994).
The crop-weather models of DSSAT have been distributed to many national and
international scientists at research, educational and private institutions and organizations, as
well as to many other individuals across the world. The generic grain cereal model CERES
and grain legume model CROPGRO have been extensively tested in North and South
America, Africa, Asia and Europe (e.g. IBSNAT, 1993; Tsuji et al., 1994). Normally, these
crop models are run on a “point” basis, i.e., input data in terms of soil and weather conditions
are assumed to relate to one location, e.g., a field or plot where an experiment was conducted
(e.g. Hoogenboom et al., 1999).
The WOFOST (WOrld FOod STudies) explanatory and dynamic crop model (e.g.
Supit et al., 1994; Van Diepen et al., 1989) Ver. 7.1 was also used in the study. This model
was developed by the DLO-Winand Staring Centre and Research Institute for Agrobiology
and Soil Fertility in Wageningen (e.g. Boogaard et al. 1998). WOFOST is a member of the
family of models developed in Wageningen by the school of C.T. de Wit. It is designed to
simulate the growth and development of annual field crops and grass during the growing
season, from sowing to maturity or harvest in daily increments. It simulates a cropping system
defined by crop, the weather conditions and the soil parameters, including the plant and soil
water balance. Outside the crop-growing period the soil water balance can be calculated for
bare soil conditions. The major processes taken into account are phenological development,
assimilation, respiration and evapotranspiration. WOFOST uses parameters and functions
describing the effects of temperature, radiation and water stress on important physiological crop
processes as a function of the development stage and crop status. For example, the
photosynthesis response curve is limited by a maximum leaf CO2 assimilation rate and initial
light use efficiency of a single leaf. These parameters are further related to temperature at a
specified carbon dioxide concentration. Biomass partitioning is a function of the development
stage of the crop, while temperature determines the development rate of the crop.
The WOFOST model is designed for simulation of three production levels. The
potential yield production level is limited only by temperature, solar radiation and the specific
back
physiological plant characteristics. Such conditions are possible in greenhouses or in very
intensive agricultural production systems (e.g. under field conditions with optimum irrigation
and nutrition). At the water-limited production level, the soil and plant water balance is also
included in the simulation of crop growth with the interactions between transpiration, stomata
opening, CO2 assimilation and water uptake being considered. The third production level is
also limited by nutrients.
Crop model validation is accomplished by imputing the user's minimum data set,
running the model and comparing outputs. To validate crop-weather models researches
compare simulated outcomes with measured results obtained from the experiments. Prior to
evaluating the models, the genetic coefficients for the varieties used are estimated. Usually
calibration consists of determining sets of model parameters/coefficients for the studied
locations to adjust timing of growth stages and yield components. Generally, calibration is
done by running crop-weather models and adjusting the coefficients to correct unreasonable
results and running the model again, repeatedly (e.g., Penning de Vries et al., 1989). Before
validation of crop-weather models a procedure of model calibration is strongly recommended.
That is why the strategy of crop-weather model adaptation, applied in this study, assumed
both calibration and verification/validation procedures.
Model parameters/coefficients may be determined either in controlled environments
or under field conditions, but since we are model users, who do not have access to controlled
environment facilities, most determinations will be made using field data. To help with this
we developed a subroutine which enables to estimate genetic coefficients from field data sets
that relate to environment, dates of phenological events, and various growth aspects by means
of an optimization procedure, suggested by Rozenbrok (e.g. Himeblau, 1975). In the purposed
subroutine, the coefficients for a crop cultivar were estimated by running the appropriate
model with approximate coefficients, comparing the model output (e.g., dates of predicted
events such as flowering and maturity dates, kernel numbers per ear and grain biomass
accumulation for each of the experiments) to actual data, and then altering the genetic
coefficient until the predicted values and measured values match (i.e., by running several
iterations). The coefficients were determined in a preset sequence, with those that relate to
phenological aspects being determined first.
The above crop-weather models require daily weather data, agrometeorological and
soil information as input. Agrotechnological (e.g., phenological development, yield,
agrotechnology applied), weather and soil data from some locations in Upper Austria, Lower
Austria (Marchfeld) and south-eastern Austria (Fig. 1a) were incorporated within the study.
Soil characteristics such as horizon designation, percentages of clay, silt, coarse fractions,
organic carbon, saturated hydraulic conductivity, total nitrogen, aluminum saturation, bulk
density, PH in water and buffer for every given soil layer were used in order to create
appropriate soil data profiles in the selected Austrian regions.
Data from crop experiments, conducted during the last two decades in Austria were
used. The Austrian registered winter wheat cultivar “Perlo” was used for Lower Austria.
