Annals of Botany 83 : 413422, 1999

Article No. anbo.1998.0832, available online at http :\\www.idealibrary.com.on

Evaluation of a Dynamic Simulation Model for Tomato Crop Growth and

Development
E. H E U V E L I N K
Wageningen Agricultural Uniersity, Department of Horticulture, Haagsteeg 3, 6708 PM Wageningen,
The Netherlands
Received : 31 July 1998

Returned for revision : 21 September 1998

Accepted : 19 December 1998

A dynamic simulation model for tomato crop growth and development, TOMSIM, is evaluated. Potential crop
growth and daily crop gross assimilation rate (Pgc, d) is computed by integration of leaf assimilation rates over total
crop leaf area throughout the day. Crop growth results from Pgc, d minus maintenance respiration rate (Rm),
multiplied by the conversion efficiency. Dry matter distribution is simulated, based on the sink strength of the plant
organs, which is quantified by their potential growth rate. Within the plant, individual fruit trusses and vegetative
units (three leaves and stem internodes between two trusses) are distinguished. Sink strength of a truss or a vegetative
unit is described as a function of its developmental stage. In this paper, emphasis is on the interactions between the
two submodels of, respectively, dry matter production and dry matter distribution. Sensitivity analysis showed that
global radiation, CO concentration, specific leaf area (SLA) and the developmental stage of a vegetative unit at leaf
#
pruning had a large influence on crop growth rate, whereas temperature, number of fruits per truss, sink strength of
a vegetative unit and plant density were less important. Leaf area index (LAI) was very sensitive to SLA and the
developmental stage of a vegetative unit at leaf pruning. Temperature did not influence the simulated Rm, as increased
respiration rate per unit of biomass at higher temperatures was compensated by a decrease in biomass. The model
was validated for four glasshouse experiments with plant density and fruit pruning treatments, and on data from two
commercially grown crops. In general, measured and simulated crop growth rates from 1 month after planting
onwards agreed reasonably well, average overestimation being 12 %. However, crop growth rates in the first month
after planting were overestimated by 52 % on average. Final crop dry mass was overestimated by 031 %, due to
inaccurate simulation of LAI, resulting partly from inaccurate SLA prediction, which is especially important at low
plant density and in a young crop.
# 1999 Annals of Botany Company
Key words : Crop growth, dry matter production, glasshouse, leaf area, Lycopersicon esculentum, partitioning,
simulation model, tomato, TOMSIM.

INTRODUCTION
Crop growth simulation models are tools that enable us to
integrate knowledge about crop growth, to test hypotheses
about how different parts of the system interact, and to
develop understanding about the system as a whole (Acock
and Reynolds, 1989). Although crop models have great
potential for practical use in agriculture in general and in
horticulture in particular, their use is still limited (Challa,
1990). One reason is the poor validation of most models,
especially at the level of the different submodels combined
with a validation of the model as a whole.
Before a model can be used it must be validated, i.e.
model output, after running the model on historical input
data recorded for the real system, has to be compared with
the real system output. Models are often validated with all
or some of the data used for model development or
calibration (e.g. Jones et al., 1991 ; Bertin and Gary, 1993),
whereas independent data, not used in model development,
should be used (McCarl, 1984). Usually, model validation is
limited to a small range of conditions only ; for example,
Jones et al. (1991), Dayan et al. (1993) and De Koning
(1994) developed models for dynamic simulation of dry
matter partitioning in tomato, but the treatments in their
validation experiments did not create a sufficient difference
0305-7364\99\040413j10 $30.00\0

in partitioning. Furthermore, in model validation the

possibility of compensating errors in the different modules is
not always recognized (e.g. Kano and Van Bavel, 1988).
Such validations do not test the general validity of the
model.
A mechanistic crop growth model for glasshouse tomato
(TOMSIM) has been developed (Heuvelink, 1995 a ; 1996 a),
and the following of its submodels (modules) validated : (1)
greenhouse transmissivity (Heuvelink, Batta and Damen,
1995) ; (2) photosynthesis (Heuvelink, 1996 b) ; (3) dry
matter production (Bertin and Heuvelink, 1993 ; Heuvelink,
1995 a) ; and (4) truss appearance rate, fruit growth period
and dry matter partitioning (Heuvelink and Bertin, 1994 ;
Heuvelink, 1996 a). Sensitivity analyses for the modules
for dry matter production and dry matter distribution
were presented previously (Bertin and Heuvelink, 1993 ;
Heuvelink and Bertin, 1994, respectively).
The aim of the present study is to evaluate the aggregated
TOMSIM model. The importance of the different parameters and climatic factors in the model is assessed with a
sensitivity analysis. Emphasis is on the interactions between
dry matter production and dry matter distribution. Validation is conducted on experiments which are most sensitive
to these interactions. Specific model inputs needed are : the
starting date for the simulation (planting date of the crop) ;
# 1999 Annals of Botany Company

