2 A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12
tant in order to assess effects of climatic variability. For the most commonly used models, this requires
long term daily values of rainfall, solar radiation, max- imum and minimum temperatures. However, many lo-
cations do not have sufficient historical weather data. Many workers have recognized this problem, and this
has led to the development of a range of weather generators such as WGEN Richarsdon and Wright,
1984, SIMMETEO Geng et al., 1986, 1988, TAM- SIM McCaskill, 1990 and others e.g. Larsen and
Pense, 1982; Bristow and Campbell, 1984; Guenni et al., 1991 which can be used to generate weather
data sets.
The quality of model output is related to the quality of weather data used as input. It follows that the test-
ing of the sensivity of model output to the quality of generated weather data is an essential prerequisite for
simulation analyses. Weather data is used in crop mod- eling not only to predict crop growth and development
in response to environmental variables but also to flag catastrophic events such as heat stress, water deficit,
frost damage and etc. Often such effects can only be assessed by using the weather data as input into the
simulation models in question because model act as a data filter and integrate the effect of deviations of
generated from actual data Richardson, 1984; Meinke et al., 1995. Meinke et al. 1995 demonstrated how
crop simulation models can be used to assess the ad- equacy and quality of weather data generation. They
showed that model complexity did not appear to influ- ence model sensivity to differences in environmental
input data.
WGEN is a well-known and widely used stochastic weather generator that requires a number of parame-
ters to generate a weather series at a site. This genera- tor has been incorporated into the WeatherMan short
for weather data manager, Pickering et al., 1994, an application program of Decision Support System for
Agrotechnology Transfer DSSAT software Tsuji et al., 1994. DSSAT is a collection of crop models and
computer programs integrated into a single software package in order to facilitate the application of crop
simulation models in research and decision making. This software is a product of the International Bench-
mark Sites Network for Agrotechnology Transfer IB- SNAT project and is distributed over the world.
Richardson 1984 and Meinke et al. 1995 tested the WGEN output applied to crop simulation models
and showed that yields obtained using generated data were not significantly different from that obtained us-
ing actual data. In these tests, parameters required by the model were obtained from long term actual data
longer than 10 years. However, we have found no reports showing the capability of WGEN for gener-
ating long term data when model parameters are de- rived from relatively short term weather data, say 3–10
years. Thus, the objective of this study was the evalua- tion of the WGEN model for generating long weather
series parameterized using short 3–10 years records of actual data to derive a specific crop model.
2. Materials and methods
Chickpea Cicer arietinum L. is the most impor- tant legume crop of Iran, and is sown on more than
50 of the total legume areas. The crop is predom- inately rainfed and only 10 is grown under irriga-
tion Sadri and Banai, 1996. Most chickpea growing areas of Iran which have cool and cold semiarid cli-
mates with terminal drought stress similar to that of Tabriz, the location chosen for this study are located
in north west NW of the country. Nearly 50 of to- tal chickpea production in Iran is concentrated in NW.
Recently, a project was initiated to analyse the bio- physical limitations in chickpea production in Iran us-
ing the crop simulation approach. However, at many locations long term weather data more than 10 years
are not available to be used in the analysis. Therefore, we are forced to use a weather data generator to pro-
vide long term weather series for use with the crop model.
2.1. The weather generation model The WGEN model provides daily values of precipi-
tation, maximum and minimum temperatures and solar radiation for a n-year period at a given location. The
precipitation is generated independently for a given day. Maximum and minimum temperature and solar
radiation are then generated depending on whether the generated day is wet or dry. The precipitation compo-
nent of WGEN is a Markov chain-gamma distribution model. The occurrence of wet or dry days is gener-
ated with a first-order Markov chain model in which the probability of rain on a given day is conditioned
A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 3
on whether the previous day was wet or dry. When a wet day is generated, the two-parameter gamma dis-
tribution is used to generate the precipitation amount. For generating daily values of maximum and mini-
mum temperature and solar radiation, the residual of the three variables are generated using a multivariate
normal generation procedure that preserves the serial correlation and cross correlation of the variables. The
final values of the three variables are determined by adding the seasonal means and standard deviations to
the generated residual elements. For more details re- fer to Richardson 1981, 1984 and Richarsdon and
Wright 1984.