“Perlo” is a well established cultivar, especially adapted for dry and warm regions such as the
region of Marchfeld. Agrotechnological, “Perlo” phenological and yield data, as well as
weather data from Grossenzersdorf (Marchfeld, Lower Austria) for the period 1985-1999
were gathered for the simulation study. Agrotechnological, phenological and yield data of two
additional winter wheat cultivars (“Renan” and “Silvius”) grown in Upper Austria and southeastern Austria for the period 1991-2000 were also used. For the same period and regions
crop data of two spring barley cultivars (namely “Meltan” and “Elisa”) were collected.
Soybean data (“Ceresia” and “Apache” cultivars) from field experiments, carried out in
Grossenzersdorf from 1992 to 1999 were also collected.
back
•
a)
Freistadt
Grossenzersdorf
Lambach
Wartberg
Gleisdorf
Eltendorf
Werndorf
b)
Kapitanovci
Dobrich
Kojnare
Pavlikeni
Burgas
Ognjanovo
Radnevo
Ljubimec
Fig. 1. Spatial distribution of the stations in Austria (a) and Bulgaria (b), used in the study
The CERES crop-weather model for maize and winter wheat was calibrated and
verified at 21 experimental variety stations in North and South Bulgaria (Fig. 1b) using field
experiments conducted during the period 1980-1993. The Bulgarian maize hybrid "Kneja
611" and winter wheat bread variety "Sadovo 1" were used. The phenological and yield data
were obtained from the Bulgarian National Variety Commission of the Ministry of
Agriculture. Daily weather data, including precipitation, maximum and minimum
temperature, and solar radiation, were collected for the same period for the nearest weather
stations, which are members of the weather network of the Bulgarian National Institute of
Meteorology and Hydrology. Different soil types (e.g., typical chernozems; leached
chernozem-smolnitza; grey forest, moderately loamy; leached cinnamonic forest; delluvialmeadow, sandy; alluvial, clayey-sandy; etc.) were taken into account in the study.
back
3. Results and discussion
3.1. Austria
3.1.1. Winter wheat (CERES and WOFOST models)
The CERES model for winter wheat was calibrated using planting and maturity dates
and yields of the wheat cultivar “Perlo” from 1985 to 1999 at station Grossenzersdorf (Lower
Austria). The calibration consisted of determining sets of five cultivar coefficients to adjust
timing of growth stages and yield components. This was done by running the model and
adjusting the coefficients to correct unreasonable results and running the model again,
repeatedly. After each run coefficients controlling flowering date and maturity date were
readjusted automatically. When the estimated growth stage dates were reasonable for a given
year, the coefficients controlling grain number and grain weight were adjusted to set
estimated yields at a reasonable level. The CERES model adequately simulates the winter
wheat growth stages duration as influenced by cultivar, planting date (day-length,
temperature) and delay due to transplanting. The difference between simulated and observed
dates of flowering and physiological maturity varies between 0 and 7 days (Fig. 2) The
simulated grain yields are in most cases in accord with the measured data, with predicted
yield results mainly within acceptable (e.g. Chirkov, 1969; Procerov and Ulanova, 1961)
limits of ± 17 % of measured yields (Fig. 3). In year 1993 where some drought periods caused
plant water stress, there is 42% difference between the simulated and measured winter yield.
The highest deviation (83%) between the simulated and measured wheat yield occurs for
1996. It is a result a growth limitation (e.g. Aggarwal et al., 1994), not considered by the
CERES model.
290
Observed (dap)
280
270
1:1 line
260
250
flowering
maturity
240
230
230
240
250
260
270
Simulated (days af ter planting)
280
290
Fig. 2. Comparison between simulated and observed flowering and maturity of winter wheat,
cultivar “Perlo” in Grossenzersdorf (1985-1999) (CERES model)
The CERES model for winter wheat was also calibrated for the wheat cultivars
“Renan” and “Silvius”, cultivated intensively in Upper Austria and south-eastern Austria
during the last decade. The “Renan” wheat cultivar is representative for Eastern Austria, while
the “Silvius” cultivar is grown especially in Upper Austria. A comparison between the
simulated and observed flowering dates both for “Renan” and “Silvius” wheat cultivars is
presented in Figure 4.
back
Simulated
Measured
8000
-16%
Wheat grain yield (kg/ha)
7%
-17%
-22%
6% -11%
-6%
7000 10%
15%
0%
6000
18%
83%
12%
42%
6%
5000
4000
3000
1986
1988
1990
1992
1994
1996
1998
Year
Fig. 3. Variations of simulated and observed winter grain yield (cultivar “Perlo”) in
Grossenzersdorf
Observed wheat f lowering (dap)
The difference between the simulated and observed wheat flowering dates in all selected
stations and years is less than 1 week. Even for the “Silvius” wheat cultivar grown in Freistadt
(Upper Austria) the above deviation from 1993 to 2000 is not higher than 1 day (Fig. 5). The
simulation error for wheat maturity is between 0 and 7 days which is considered as a good
result (Fig. 6). In Freisdadt the maximum deviation between the simulated and observed
physiological maturity of winter wheat is 3 days in 1996 and 2000 (Fig. 5). It is obviously that
the variations of the simulated phenological stages by the CERES model are following the
variations of the observed ones.