414

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment

the initial dry mass and the flowering date of the first truss ;
the number of fruits per truss ; and climatic data. Contrary
to the validation of the modules separately, interactions
among the submodels then become manifest. To analyse
possible discrepancies, results are also compared with
simulations where leaf area index (LAI), dry matter
distribution and biomass present (i.e. above-ground crop
mass, not including leaves and fruits already removed) are
input into the model.
MODEL DESCRIPTION

Source/sink
+ ratio

Number of
+ fruits
Pnc
+

Wt

Dry matter production

This module has been described (Heuvelink, 1995 a) ; it
calculates potential crop growth, and daily crop gross
assimilation rate is computed by integration of leaf
assimilation rates over total crop leaf area and over the day.
The module assumes a uniform crop canopy and photosynthetic characteristics of single leaves are assumed to be
identical to all leaves in the canopy. Ground reflectance of
radiation was taken into account as described by Gijzen
(1992).
Daily dry matter (DM) production is calculated according
to :
(1)
dW\dt l Cf(Pgc, dkRm)
with dW\dt the crop growth rate (g DM m# d"), Cf the
conversion efficiency from assimilates to dry matter
(g DM g" CH O), Pgc, d the crop gross assimilation rate per
#
unit ground area (g CH O m# d") and Rm the maintenance
#
respiration rate per unit ground area (g CH O m# d").
#
Rm is calculated from the dry mass of the plant parts
multiplied by their specific maintenance coefficients and
depends on temperature (Spitters, Van Keulen and Van
Kraalingen, 1989). However, it is clear that specific
maintenance respiration coefficients are not constant, but
depend on the metabolic activity of the crop (Amthor,
1989). Mean relative crop growth rate was used as a
measure for metabolic activity according to Heuvelink
(1995 a) :
(2)
Rm l Rm(1kefRGR)
in which Rm is the maintenance respiration rate
(g CH O m# d"), Rm the maximum maintenance respir#
ation rate (g CH O m# d"), f is a coefficient (33 d ;
#
Heuvelink, 1995 a) and RGR is the simulated relative
growth rate of the crop (d"), averaged over the preceding
5 d.
The module runs on an hourly basis. Parameter values are
equal to those used by Bertin and Heuvelink (1993) and
Heuvelink (1995 a), except that Bertin and Heuvelink (1993)
modelled Rm independently of mean relative crop growth
rate [ f l _ in eqn (2)].

Wveg
LAI

+
+

SLA

F. 1. A simplified representation of two important interactions

(feedback mechanisms) between dry matter production and dry matter
partitioning in an indeterminate tomato crop : (j) indicates positive
influence, (k) indicates negative influence. Solid line represents C flow,
dashed lines represent information flow. LAI, Leaf area index ; Pnc,
crop net assimilation rate ; SLA, specific leaf area ; Wt, total crop dry
mass ; Wveg, vegetative crop dry mass.

individual fruit trusses and vegetative units (three leaves and

stem internodes between two trusses) are distinguished. The
appearance rate and harvest rate of trusses depend on
temperature only. The vegetative unit starts to grow 3 weeks
before the corresponding truss and its growth period is
assumed to be equal to that of the corresponding truss. Note
that the vegetative sink strength is not a constant, as in the
simple version of the module (Heuvelink, 1996 a).
All sinks derive their assimilates for growth from one
common assimilate pool. Daily available biomass is distributed among the total number of sinks per plant,
according to their relative sink strength, which is defined as
their potential growth rate, relative to the total sink strength
of all sinks together. Sink strength of a truss is equal to the
potential growth rate of a tomato in the truss multiplied by
the number of fruits on a truss, the latter being an input to
the model. Root sink strength was set at 15 % of aboveground vegetative sink strength. Partitioning within the
vegetative plant part is 7 : 3 : 1n5 for leaves, stem and roots,
respectively.
The module runs on a daily basis. When the available
biomass equals or exceeds the total sink strength, each sink
organ will grow at its potential rate. The assimilates not
used for growth are stored as reserves. The next day these
reserves are added to the newly formed assimilates. Model
parameters and initial settings (e.g. the developmental stage
of vegetative units at first anthesis) are the same as in the
validation of the dry matter distribution module (Heuvelink,
1996 a). The model starts at anthesis of the first truss.

