approach using a dynamic model of crop growth. The model must be conceptually appropriate for
the research in hand and it must have input requirements which can be met and must give
reasonable predictions. Models are presently available for many crops, but these have generally
been developed in temperate locations.
We chose the CERES-Maize model Jones and Kiniry, 1986, because we thought it appropriate
for fulfilling these criteria. It was tested under temperate conditions characterised by the absence
of water stress in North America Hodges et al., 1987; Piper and Weiss, 1990 and in Europe,
precisely in Belgium Lahrouni et al., 1993, in France Lorgeou, 1991; Plantureux et al., 1991
and in Germany Entenmann et al., 1989. To our knowledge, the model has not been tested under
Mediterranean
conditions Ruget
and Bon-
homme, 1991, characterised by both atmospheric and soil drought Hamdy and Lacirignola, 1999.
Results of these tests show some limitations in the use of CERES-Maize. Carberry et al. 1989
tested the model in semi-arid regions of tropical Australia and proposed new modules for phenol-
ogy, leaf development and increase of biomass. Phenological module is not precise in the case of
water excess in the soil Hodges et al., 1987; Carberry et al., 1989 or in the case of cold
seasons Liu et al., 1989. Finally, Carberry 1991 underlines that water stress does not have any
effect on phenology and proposes a correction. Lahrouni and Ledent 1991 show the conse-
quences of biased estimations of phenological stage duration on leaf area and dry matter parti-
tioning simulated by the CERES-Maize. Finally, grain yield and its components can be overesti-
mated or underestimated by the model in rainy or dry seasons, respectively Wu et al., 1989.
The objective of this study was to compare the predictions of CERES-Maize with observed data
from field experiments, in which the moisture regime varied, but nitrogen nutrition was always
adequate. The results of this study allow us to evaluate model performance under contrasting
soil water conditions in a Mediterranean environ- ment. Moreover, this study allows us to identify
model subroutines that should be modified to improve to fit better in the observed Mediter-
ranean conditions.
2. Materials and methods
2
.
1
. The field experiment
2
.
1
.
1
. Site and climate Field studies examining the response of maize
cv Maltus to water deficit were made at Rutigliano on the experimental farm of the Isti-
tuto Sperimentale Agronomico latitude 41°01N, longitude 17°1724¦E and altitude 122 m a.s.l.
during 1996 and 1997.
The Rutigliano climate is characterised by warm dry summers, with maximum air tempera-
ture sometimes higher than 40°C, and minimum relative air humidity often less than 20. Mean
annual rainfall is 600 mm, almost exclusively con- centrated in spring and autumn. Fig. 1 shows the
Fig. 1. Maximum and minimum air temperature; solar radia- tion, and precipitation columns: empty, 1996; solid, 1997
measured at the agrometeorological station of Rutighano.
time course of maximum and minimum tempera- ture, solar radiation, and precipitation measured
during the cropping seasons 1996 and 1997 at the agrometeorological station located about 50 m
from the experimental site. The year 1997 was sunnier and warmer than 1996, especially early in
the crop season. For both the years the central period of growth was characterised by an ex-
tended drought except for a short rain in 1997. Precipitation was more frequent and heavier at
the end of the 1996 crop season.
The soil is silty clay loam, a well-drained red earth or ‘Alfic Xerarent Mixed Thermic Fine’
USDA Soil Taxonomy. The soil profile is shal- low up to 0.6 m in depth, and consequently
water availability is low about 110 mm, calcu- lated between the field capacity and the wilting
point, so irrigation is necessary during the maize crop season.
2
.
1
.
2
. Crop management and growth analysis Crops were sown on 24th May 1996 and on
27th May 1997 144 and 147th day of the year, respectively in rows 60 cm apart and at a rate of
10 seeds m
− 2
. The final density was 5 plants m
− 2
. All the crops were grown under high-input condi-
tions 120 kg P
2
O
5
ha
− 1
before sowing and 100 kg ha
− 1
of nitrogen as ammonium nitrate 26 N, in two split applications, the first one early in
the crop season and the second one at the jointing stage.
The experimental design was a randomised block replicated three times nine plots in total.
Each plot was 9 m long and 9.6 m wide. Dates of the main phenological stages were
collected during the 2 years of trial and included: 1 seedling emergence; 2 fourth leaf collar visi-
ble; 3 tenth leaf collar visible; 4 50 tassel- ing; 5 milky maturity; 6 dough maturity; 7
maturity.
Leaf area
and above-ground
dry biomass were determined at each phenological
stage by sampling the plants over a 1.5 m
2
area per plot. Leaf area was measured with a LiCor
1300 leaf metre and dry biomass, separated into different plant parts, was measured after drying
at 80°C for 48 h in an oven.
2
.
1
.
3
. Irrigation scheduling and experimental design
Irrigation scheduling tried to produce three dif- ferent soil water conditions. The adopted method-
ology is based on plant water relationships. It was observed that in maize, stomatal conductance
does not limit gas exchange when the pre-dawn leaf water potential C is higher than −0.3
MPa, but if C becomes more negative than this threshold value, maize stomata tend to close
Katerji and Bethenod, 1997. Moreover if C decreases to − 1.5 MPa, stomatal conductance
values do not change. Therefore, − 0.3 MPa rep- resents a threshold value, we chose to separate no
water stress from stress conditions.
During each trial year, plants were well-watered until leaf area index LAI was just higher than 1.