260
1:1 line
250
240
230
220
210
210
220
230
240
250
260
Simulated wheat f lowering (days after planting)
Fig. 4. Comparison between simulated and observed flowering date of winter wheat (“Renan”
and “Silvius” cultivars) at the stations in Upper Austria and south-eastern Austria
(1991-2000)
back
310
Days af ter planting
300
290
flowering (simulated)
280
flowering (observed)
270
maturity (simulated)
maturity (observed)
260
250
240
230
1993
1994
1995
1996
1997
1998
1999
2000
Year
Observed wheat maturity (dap)
Fig. 5. Variations of simulated and observed flowering and maturity of winter wheat
(“Silvius” cultivar) in Freistadt
305
1:1 line
300
295
290
285
280
275
275
280
285
290
295
300
305
Simulated wheat maturity (days af ter planting)
Fig. 6. Comparison between simulated and observed maturity date of winter wheat (“Renan”
and “Silvius” cultivars) at the stations in Upper Austria and south-eastern Austria
(1991-2000)
The CERES model parameters impacting wheat growth and yield formation were
calibrated using the “Renan” cultivar data from Lambach (Upper Austria). Figure 7a
compares the variations of the simulated and measured winter yield between 1991 and 2000.
Only in 2 years the deviation between the simulated and measured wheat grain yield is above
20%. It should be noted that these 2 years are characterized with very high measured wheat
grain yield – near 9000 kg/ha. The simulation error for wheat grain yield is less or equal to
10% from 1993 to 1999 (Fig. 7a).
back
For verification, crop models are used to simulate crop responses under specific
experimental conditions for which observed data are available. Usually, verification refers to
the testing of the model on an independent data set. When this is done, we have an empirical
indication as to the model's applicability to years and/or locations other than those for which it
was fitted or calibrated. Figure 7b represents, for example, a verification of the calibrated
CERES model parameters from Lambach on a neighboring station (Wartberg, Upper Austria).
The deviation between the simulated and measured yield from 1994 to 2000 in Wartberg
exceeds the ±20% interval only in 1995. It is necessary to point out that the measured winter
wheat grain yield was very low in this year (Table 1) at the most stations used in the study,
although weather conditions were favorable for normal crop growth (at least without water
stress). It should be also specified that the eventual impact of pest and diseases on crop
growth, development and yield formation was not account within the study.
Wheat yield (kg/ha)
9000
-24%
a)
-2%
8000
13% -23%
7000
10%
6% -8%
-2%
7%
7%
6000
5000
4000
1990
1992
1994
1996
1998
2000
Year
Wheat yield (kg/ha)
9000
b)
simulated
8000
measured
-1%
9% 34%
7000
5%
4%
-11%
18%
6000
5000
4000
1990
1992
1994
1996
1998
2000
Year
Fig. 7. Variations of simulated and measured winter wheat (cultivar “Renan”) yield in
Lambach (a) and Wartberg (b)
back
Table 1. Deviation between simulated and measured winter wheat grain yield in 1995
Wheat cultivar “Renan”
Station
Simulated Measured Deviation
(kg/ha)
(kg/ha)
(%)
7035
6560
7
Lambach
Wartberg
7059
5260
34
Freistadt
6662
5440
22
Gleisdorf
8593
9020
-5
Eltendorf
8502
4720
80
*- not in Fig. 8
Wheat cultivar “Silvius”
Simulated Measured Deviation
(kg/ha)
(kg/ha)
(%)
7465
5970
25
7465
5380
39
7338
6200
18
8074
9470
-15
8209
3820
114*
The low measured winter wheat grain yield in Upper and south-eastern Austria in
1995 versus the relatively simulated high wheat yield caused most of the significant
deviations beyond the ±20% error interval presented in Figure 8. The comparison of the
simulated and measured yields in this figure indicates satisfactory performance of the CERES
model for the selected regions.