Dry matter partitioning

In this module, based on Heuvelink and Marcelis (1989)
and De Koning (1994), and described and validated as a
more detailed approach by Heuvelink (1996 a), dry matter
distribution is primarily regulated by the sinks, and

Linking dry matter production and dry matter distribution

Two important interactions (feedback mechanisms) between dry matter production and dry matter distribution in
tomato can be distinguished (Fig. 1) : (a) flower and\or fruit

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment

415

T 1. Details of model alidation experiments conducted in glasshouses

Experiment

Year

Planting
date

1a

1992

2 March

2b

1992

3c
4d

1989
1991

14 April

7 March
7 June

Treatments
Three plant densities :
3n1, 2n1 and 1n6 plants m#
Fruit pruning in combination with plant
density :
3 fruits\truss and 4n8 plants m#
5 fruits\truss and 2n9 plants m#
7 fruits\truss and 2n1 plants m#
Fruit pruning : 3 or 7 fruits\truss
No truss pruning (control) or every other
truss removed at anthesis

Average 24 h
temperature
(mC)

Average global
radiation outside
(MJ m# d")

20n4

13n7

21n2

17n5

20n0
23n2

15n3
15n3

Reported by : a Heuvelink (1995 a) : Expt 13 ; b Heuvelink (1995 a) : Expt 14 ; c Heuvelink and Buiskool (1995) : Expt 2 ; d Heuvelink and Buiskool
(1995) : Expt 5.

abortion at a low source\sink ratio (De Koning, 1994 ;

Bertin, 1995), resulting in fewer fruits on the plant and
hence decreased sink strength and increased source\sink
ratio ; and (b) partitioning to the vegetative parts determining
LAI and hence future light interception and dry matter
production. Note that in TOMSIM dry matter distribution
is fully independent of dry matter production ; indirect
influence via flower and\or fruit abortion is not modelled.
Linking both modules results in some changes, compared
to the separate validation of the modules (Heuvelink,
1995 a, 1996 a). Assimilate requirements (Cf) are calculated
based on simulated dry matter allocation. Simulated rather
than measured dry mass, based on simulated crop growth,
simulated harvest of fruits and simulated pruning of leaves,
is used to calculate Rm.
Leaf area is simulated, based on simulated leaf dry mass
and specific leaf area (Fig. 1). Leaf dry mass results from
leaf growth and pruning of leaves. Leaves are removed
when the developmental stage of the corresponding truss is
0n9 (at 20 mC trusses grow for 60 d and thus leaves are
removed 6 d before the corresponding truss is harvest-ripe).
In practice, specific leaf area (SLA ; cm# g") depends on
many factors, e.g. light intensity (Bruggink, 1992), temperature (Harssema, 1977), CO concentration (Madsen,
#
1973) and sink-source ratio (Heuvelink and Buiskool, 1995).
In the model, it is a forcing function (Heuvelink, 1995 b),
which depends on the day of the year (t ; day 1 is 1 January) :
SLA l 266j88 SIN(2(tj68)\365)

(3)

Thus only seasonal effects (mainly radiation ; Heuvelink,

1995 b) were taken into account. At present, a suitable, more
explanatory, well-validated and therefore satisfactory prediction of SLA is not available. Difficulties in simulation of
SLA are reviewed by Marcelis, Heuvelink and Goudriaan
(1998).

Selection of experiments for model alidation

The separate validation of the dry matter production and
dry matter distribution modules showed good agreement

between measurements and simulation (Heuvelink, 1995 a,

1996 a). Therefore, as pointed out before, validation of the
aggregated model should focus on the interaction between
the submodels. Such interactions play a role in the simulation
of LAI and Rm. Hence, validation was conducted in two
plant density experiments and two fruit pruning experiments. Plant density influences SLA, LAI and dry mass and
the latter affects Rm directly. Fruit pruning increases
assimilate allocation to the vegetative plant parts and
decreases specific leaf area (Heuvelink and Buiskool, 1995)
and, therefore, would influence LAI, although this is not
taken into account in eqn (3). Furthermore, a reduced fruit
load will increase dry mass per unit of ground area, because
of higher weights of the vegetative plant parts (Heuvelink
and Buiskool, 1995). Validations were made for different
seasons to make them more general.
In addition, two data-sets from commercially grown
crops (Rijsdijk, De Ruiter and Bergman, 1993) were used
for model validation. These cover the whole growing season,
resulting in tall, old plants, whereas the experiments (Table
1) lasted only for about 100 d after planting. Also, in
commercially grown crops, large biomass is present at the
end of the season under low light, making Rm relatively
important.

MATERIALS AND METHODS

Sensitiity analysis
In a sensitivity analysis, dry matter production and dry
matter distribution were simulated for a year-round tomato
crop, using the global radiation data of selected months
from the 19711980 weather records of De Bilt, Netherlands
(Breuer and Van de Braak, 1989). This so-called selected
year results in the same average irradiance as observed for
the 30-year average (19511980) global radiation data at De
Bilt ; however, it contains a representative variation in
radiation. Simulation started at anthesis of the first truss,
which was set at day 1 (1 January). The simulation period
ended at day 250 (7 September), a reasonable estimate for
the date when plants are stopped in practice. The tem-

416

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment

T 2. Output of the model at reference conditions and partial sensitiity [O\O\I\I, see eqn (4)] of output to a change
in inputs
Model output