2.2. The crop growth model To evaluate the sensivity of crop models to observed
versus generated data, we used the chickpea model of Soltani et al. 1999 with each of the observed and
generated weather files. This model is similar to the models of Sinclair 1986, Amir and Sinclair 1991,
and Hammer et al. 1995. The model is based on a daily time step and simulate crop growth and devel-
opment as a function of temperature, solar radiation and water availability. Crop phenology is divided into
two growth stages before and after beginning seed growth, the duration of which are predicted based on
daily temperature and water deficit. Leaf area devel- opment is calculated as functions of expansion and
senescence of leaves. These functions are sensitive to temperature and water deficit. Daily biomass produc-
tion is predicted from leaf area index, light extinction coefficient and radiation use efficiency. Transpiration
is calculated as a function of daily biomass produc- tion, transpiration efficiency coefficient and daily va-
por pressure deficit. Daily biomass production and leaf growth decrease below potential values once the frac-
tion of transpirable soil water FTSW declines below a threshold value about 0.37. After the beginning of
seed growth the accumulated biomass is partitioned into the grains, the rate of which depending on cli-
matic conditions at and after beginning of seed growth. Soil evaporation, soil water drainage, and runoff are
also calculated in the soil water balance sub-model. The model uses readily available weather and soil in-
formation. The model was tested using independent data from a range of Iran’s environmental conditions.
In most cases, simulated grain yields were similar to that of observed yields. At this stage the model does
not account for the effects of pests, diseases and soil fertility.
2.3. Procedure Tabriz 38
◦
8
′
N, 46
◦
17
′
E and 1361 m above sea level was selected as the site for comparison of chick-
pea model yield estimates made with actual weather data and data generated by WGEN as implemented
in WeatherMan Pickering et al., 1994. Precipitation and maximum and minimum temperature data were
available for 30 years 1965–1995. Solar radiation data were calculated from sunshine hours and ex-
traterrestrial radiation as outlined by Doorenbos and Pruitt 1977. With solar radiation data estimated,
a complete 30-year daily data set of the required weather variables were available.
Four 100-year samples of weather data were gener- ated for Tabriz. The parameters used to generate data
were obtained from daily weather data of 3, 5, 7, and 10 recent years as base periods W3, W5, W7, and
W10 thereafter. Monthly means of rainfall, solar ra- diation, maximum temperature and minimum temper-
ature for 3, 5, 7 and 10 recent years are shown in Table 1. The actual and generated weather series were
each used as input to the chickpea model. The same soil, cultivar and management inputs were used for all
simulations. The soil that was chosen was sandy loam with a plant available water of 0.13 m
3
m
− 3
, a depth of 100 cm and a curve number of 80. Curve number
of soil is required to estimate run off using the curve number technique Knisel, 1980. Curve numbers are
assigned to various soil categories depending on their texture and condition and then modified to take ac-
count of standing crop cover. The chickpea cultivar Jam was chosen for simulations. Characteristics of this
variety are given in Soltani et al. 1999. Simulations were run under irrigated and rainfed conditions with
three sowing dates — 26 March, 10 and 25 April. The selected sowing dates do not necessarily reflect