Measured wheat yield (kg/ha)
10000
calibration
verification
(-20%;+20%)
9000
8000
7000
1:1 line
6000
5000
5000
6000
7000
8000
9000
Simulated wheat yield (kg/ha)
Fig. 8. Comparison between simulated and measured winter wheat (“Renan” and “Silvius”
cultivars) grain yields at the stations in Upper Austria and south-eastern Austria (1991-2000)
Calibration of the WOFOST model for Austrian environment conditions
(Grossenzersdorf, Lower Austria) had been initiated by Eitzinger et al. (2000). The simulated
crop-growing duration of the “Perlo” winter cultivar from 1985 to 1999 correlated well with
the observed data, although in some years (i.e. 1986, 1992 and 1993) larger differences (11-14
days) were also noted. Nevertheless, the mean difference between the simulated and observed
crop-growing season was only 1 day and on acceptable range from 0 to 8 days, excluding the
above three years (Eitzinger et al., 2000).
The WOFOST simulation model was also calibrated for the winter wheat cultivars
“Renan” and “Silvius” for the selected locations in Upper Austria and south-eastern Austria.
Some results, obtained for station “Lambach (Upper Austria) are presented in Figures 9-10.
back
Generally, the variations of the simulated and observed phenological stages (including
flowering and physiological maturity) of the “Renan” winter wheat cultivar are similar during
the investigated period (Fig. 9). The difference between the simulated flowering and maturity
dates is not higher than 1 week, except in 1994 for flowering (8 days). The obtained
deviations of the simulated winter wheat yield, relative to the observed one is less than 20%
from 1991 to 1999. A comparison between the results presented in Figures 7a and 10 shows
that the WOFOST model also simulates not well the higher yields in 1992 and 2000. It should
be noted that the effect of fertilizers on the crop growth and final yield formation was not
account for the crops in Upper Austria and south-eastern Austria. It might be a reason for the
simulated underestimation of the winter wheat yield in these two years (Fig. 10).
220
210
Julian day
200
flowering (simulated)
190
flowering (observed)
180
maturity (simulated)
170
maturity (observed)
160
150
140
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year
Fig. 9. Variations of simulated and observed flowering and maturity of winter wheat
(“Renan” cultivar) in Lambach; WOFOST model
9000
simulated
-18%
-21%
Wheat yield (kg/ha)
measured
17%
8000
15%
0%
-4%
10%
-1%
7000
-6%
5%
6000
5000
4000
1990
1992
1994
1996
1998
2000
Year
Fig. 10. Variations of simulated and measured winter wheat (cultivar “Renan”) yield in
Lambach; WOFOST model
back
3.1.2. Spring barley (CERES model)
Observed barley flowering (dap)
In a similar way the CERES model was calibrated and verified for spring barley
(“Meltan” and “Elisa” cultivars) at some locations in Upper and south-eastern Austria. Some
of the obtained results are presented in Figures 11-14. The simulated by the CERES model
phenological stages are compared with the observed flowering and maturity dates in Figures
11 and 12.
100
95
90
1:1 line
85
80
75
70
65
65
70
75
80
85
90
95
100
Simulated barley f lowering (days after planting)
Observed barley maturing (dap)
Fig. 11. Comparison between simulated and observed flowering date of spring barley
(“Meltan” and “Elisa” cultivars) at the stations in Upper Austria and south-eastern Austria
(1992-2000)
135
130
1:1 line
125
120
115
110
105
100
95
95
100
105
110
115
120
125
130
135
Simulated barley maturity (days after planting)
Fig. 12 Comparison between simulated and observed maturity date of spring barley (“Meltan”
and “Elisa” cultivars) at the stations in Upper Austria and south-eastern Austria (1992-2000)
back
Barley yield (kg/ha)
The deviation of the simulated flowering dates from the observed ones varies between 0 and 9
days. The difference between the simulated and observed maturity dates of spring barley also
is not higher than 9 days except 2000 in Freisdat (Upper Austria). The 2000 simulated
maturity date of the “Elisa” cultivar is 12 days later than the observed physiological maturity
date. Nevertheless, it might be expected that the CERES model would simulate spring barley
maturity dates that are not later or earlier than 10 days in comparison to the observed/real
ones.