Time period 1*
Reference output
Partial sensitivities
Temperature
Global radiation
CO concentration
#
SLA
Number of fruits per truss
Vegetative sink strength
Developmental stage of leaf pruning
Plant density
Time period 2
Reference output
Partial sensitivities
Temperature
Global radiation
CO concentration
#
SLA
Number of fruits per truss
Vegetative sink strength
Developmental stage of leaf pruning
Plant density

Crop growth
rate
(g m# d")

LAI
(m# m#)

2n86

1n32

k0n07
1n00
0n55
0n72
k0n11
0n11
0n56
0n24

k0n07
0n51
0n27
1n35
k0n19
0n19
0n78
0n63

14n70
k0n05
0n89
0n74
0n54
k0n29
0n30
0n53
0n01

Total
dry mass
(g m#)

Rm
(g CH O m# d")

Fruit\plant
ratio
(g g")

0n76

0n54

k0n06
0n66
0n34
0n42
k0n01
0n01
0n48
0n51

0n00
0n82
0n41
0n51
k0n14
0n14
0n54
0n39

0n02
0n00
0n00
0n00
0n33
k0n34
0n00
0n00

2n12

472n2

3n29

0n74

k0n10
0n92
0n75
1n58
k0n99
0n99
1n71
0n02

k0n10
0n89
0n73
0n55
k0n45
0n45
0n77
0n05

0n00
0n90
0n73
0n54
k0n52
0n52
0n83
0n03

0n00
0n00
0n00
0n00
0n25
k0n26
0n00
0n00

83n4

* Average values over two time periods are presented : (1) day 1 to day 60 (representing the period of anthesis of the first truss until first harvest
of ripe fruits) ; and (2) day 61 to day 250 (producing crop).
Partial sensitivity of temperature was expressed per change of 1 K instead of per relative change.

perature was set to 20 mC and the CO concentration was

#
350 mol mol".
Initial dry mass was set to 20, 9, 9 and 0 g m# for leaves,
stem, roots and fruits, respectively. Plant density was
2n5 m# and no side shoots were retained. Glasshouse
transmissivity for diffuse radiation was 0n70, the number of
fruit per truss was 8 and the ground reflection for light was
0n5 (whole-season estimate for soil covered with white
plastic sheet).
The effects of a change in model input or parameters were
investigated by calculating the relative partial sensitivity of
model output :
O\O
(4)
I\I
where O\O is the relative change in model output, and I\I
is the relative change in the value of a parameter or input
data. Sensitivity was calculated as the average sensitivity to
a change in parameter or input data by k10, k5, 5 and
10 %. Sensitivity to temperature was calculated as the
average sensitivity to a change in temperature by k2, k1,
1 and 2 K. Model outputs examined were : (1) average crop
growth rate ; (2) average LAI ; (3) average weight of aboveground biomass ; (4) average Rm ; and (5) average fraction of
dry matter partitioned into the fruits.
As it was expected that the response of a young crop may
differ from that of a producing crop, output was averaged
over two periods : (1) the time from first flowering (day 1)

until 60 d (fruit growth period at 20 mC) later, representing

the transition from a young non-producing crop to a
producing crop ; and (2) a subsequent period of 190 d
covering the time up to topping.
Model alidation
The model was validated on four glasshouse experiments
involving Lycopersicon esculentum Counter , all published
previously (Table 1) and on data from commercially grown
crops of Pronto (Rijsdijk et al., 1993), described and used
before by Heuvelink (1995 a). Leaves were picked only
twofour times during an experiment and not weekly, as in
commercial practice, and as was simulated by the model at
a temperature of about 20 mC. Therefore, in the validation
on the experiments, the dates at which leaves of a vegetative
unit were removed from the plants were an input to the
model.
From the end of March onwards, a side shoot was
retained on each fourth plant in the commercially grown
crops, increasing shoot density by 25 %. Therefore, in the
simulations, from day 100 (first anthesis of side shoot)
onwards, for all newly formed trusses and for all vegetative
units at a higher or equal level as the one below the first truss
on the side shoot, sink strengths were multiplied by a plant
density 25 % higher than the original plant density. This was
to calculate sink strengths per unit ground area. The
number of fruit per truss was set to 9 in the commercially

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment
grown crops, based on the measurements of De Koning
(1994 ; Fig. 6.3.9.b) for Nursery II.
The ground reflection coefficient for light was assumed to
be 0n15 for the experiments (bare dark soil ; Rosenberg, Blad
and Verma, 1983) and 0n5 for the commercially grown crops
as, in the latter, the soil was covered with white plastic sheet.
To compare measured and simulated crop growth, the
model was also run with LAI, dry matter distribution and
biomass measured as inputs. The latter was used for the
calculation of Rm only. Cubic spline functions, fitted to
observed values for LAI and dry mass of leaves, stem and
fruits were used. A complete description of this procedure,
used for testing the dry matter production module separately, has been given (Heuvelink, 1995 a).