common practices, but were rather selected to trigger chickpea model responses to stressfull conditions to
enable comparisons of model sensivity to weather data generation and to test the base periods used for param-
eter estimation at various times of the year. Planting
4 A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12
Table 1 Comparison of monthly rainfall amount, solar radiation, maximum and minimum temperatures averaged over recent 3, 5, 7 and 10 year
periods Month
Rainfall Solar radiation
amount mm MJ m
− 2
d
− 1
3 5
7 10
3 5
7 10
1 24.8
18.6 15.5
15.5 7.8
8.4 8.4
8.4 2
8.7 14.5
11.6 17.4
9.6 9.8
9.9 9.7
3 46.5
37.2 34.1
34.1 13.9
13.9 13.9
13.7 4
66.0 54.0
48.0 45.0
17.8 19.2
18.9 18.7
5 49.6
49.6 40.3
37.2 23.8
23.2 23.7
23.9 6
21.0 21.0
15.0 18.0
26.4 26.7
26.9 27.0
7 3.1
6.2 3.1
3.1 27.5
27.5 27.8
27.6 8
3.1 3.1
3.1 6.2
24.7 24.4
24.8 25.3
9 6.0
3.0 3.0
3.0 20.7
20.5 20.4
20.8 10
18.6 12.4
21.7 24.8
15.0 15.2
14.7 14.7
11 36.0
36.0 30.0
27.0 9.0
9.4 10.0
10.1 12
18.6 24.8
21.7 31.0
8.9 8.1
8.0 7.8
Annual 302.0
280.4 247.1
262.3 17.1
17.2 17.3
17.3 Maximum
Minimum temperature
◦
C temperature
◦
C 1
2.6 2.4
1.6 2.3
− 4.7
− 5.6
− 6.9
− 6.4
2 5.6
5.0 3.9
4.6 −
2.9 −
3.8 −
4.8 −
3.9 3
11.0 9.8
10.1 10.0
0.7 0.3
0.4 0.4
4 17.3
17.2 17.6
17.3 6.6
6.5 6.6
6.5 5
22.4 21.4
22.2 22.5
10.9 10.3
10.5 10.6
6 27.8
27.8 28.6
28.6 15.3
15.4 15.8
15.7 7
32.1 32.5
33.0 33.0
19.2 19.4
19.9 19.9
8 32.8
32.3 32.4
32.2 19.6
19.3 19.3
19.1 9
28.0 27.9
28.2 28.3
14.7 14.6
14.5 14.5
10 20.5
20.8 20.8
20.2 8.3
8.5 8.6
8.5 11
10.6 11.3
12.1 11.8
1.9 2.3
2.7 2.4
12 3.6
3.6 4.5
4.7 −
3.5 −
3.1 −
2.8 −
2.8 Annual
17.9 17.7
18.0 18.0
7.2 7.1
7.1 7.1
densities were 25 and 50 plants per m
2
under rainfed and irrigated conditions, respectively. Under irrigated
conditions, the water balance sub-model was turned off. The management inputs were the same for each
year of crop growth model runs. Thirty year simula- tions were made with the actual data and 100 years
with the four series of the generated weather data. The daily water balance only for rainfed conditions and
crop growth and yield were simulated for each year.
In each case, generated data were compared to the related base period data and to the entire 30-year
length of record. The former tests indicate whether WGEN assumptions are valid and the latter tests show
whether the base periods 3–10 are of sufficient length to adequately represent the entire data set. t-Tests were
used to compare the means of precipitation, number of wet days, solar radiation, maximum temperature, min-
imum temperature, number of days with a maximum temperature greater than 35
◦
C and number of days with a minimum temperature less than 0
◦
C for each month. Means of simulated yields and some other
crop related variables using actual and four series of generated data were also compared by t-tests.
Cumulative distribution functions CDFs of each weather variable and simulated grain yield were
calculated. In the case of grain yield, data from different sowing dates were combined into one
CDF. The CDFs were then compared with CDFs of actual weather data or simulated grain yield us-
ing actual weather data and tested for significant
A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 5
differences using Kolmogorov–Smirnov test. The Kolmogorov–Smirnov statistic is simply the max-
imum absolute difference between the cumulative probability distributions of the two samples in ques-
tion. The smaller the calculated test statistic, the smaller are the differences between the distributions
in question. Statistical comparisons were performed using SAS SAS Institute, 1989.
3. Results and discussion