8000
a)
5% -17%
15%
7000
14% -17%
15%
6000
4%
9%
10%
5000
4000
1992
1994
1996
1998
2000
Barley yield (kg/ha)
Year
8000
b)
simulated
-10%
7000
6000
measured
23%
5000
4000
1992
1994
1996
1998
2000
Barley yield (kg/ha)
Year
8000
c)
-33%
-6%
7000
-7%
15%
-6%
6000
4%
6%
5000
4000
1992
1994
1996
1998
2000
Year
Fig. 13. Variations of simulated and measured spring barley (“Meltan” cultivar) yield in
Lambach (a), Wartberg (b) and Gleisdorf (c)
back
Measured wheat yield (kg/ha)
The calibrated CERES model parameters optimized for station Lambach (Upper Austria)
were applied for verification in stations Wartberg and Gleisdorf (Fig. 13). Generally, the
variations in simulated spring barley grain yield follows the variations of the measired yield,
especially from 1992 to 1996. The observed yield decreases in 1992, 1993, 1994 and 1995 are
well simulated by the crop model. A difference of 33% between the simulated and measured
barley yield is seen for 1997 in Gleisdorf (south-eastern Austria). The measured yield in that
year was very high in comparison to the yield of the rest years and stations – above 8000
kg/ha. It seems that the CERES model can not account great positive variations of crop yield
due to model limitations themselves and/or input data limitations. The comparison between
the simulated and measured spring barley grain yield of the “Meltan” cultivar as well as the
“Elisa” cultivar for 3 stations are also presented in Figure 14.
8000
1:1 line
7000
6000
5000
calibration
verif ication
(-20%;+20%)
4000
4000
5000
6000
7000
8000
Simulated wheat yield (kg/ha)
Fig. 14. Comparison between simulated and measured spring barley (“Meltan” and “Elisa”
cultivars) grain yields at stations Lambach, Wartberg and Gleisdorf
3.1.3. Soybean (CROPGRO model)
The CROPGRO model was calibrated and verified for two soybean cultivars
(“Ceresia” and “Apache”) using agrometeorological data from Grossenzersdorf (Lower
Austria) for the period 1992-1999. The model simulates flowering and physiological maturity
dates well for each cultivar in all considered years, used for model calibration (1995-1999) as
well as for model verification (1992-1994). The difference between the simulated and
measured dates of these two phenological stages varies from 0 up to 5 (Fig. 15). The deviation
between the simulated and measured soybean yield by the cultivar ”Ceresia” varies in most
years between 1% (1994) and 20% (1997, 1999) (Fig. 16a). The difference between measured
and simulated yield by the cultivar ”Apache” varies between 6 and 20 % except in 1994 (Fig.
16b). The highest differences between the simulated and measured yield in 1994 for both two
cultivars (53 and 54%, respectively) are caused due to simulated insufficient soil moisture
during the crop-growing season. It should be noted that the CROPGRO model is very
sensitive to water stress. 1994 was a very dry year, with dry spells during the summer season.
The model simulates very well the variations of measured soybean yield during the studied
period (Fig. 16), which is a precondition for a good model performance in the selected region.
back
Verification
flowering
Observed (days after planting)
140
maturity
120
1:1 line
100
80
Calibration
flowering
60
maturity
40
40
80
100
120
140
Simulated (days af ter planting)
Fig. 15. Comparison between simulated and observed phenological stages of soybean
(“Ceresia” and “Apache” cultivars) in Grossenzersdorf (1993-1999)
Soybean seed yield (kg/ha)
5000
60
a)
Simulated
Measured
20%
10%
4000
3000
2000
7%
-16%
-1%
-20%
-14%
-53%
1000
0
Soybean seed yield (kg/ha)
5000
4000
1992 1993 1994 1995 1996 1997 1998 1999
Year
-20%
b)
-8%
12%
-9%
-6%
14%
3000
2000
-54%
1000
0
1992 1993 1994 1995 1996 1997 1998 1999
Year
Fig. 16. Variations of simulated and measured seed yield of soybean cultivars “Ceresia (a)
and Apache (b), in Grossenzersdorf (Lower Austria)
back
3.2. Bulgaria
3.2.1. Maize (CERES model)
As a first step, the CERES model parameters determining maize development and
phenological stage occurrence were calibrated. Figure 17 represents a comparison between
the simulated and observed silking and maturity dates of maize for the investigated 21
experimental crop variety stations across the country. The difference between the simulated
and observed silking dates during the period of calibration (1984-1990) in most cases (108
cases from total 112 field experiments) is up to 1 week. The deviation of the simulated
maturity dates, relative to the observed ones is less than 2 weeks, except 2 cases. A better
view for the distribution of the simulation error regarding these two phenological stages can
be obtained in Figure 18.