Sensitiity analysis
Sensitivity analysis showed that global radiation, CO
#
concentration, SLA and the developmental stage of a
vegetative unit at leaf pruning had a large influence on crop
growth rate, whereas temperature, number of fruits per
truss, vegetative sink strength and plant density were less
important (Table 2). Sensitivity of crop growth rate to the
changes in parameters or factors tested differed little between
young and producing crops (i.e. when harvest had commenced), except for sensitivity to plant density, to the
number of fruits per truss and to the sink strength of a
vegetative unit relative to one fruit.
In the young crop, crop growth rate decreased a little at
higher temperature, because more assimilates were partitioned into the fruits, resulting in a decreased LAI (Table
2). Growth rate was proportional to LAI : a 7 % decrease in
LAI as a result of a 1 K rise in temperature resulted in a 7 %
decrease in crop growth rate. Increased dry matter partitioning to the fruits at increased temperature resulted from
increased earliness (increased flowering and fruit development rate).
In a producing crop, temperature did not influence dry
matter partitioning. However, LAI was also negatively
influenced by increased temperature as final leaf size
decreased. For a producing crop, growth rate was not
proportional to LAI : a 10 % decrease in LAI resulting from
a 1 K temperature increase reduced crop growth rate by
5 %. A change in temperature did not influence Rm. At cool
temperatures a larger biomass compensated for the lower
respiration rate per unit biomass, whereas at warmer
temperatures biomass was reduced, which in turn compensated for the increase in respiration rate per unit biomass
(Table 2).
Global radiation and CO concentration influenced crop
#
growth rate positively (Table 2), mainly as a result of a
change in rate of leaf photosynthesis. Furthermore, LAI
increased with global radiation and CO concentration,
#
which augmented the subsequent crop growth rates. A
higher global radiation or CO concentration increased Rm,
#
because of larger biomass.
Larger SLA resulted in a higher crop growth rate,
explained by a strong positive influence of SLA on LAI and

16
A
12

08

04

00
60

100

140

180

220

100

140

180

220

100

140
Day of year

180

220

16
Total dry weight (kg m2)

RESULTS

417

B
12

08

04

00
60
16
C
12

08

04

00
60

F. 2. Measured (
, , $, #, >, =) and simulated [aggregated
model (IIII) ; model with LAI, dry matter partitioning and biomass
present as inputs (II II)] dry matter production at (A) low, (B)
medium and (C) high plant density. Expt 1 ( , #, =) ; expt 2 (
, $,
>). Plant density was 1n6 m# ( ), 2n1 m# ($, #), 2n9 m# (=),
3n1 m# (>) and 4n8 m# (
). Vertical bars indicate s.e.m. larger than
symbols.

partly counteracted by a greater biomass and therefore a

higher Rm (Table 2). Both in the young and the producing
crop, an increase in SLA resulted in a more than
proportional increase in LAI.
An increase in the number of fruits per truss decreased
crop growth rate, as LAI was negatively influenced (Table
2). Exactly the opposite sensitivity was observed for
increased sink strength of a vegetative unit relative to sink
strength of a fruit. Influences on LAI resulted from effects
on dry matter partitioning into the vegetative parts. Growth
rate of a producing crop was three-times more sensitive to
a change in the number of fruits per truss or the sink
strength of a vegetative unit, than growth rate of a young

418

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment

T 3. Measured and simulated aerage crop growth rates for the first month after planting* (A) and for the rest of the
experimental period (B)
Crop growth rate (g m# d")
Experiment
A
1
2
3
4

Treatment

Measured

Aggregated model

Aggregated model

With LAI input

1n6 plants m#
2n1 plants m#
3n1 plants m#
2n1 plants m#
2n9 plants m#
4n8 plants m#
3 fruits per truss
7 fruits per truss
control
k50 % trusses

2n21
2n93
4n04
6n29
7n68
8n23
3n96
4n14
4n72
4n46

4n48
4n96
5n59
10n60
11n35
12n09
6n53
6n52
5n10
5n17

2n03
1n69
1n38
1n69
1n48
1n47
1n65
1n57
1n08
1n16
1n52

1n48
1n36
1n23
1n21
1n21
1n33
1n25
1n21
1n14
1n19
1n26

14n2
14n3
14n6
16n1
16n3
16n1
15n2
15n3
10n2
11n0

1n23
1n23
1n12
1n18
1n11
0n94
1n21
1n19
0n99
0n99
1n12

1n15
1n23
1n12
1n11
1n09
0n96
1n18
1n11
1n08
1n03
1n11

Average
B
1

1n6 plants m#
2n1 plants m#
3n1 plants m#
2n1 plants m#
2n9 plants m#
4n8 plants m#
3 fruits per truss
7 fruits per truss
control
k50 % trusses

2
3
4

Simulated\measured crop growth rate

11n6
11n7
13n0
13n6
14n7
17n1
12n6
12n9
10n3
11n1

Average

* Calculated as (WendkWstart)\(tendktstart) ; W, Total crop dry mass ; t, time.