Thereafter, three water treatments were imposed: full irrigation IRR, moderate water stress
STR1 and severe stress STR2. Irrigation was scheduled with a low-pressure system drip irriga-
tion, whenever C, measured daily, equalled − 0.3 MPa IRR, − 0.6 MPa STRI and − 1.2
MPa STR2, respectively. The pre-dawn leaf wa- ter potential C was measured on the last devel-
oped leaves before sunrise. Five leaves per plot were harvested from the three treatments and the
water potential was measured with a pressure chamber Scholander et al., 1965. C is not re-
quired for testing CERES-Maize, but it is re- quired for providing the protocol for irrigation
scheduling, and C is necessary for calibrating the new versions of the model.
2
.
1
.
4
. E6apotranspiration estimation Evapotranspiration was estimated by using a
different simplified soil water balance approach. At our site, runoff and capillary rise can be
neglected, owing to the flat ground and to the presence of a cracked rocky layer which limits soil
depth and ascending water. The equation for soil water balance can be expressed as
ET = P + DS
w
− D
r
1 where; ET, is crop evapotranspiration mm; P,
precipitation or irrigation mm; DS
w
, the differ- ence in soil water content, measured with time
domain reflectometry TDR, Tektronic, model
Table 1 Input data required by CERES-Maize model
Acronym Parameter or variable
Units Location data
Latitude Degrees
LAT Planting data
cm SDEPTH
Sowing depth Plant population
Plant m
− 2
PLANTS Climatic data
Day JDATE
Day of the year
Maximum temperature °C
TEMPMAX TEMPMN
Minimum temperature °C
MJ m
− 2
SOLRAD Solar radiation
RAIN Rainfall
mm Culti6ar data
Thermal time from °C
P1 emergence to end of
juvenile stage P2
Photoperiod sensitivity Day h
− 1
coefficient Thermal time from silking
°C P5
to physiological maturity
Potential kernel number Kernel per
G2 plant
Potential kernel growth G3
mg per kernel per day
rate Soil data
Soil albedo SALB
U mm per day
Stage-1 soil evaporation coefficient
Whole-profile drainage SWCON
cm per day rate coefficient
Runoff curve number CN2
– LL
cm
3
cm
− 3
Lower limit of soil water content
DUL Drained upper limit
cm
3
cm
− 3
cm
3
cm
− 3
SAT Saturated water-content
WR Root growth factor
–
showed that ET estimates were similar to those provided by the Bowen ratio method differences
between the two methods were less than 10. The Bowen ratio method has been thoroughly
tested and its validity as a reference evapotranspi- ration measurement has been well established
Fuchs and Tanner, 1970; Sinclair et al., 1975, even for high canopies Rana and Katerji, 1998.
2
.
2
. CERES-Maize model CERES-Maize is a deterministic simulation
model, designed to simulate the effects of cultivar, planting density, weather, soil water and, in one
of the versions, nitrogen on crop growth, develop- ment and yield Jones and Kiniry, 1986. To
simulate accurately maize growth, development and yield, the model takes into account the fol-
lowing processes.
Phenological development, especially as it is affected
by genetics,
temperature, and
photoperiod.
Leaf area development and growth of stems and roots.
Biomass accumulation and partitioning to grains and other organs.
Soil water balance and water used by the crop.
Soil nitrogen uptake and its partitioning among plant organs.
CERES standard version V3.0, used in our study, assumes that nitrogen supply is not limit-
ing. The input parameters needed to run CERES- Maize are listed in Table 1.
2
.
2
.
1
. Model calibration Data from treatment IRR, collected during the
1996 season, were used to calibrate CERES- Maize. The remainder of the data was used for
validation of the model IRR in 1997; STR1 and STR2 in both the years. This procedure allowed
us to validate the model using experimental data that are effectively independent from those used
for its calibration.
Thermal time from seedling emergence to the end of juvenile stage P1 and the photoperiod
sensitivity coefficient P2 were obtained by cali- bration based on observed silting date, according
to an approach presented by Ritchie and Aloga- 1502 C in two layers, 0 – 30 and 30 – 60 cm, in
only one block and with two replicates per plot. For the calculation of volumetric soil water con-
tent a local calibration curve was used. The drainage D
r
was estimated as the amount of water exceeding maximum water capacity in the
whole soil profile. This approach to ET estimation was validated Mastrorilli et al., 1998, and
raswamy 1989. Potential kernel number G2 and growth rate G3 were found by calibration
of kernel weight and kernel number. The duration of grain-filling P5 was the value observed in the
data-set used for calibration IRR 96. Values for the lower, drained upper and saturated limits LL,
DUL, SAT were based on those measured in soil samples collected in an independent trial Castri-
gnano` et al., 1994.
2
.
3
. Model 6alidation The accuracy of model predictions of crop de-
velopment and growth LAI and above-ground biomass and grain yield was estimated using two
statistical procedures. The first consisted of a linear regression between measured and predicted
values on all the observation dates for each of the three variables under study. Two Student’s t-tests
were then applied to verify the following two ‘null’ hypotheses: intercept = 0 and slope = 1.
The second procedure followed the methodol- ogy proposed by Addiscott and Whitmore 1987
and Whitmore 1991 and consisted in partition- ing the sum of the squares of the residuals into
pure error i.e. random variation and lack of fit i.e. systematic variation.
The accuracy of maize evapotranspiration sim- ulations was tested by linear regression between
ET values provided by the model and ET esti- mates from the simplified soil water balance, pre-
sented above,
and based
on soil
water measurements with the TDR. Also for the ET
validation, estimates from IRR 96 were not considered.
3. Results and discussion