22.X
Observed date
07.X
17.IX
1:1 line
28.VIII
silking
08.VIII
verification
19.VII
maturity
verification
29.VI
29.VI
19.VII
08.VIII
28.VIII
17.IX
07.X
22.X
Simulated date
Fig. 17. Comparison between simulated and observed silking and maturity dates of maize in
Bulgaria during the periods of calibration (1984-1990) and verification (1991-1993)
20
Number
15
silking
maturity
10
5
0
-16
-12
-8
-4
0
4
∆n (days)
8
12
16
Fig. 18. Number of deviations (∆n) between simulated and observed silking and maturity
dates of maize in Bulgaria during the period of calibration (1984-1990).
back
After identification of the CERES model parameters, characterizing the vegetative and
reproductive periods of maize crop, the model parameters related to maize yield formation
were optimized. The biological meanings of these parameters are related to the maximum
grain number for one crop and the rate of grain filling. Within 78.8% of the total maize field
experiments the error of the simulated number of grains/m2, relative to the measured one is
less than 20%, whereas the deviation of the simulated maize grain weight from the measured
weight is higher than 20% only in 6 cases (Fig. 19).
a)
C
8%
D
7%
b)
E
4% F
3%
B
33%
CD
E
2%
1%
3%
B
24.3%
A
54.5%
A
61%
Legend: A - 0% ≤ ∆ ≤ 10%, B - 10% <∆ ≤ 20%, C - 20% < ∆ ≤ 25%,
D - 25% < ∆ ≤ 30%, E - 30% < ∆ ≤ 40%, F - 40% < ∆ ≤ 50%
Fig. 19. Deviation (∆) between simulated and measured number of grains/m2 (a)
and grain weight (b) of maize in Bulgaria (1984-1990)
The deviation between the simulated maize grain yield and the measured one is less
than 10% for more than 1/3 (44%) of all field experiments and it is less than 20% for 73% of
all cases *Fig. 20 ands 21). The difference from 25 and 50% is observed for 21 field
experiments (from total 112).
15200
Measured yield (kg/ha)
1:1 line
12700
10200
7700
5200
calbration
verification
(-20%;+20%)
2700
2700
5200
7700
10200
12700
15200
Simulated maize yield (kg/ha)
Fig. 20. Comparison between simulated and measured maize grain yield in Bulgaria during
the periods of calibration (1984-1990) and verification (1991-1993)
back
C
9%
D
4%
E
8%
F
6%
B
29%
A
44%
Legend: A - 0% ≤ ∆ ≤ 10%, B - 10% <∆ ≤ 20%, C - 20% < ∆ ≤ 25%,
D - 25% < ∆ ≤ 30%, E - 30% < ∆ ≤ 40%, F - 40% < ∆ ≤ 50%
Fig. 21. Deviation (∆) between simulated and measured maize grain yield in Bulgaria
(1984-1990)
It is necessary to specify again that up to 20% deviation of the simulated crop yield, relative
to the observed one is considered as a good result in this study, following for example
Chirkov, (1969), Procerov and Ulanova (1961), Slavov and Vitanov (1977).
In order to validate the calibrated CERES model parameters for maize, the model was
verified on independent agrometeorological data set including the periods 1980-1983 and
especially 1991-1993. The differences between the simulated and observed silking and
maturity dates of maize are above 7 and 14 days, respectively only in 6 cases from 1991 to
1993 (from total 50 field experiments) (Fig. 17). The deviation between the simulated and
observed maturity dates varies up to 1 week in 1/3 of the experiments, used for model
verification. The simulation error, related to the number of grains/m2 and grain weight is less
than 20% within 60 and 84% of the cases, respectively. As a result, the deviation between the
simulated and measured maize grain yield is on the interval 0-25% for 33 field experiments
(66% of the total), carried out during the period 1991-1993 (Fig. 20). The agrometeorological
data from 1980 to 1983, available for some of the experimental crop variety stations, were
also used for model verification. For example, the variations of the simulated and measured
yield of maize at station Kojnare (North Bulgaria) are presented in Figure 22.
3.2.2. Winter wheat (CERES model)
The CERES model for winter wheat was also adapted for the environmental
conditions in Bulgaria by applying crop, weather and agrotechnological data from 1980 to
1993. The years from 1984 to 1990 were used for calibration of the model parameters and the
rest years were left for model verification. The obtained results are presented in Figures 2328.
Maize grain yield (kg/ha)
back
simulated
15400
measured
12900
10400
7900
5400
1980
1982
1984
1986
Year
1988
1990
1992
Fig. 22. Variations of simulated and measured grain yield of maize
at station Kojnare (North Bulgaria) during the periods of calibration (1984-1990)
and verification (1980-1983 and 1991-1993)
Observed date
19.VII
29.VI
1:1 line
09.VI
flowerring
verification
maturity
verification
20.V
30.IV
30.IV
20.V
09.VI
Simulated date
29.VI
19.VII
Fig. 23. Comparison between simulated and observed flowering and maturity dates of winter
wheat in Bulgaria during the periods of calibration (1984-1990) and verification (1991-1993)
The departure of the simulated dates of winter wheat flowering and maturity, relative
to the observed ones is less than 3 days in 80 and 68%, respectively for all 132 field
experiments executed from 1983 to 1990. This difference is higher than 1 week only for 4 and
6 cases (from total 58 field experiments), respectively in the years of model verification from
1991 to 1993 (Fig. 23 and 24). The model simulation error for the number of wheat grains/m2
and grain weight is less than 20% for 74 and 93%, respectively of the field experiments,
carried out from 1983 to 1990г (Fig. 25).