Calculated as the slope of a linear regression of W against time.
Simulation with LAI, dry matter partitioning and biomass present being model inputs ; the latter was used for prediction of Rm only.

crop. This was caused by a five-times smaller sensitivity of

LAI to the number of fruits per truss or vegetative sink
strength in the young crop, as fruit number on the plant is
increasing in the young crop : it starts without fruit and at
that moment the number of fruits per truss or vegetative
sink strength relative to generative sink strength does not
play a role at all. A change in number of fruits per truss or
sink strength of a vegetative unit influenced above-ground
biomass (leaf and stem weights), and therefore Rm. However,
these effects were quite small in a young crop.
The developmental stage of a vegetative unit at which the
leaves are removed influenced LAI strongly (more than
proportional effect in the producing crop) and therefore
crop growth rate (Table 2). Early leaf pruning decreased
LAI as well as biomass and therefore Rm.
Denser planting increased crop growth rate only for the
young crop because of its positive influence on LAI (Table
2). For a producing crop, none of the crop characteristics
investigated was influenced by plant density.
Model alidation
For expts 1 and 2, simulated crop growth exceeded
measured growth (Fig. 2 and Table 3). Final dry mass was
overestimated by 031 %. In both experiments, at the
highest plant density, the aggregated model provided
identical growth curves whether or not LAI, dry matter
partitioning and biomass were input to the model (Fig. 2 C).

At the lowest plant densities, final dry matter production

was greater with the aggregated model compared to the
simulations with LAI, dry matter distribution and biomass
being inputs. This discrepancy originated mainly from a
higher simulated growth rate in the first month after
planting (Fig. 2 A, Table 3).
Crop growth rate was also overestimated in the first
weeks after planting when LAI, dry matter partitioning and
biomass present were inputs to the model (Fig. 2, Table 3).
In the first month after planting, crop growth rate was
overestimated by 48, 36 and 23 % in expt 1 at low, medium
and high plant density, respectively. In expt 2 these values
were 21, 21 and 33 %.
For expt 1, simulated LAI was in reasonable agreement
with the measurements, except for the first month after
planting, where LAI was largely overestimated (Fig. 3 A).
For the lowest plant density, LAI was overestimated over
the whole experimental period (Fig. 3 A). For all three
treatments, the time course of simulated SLA showed
reasonable agreement with the measurements. For the
smaller and medium plant density, SLA was over-estimated
on average by 7 % (values between k6 and 34 %) and 1 %
(values between k6 and 17 %), respectively, whereas for the
largest plant density SLA was underestimated by 4 %
(values between 3 % overestimation and 9 % underestimation). Note that control treatment from expts 1, 3
and 4 were used to determine eqn (3) and hence validation
of SLA was partly on the same data as used for model

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment
7

50

419

40

5
30

4
3

LAI (m2 m2)

2
1
0
60
7

80

100

120

140

160

180

6
5

Total dry weight (kg m2)

20
10

0
50

50

100

50

100

150

200

250

300

150
200
Day of year

250

300

40
30

4
20

3
2

10

1
0
100

0
120

140

180
160
Day of year

200

220

F. 3. Measured (
, , $, #, >, =) and simulated (IIII,II II,
I I I) LAI at three plant densities in (A) expt 1 and (B) expt 2. Plant
density was 1n6 m# ( ), 2n1 m# ($, #), 2n9 m# (=), 3n1 m# (=) and
4n8 m# (
). Vertical bars indicate s.e.m. larger than symbols.

development. However, determination of eqn (3) on the

data excluding these three control treatments resulted in an
almost identical equation (not shown). For expt 2, LAI
was overestimated in all three treatments. This resulted
mainly from a too rapid increase in simulated LAI in the
first month after planting (Fig. 3 B). This was at least partly
caused by overestimation of SLA for the first month after
planting by 25, 17 and 10 % for the small, medium and large
plant densities, respectively. Thereafter average overestimation of SLA was reduced to 11, 11 and 8 %, respectively.
Final fruit yield was overestimated by 2942 %, except for
the highest plant density in expt 2, where simulated and
measured fruit yield were equal (not shown). Overestimation
of fruit yield resulted from overestimation of crop growth
(Fig. 2), whereas dry matter partitioning was simulated well
(Heuvelink, 1996 a).
For both treatments in expt 3 the crop growth rate was
overestimated, but when LAI, dry matter partitioning and
biomass present were inputs to the model, crop growth rate
was predicted reasonably well (Table 3). Overestimation of
final dry mass (1722 %) mainly resulted from overestimation of crop growth rate in the first month after planting
(Table 3), caused by an overestimation of LAI, which could
be explained by a strong overestimation of SLA over the
whole experimental period. The SLA overestimation was
23 % for the control and 37 % for the fruit pruning
treatment. On day 94, simulated LAI was more than twice