back
25
flowering
maturity
Number
20
15
10
5
0
-10
-8
-6
-4
-2
0
2
4
6
8
10
∆n (days)
Fig. 24. Number of deviations (∆n) between simulated and observed flowering and maturity
dates of winter wheat in Bulgaria during the period of calibration (1984-1990).
a)
C
11%
D
4%
b)
E
8%
F
2%
B
14%
C
5% D
2%
B
30%
A
45%
A
79%
Legend: A - 0% ≤ ∆ ≤ 10%, B - 10% <∆ ≤ 20%, C - 20% < ∆ ≤ 25%,
D - 25% < ∆ ≤ 30%, E - 30% < ∆ ≤ 40%, F - 40% < ∆ ≤ 50%
Fig. 25. Deviation (∆) between simulated and measured number of grains/m2 (a)
and grain weight (b) of winter wheat in Bulgaria (1984-1990)
As a result of the above good model results, the deviation between the simulated and
measured grain yield of winter wheat is also considered as a satisfactory performance of the
CERES model (Fig. 26-28). The simulation error for the main periods of calibration (19841990) and verification (1991-1993) is beyond the ±20% interval only for 28% of the
considered all 190 field experiments. In a similar way, as it was done for maize crop,
agrometeorological data from 1980 to 1983, when available, were also applied for additional
model verification (Fig. 28).
Measured wheat yield (kg/ha)
back
1:1 line
7750
6250
4750
calibration
verification
(-20%;+20%)
3250
3250
4750
6250
Simulated wheat yield (kg/dka)
7750
Fig. 26. Comparison between simulated and measured grain yield of winter yield in Bulgaria
during the periods of calibration (1984-1990) and verification (1991-1993)
C
9%
D
9%
E
4% F
3%
B
27%
A
48%
Legend: A - 0% ≤ ∆ ≤ 10%, B - 10% <∆ ≤ 20%, C - 20% < ∆ ≤ 25%,
D - 25% < ∆ ≤ 30%, E - 30% < ∆ ≤ 40%, F - 40% < ∆ ≤ 50%
Fig. 27. Deviation (∆) between simulated and measured grain yield of winter wheat
in Bulgaria (1984-1990)
4. Conclusions
There was found good agreement between real phenological stages and yield of
important crops in Austria and Bulgaria and phenology and yield, estimated with the CERES,
CROPGRO and WOFOST models. Overall, the model test results were very good. The results
obtained indicated a satisfactory performance of the applied simulation models for winter
wheat, spring barley, maize and soybean in the selected environments.
It is necessary to emphasize, however, that the above crop-weather models embody a
number of simplifications and limitations. For example, many physiological processes and
back
their response to diurnal weather conditions are not simulated on hourly base but using daily
time step input weather data. The impacts of weeds, diseases, and insect pests on crop growth,
development and final yield formation were also not assumed within the study. These
limitations might be related to some of the high deviations between the model crop
simulations and real measured crop data.
8000
Winter grain yield (kg/ha)
simulated
measured
6000
4000
2000
1980
1982
1984
1986
Year
1988
1990
1992
Fig. 28. Variations of simulated and measured grain yield of maize
at station Pavlikeni (North Bulgaria) during the periods of calibration (1984-1990)
and verification (1980-1983 and 1991-1993)
Crop-weather models can play an important role at different levels of applications,
ranging from decision support for crop management at a farm level to advancing
understanding of sciences at a research level. The main goal of most applications is to predict
final yield in the form of either grain yield, fruit yield, root or tuber yield, biomass yield for
fodder, or any other harvestable product (Hoogenboom, 2000). Certain applications link the
price of the harvestable product with the cost of inputs and production to determine economic
returns.
According to Hoogenboom (2000), the management applications of crop-weather
models can be defined as strategic applications, tactical applications, and forecasting
applications. In strategic applications, the crop models are run prior to planting of a crop to
evaluate alternative management strategies. In tactical applications, the crop models are run
before planting or during the actual growing season. Both strategic and tactical applications
provide information for decision making by either a farmer, consultant, policy maker, or other
person involved directly with agricultural management and production. Forecasting
applications can be conducted either prior to planting of a crop or during the growing season.
The main objective is to predict yield; this information can be used at a farm-level for
marketing decisions or at a government level for policy issues and food security decisions.
The models applied within the study can be used for within-year crop decisions, multiyear risk analysis for strategic planning, crop yield prediction, basic economic assessments of
agricultural cereal production and definition of research needs in Austria and Bulgaria and
surrounding countries in Central and Eastern Europe. For example, these models have
potential for generating forecasts of environmental yields in advance of harvest or maturity, as
well as at the time of harvest. In this case, online assessments of current agrometeorological
conditions and expected yield can be done. Expected crop yields can be predicted some time
back
before harvest by using climate or expected weather data. Because long-range weather
forecasts are not yet reliable, predicting weather is not yet very exact. However, using a range
of reasonable weather patterns, a “fork” of yield expectations can be determined. Such
estimates will be of value to growers seeking to sell their crops, to the transportation industry
in planning to move the crop, and to national governments estimating the effects of
production on future prices.
Once the CERES, CROPGRO and WOFOST crop-weather models have been adapted
for the environmental conditions in the selected regions in Austria and Bulgaria, numerous
computer experiments can be run. These experiments can involve an assessment and
comparison of new and traditional varieties and their response to different fertilizer and
irrigation regimes, to different soil types and climatic conditions (e.g. climate change
impacts). The major goal of these simulations will be determination of the optimum crop
management practices, necessary to obtain high crop yields and gains.
References
Aggarwal, P.K., 1995. Uncertainties in crop, soil and weather inputs used in growth models - implications for
simulated outputs and their applications. Agricultural Systems, 48 (3), 361-384.
Baier, W., 1979. Note on the terminology of crop-weather models.Agric. Meteorol., 20, 137–145.
Boogard, H., C. van Diepen, R. Roetter, J. Cabrera and H van Laar, 1998. WOFOST 7.1 User’s guide for the
WOFOST 7.1 crop growth simulation model and WOFOST control Center 1.5. SC-DLO, IRRI,
Wageningen, 144 pp.
Chirkov, U., 1969. Agrometeorological Conditions and Maize Productivity. Gidrometeoizdat, Leningrad (in
Russian).
Eitzinger, J., Z. Žalud, C.A. van Diepen, M. Trnka, D. Semerádová, M. Dubrovský, and M. Oberforster, 2000.
Calibration and evaluation of the WOFOST model for winter wheat. 8th International Poster day '
Transport of Water, Chemicals and Energy in the System Soil-Crop Canopy-Atmosphere' 16.11.2000,
Bratislava. (CD version, ISBN 80-968480-0-3, Institute of Hydrology, Slovak Academy of Sciences).
Himeblau, D., 1975. Applied non-linear programming. Moscow, 123 pp. (in Russian).
Hoogenboom, G., 2000. Contribution of agrometeorology to the simulation of crop production and its
application. Agricultural and Forest Meteorology, 103, 137-157.
Hoogenboom, G, P. Wilkens and G. Tsuji, 1999. DSSAT v3, Vol.4. University of Hawaii, Honolulu, Hawaii.
International Benchmark Sites Network for Agrotechnology Transfer (IBSNAT), 1993. The IBSNAT Decade.
University of Hawaii, Honolulu.
Penning de Vries, F.W.T., D.M. Jansen, H.F.M ten Berge and A. Bakema, 1989. Simulation of Ecophysiological
Processes of Growth in Several Annual Crops. Pudoc, Wageningen, The Netherlands, 270 pp.
Procerov, A. and E. Ulanova, 1961. Using Methods of Agrometeorological Prognoses in the USSR.
Gidrometeoizdat, Moscow, 156 p. (in Russian)
Slavov, N. and N. Vitanov, 1977. Evaluation of the methods for agrometeorological prediction of maize growth
in Bulgaria. Agricultural Meteorology, Vol. 1, GUHM, Sofia, pp. 53-64 (in Bulgarian).
Supit, I., A.A. Hooijer and C.A. van Diepen, 1994. System Description of Wofost 6.0 Crop Simulation Model
implemented in CGMs. European Commission Joint Research Centre, Luxembourg, pp 144
Tsuji, G., G. Hoogenboom and P. Thornton, 1998. Understanding Options for Agricultural Production. Kluwer
Acad. Publ., 399 pp.
Tsuji, G., G. Uehara and S. Balas, (eds.), 1994. DSSAT version 3, Vol. 1-3, IBSNAT, University of Hawaii,
Honolulu.
Van Diepen, C.A., J. Wolf, H. van Keulen and C. Rappolt, 1989. WOFOST: a simulation model of crop
production. Soil Use Manage, 5, 16-24.