F. 4. Measured ($) and simulated [aggregated model (IIII) ;

model with LAI, dry matter partitioning and biomass present as
inputs (II II)] dry matter production for commercially grown crops
at Nursery I (A) and Nursery II (B).

the measured value (3n2 instead of 1n5 m# m#), for both

treatments. For the fruit pruning treatment, LAI at the end
of the experiment was simulated to be 6n8, whereas measured
LAI was only 3n9. In addition to overestimation of SLA,
this was also caused by slight overestimation of the fraction
of dry matter partitioned into the vegetative plant parts
(Heuvelink, 1996 a).
In expt 4, simulated and measured crop growth were in
good agreement (Table 3). For both treatments, simulated
growth was smaller for the aggregated model than when
LAI, dry matter partitioning and biomass present were
inputs to the model (Table 3). This could be explained by
underestimation of LAI in the period from day 190 until
day 210 (not shown), which resulted mainly from a 17 %
underestimation of SLA in the first month after planting.
Averaged over the four experiments, crop growth rate
was overestimated, especially for the young crop (Table 3).
In the first month after planting (Table 3), the aggregated
model overestimated crop growth rate by 52 %, whereas this
was 26 % when LAI, dry matter partitioning and biomass
present were inputs to the model. During the rest of the
experimental period (Table 3), these values were 12 and
11 %, respectively.
For the commercially grown crops, simulated crop growth
(Fig. 4) and dry matter partitioning (not shown) were in
reasonable agreement with the measurements, except for
partitioning in the period 50100 d after first anthesis,
where the average simulated fraction partitioned into the

420

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment
total crop growth rate was largely underestimated for both
crops, due to underestimation of LAI (not shown).

DISCUSSION

LAI (m2 m2)

50

100

50

100

150

200

250

300

150
200
Day of year

250

300

F. 5. Measured ($, II II) and simulated (IIII) LAI for commercially

grown crops at Nursery I (A) and Nursery II (B).

300

SLA (cm g )

400

200

100

50

100

150

200

250

300

Day of year
F. 6. Measured ($,
) and simulated (IIII) SLA for commercially
grown crops at Nursery I ($) and Nursery II (
).

fruits was 0n74, whereas the measured fraction was 0n64. For
Nursery I, simulated growth was somewhat higher than
when LAI, dry matter partitioning and biomass present
were inputs to the model (Fig. 4 A). This was caused by
overestimation of LAI between days 125 and 250 (Fig. 5 A).
For Nursery II, both simulated crop growth curves were
very similar (Fig. 5 B) and simulated LAI agreed well with
measured LAI (Fig. 5 B). This agreement, however, was the
result of compensating errors in the model, as SLA was
strongly overestimated for both commercially grown crops
(Fig. 6). When measured SLA was an input to the model,

For all experiments, except expt 4, agreement between

measured and simulated crop growth was worse for the
aggregated model than when LAI, dry matter partitioning
and biomass were inputs to the model (Table 3). There is a
greater risk of propagation of errors using the aggregated
model. Measured and simulated crop growth rates differed
mainly at low LAI (Table 3). At low LAI, light interception
and thus crop growth is very sensitive to LAI (Bertin and
Heuvelink, 1993). At high LAI (above 3) this sensitivity is
much reduced and an overestimation of LAI in that case
hardly influences crop growth. This is shown for the highest
plant density in expt 2 : although LAI was largely
overestimated (Fig. 3 B), simulated crop growth agreed very
well with the measurements (Fig. 2 C).
The large overestimations of crop growth rate in the first
weeks after planting (Table 3), with LAI, dry matter
partitioning and biomass as model inputs, are the main
reason for the discrepancy between crop growth curves
provided by the aggregated model and the simulation with
LAI, dry matter partitioning and biomass as inputs (Fig. 2).
This overestimation of crop growth rate is amplified in the
aggregated model because an overestimation of crop growth
rate has a positive feedback on subsequent crop growth rate
through an overestimation of LAI (Fig. 1). This positive
feedback is one of the major problems of dynamic crop
growth simulation.
Possible reasons for overestimation of crop growth rate in
the first weeks after planting when LAI, dry matter
partitioning and biomass present were inputs to the model,
may be clustering of leaves (row crop of small plants not yet
touching each other) and a possibly lower light intensity at
the top of the young plants than simulated at the top of a
producing crop in an infinite glasshouse. The model assumes
a homogenous distribution of leaf area over the glasshouse
surface, which will overestimate crop growth rate if this is
not the case. Simulation can be improved by using a module
for light interception in a row crop (e.g. Gijzen and
Goudriaan, 1989). The glass walls of the 12i12n8 m
glasshouse compartments in which experiments were conducted will cause a vertical light gradient and hence
overestimation of radiation above the crop at low crop
height. This explanation is quite likely, as in expts 3 and
4 (Table 3) and in all 12 validation experiments of
Heuvelink (1995 a) an overestimation of crop growth rate in
the first weeks after planting was also observed [see Fig. 1 in
Heuvelink (1995 a)], in simulations where LAI, dry matter
partitioning and biomass were inputs to the model. Another
reason for overestimation of crop growth rate, which also
holds for a producing crop (Table 3), is the assumption of
potential crop growth (see Model Description).
The large underestimations of crop growth for the
commercially grown crops, with measured SLA as input
(not shown), probably resulted from more leaf pruning in
the model than in reality. Although precise moments of leaf
picking were not known, it could be concluded from the

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment
destructive measurements per vegetative unit that indeed
the number of leaves retained on the plants in summer was
greater than predicted by the model. Sensitivity analysis
showed that a small shift in leaf picking can greatly
influence crop growth (Table 2). This is especially true if, as
in the case of the commercially grown crops, LAI is mostly
below 2n5 (Fig. 5), when any variation will have a relatively
large effect on crop photosynthesis and thus crop growth
(Bertin and Heuvelink, 1993).
The validation experiments and the sensitivity analysis
demonstrate the importance of accurate prediction of SLA.
For the commercially grown crops (Fig. 6), as well as for
expt 3, SLA was largely overestimated. Thicker leaves in
the commercially grown crops may have resulted from the
exposure to high CO concentrations (Madsen, 1973).
#
Equation (3) is based on experiments without CO en#
richment, resulting in CO concentrations between 300 and
#

400 mol mol " (Heuvelink, 1995 b), between 1000 and
1600 h, averaged over the whole growing period. For
Nurseries I and II, CO concentration was much higher,
#
being 510 and 549 mol mol", respectively, between 1000
and 1600 h, averaged over the whole growing period.
Overestimation of SLA in expt 3 remains unexplained :
expt 1 (2n1 plants m#) was conducted in the same period
of the year, at almost the same average radiation,
temperature (Table 1) and CO concentration (not shown),
#
but showed good agreement between measured and simulated SLA.
The comparison with commercially grown crops clearly
shows the necessity of model validation at submodel level.
On the basis of total dry matter production (Fig. 4) one
would have concluded that the aggregated model is accurate.
However, the analysis showed that the correct simulation
resulted from overestimation of SLA (Fig. 6) which was
compensated by another error, probably too early leaf
picking. It also shows the necessity of elaborate data sets for
validation.
Note that the results of the sensitivity analysis (Table 2)
depend on the definition of the reference crop . For
example, if the reference crop contained only four fruits per
truss instead of eight, sensitivity of crop growth rate of the
producing crop to the number of fruits per truss was much
reduced (k0n05 instead of k0n3), whereas sensitivity of
fruit\plant ratio was increased (0n4 instead of 0n3). At a
lower number of fruits per truss, less assimilate is partitioned
into the fruits and LAI and existing biomass are higher. The
reduced sensitivity of crop growth rate to change in number
of fruits per truss resulted from a reduced relative effect on
LAI, as average LAI was 70 % higher in the new reference
situation with four fruits per truss, and a reduced relative
influence on Rm as the average biomass present was 24 %
higher. A large LAI, resulting from less leaf removal, is also
the reason that Heuvelink and Buiskool (1995) observed no
influence of fruit pruning (range : twoseven fruits per truss)
on total crop dry matter production, whereas sensitivity
analysis (Table 2) showed decreased crop growth with more
fruit.
Temperature did not influence Rm in the long term,
because after decreased temperatures there was more
biomass present, which compensated for a smaller res-

421

piration rate per unit of biomass and ice ersa. This is in

agreement with a theoretical analysis of De Koning (1994),
based on the assumption that dry matter distribution and
biomass production do not change with temperature.
Prolonging low glasshouse temperatures, to reduce Rm, is
not practical as biomass present will increase due to low
organ development rate.
In the present sensitivity analysis a temperature increase
of 1 K did not influence Rm, whereas De Koning (1994)
calculated a decrease by as much as 15 %, mainly as a result
of a 12 % decrease in crop growth rate. This difference is
explained by the strong direct influence of temperature on
dry matter partitioning observed and modelled by De
Koning (1994). In his model, temperature increase reduces
partitioning to the vegetative parts and thus LAI and crop
growth rate. In TOMSIM, no direct influence of temperature
on partitioning is assumed, based on Heuvelink (1995 c) and
supported by the simulation of a climate room experiment
(Heuvelink, 1996 a). Van den Boogaart and Schellekens
(1990) observed a decrease in total growth rate of only 1 %
per K, when a tomato crop was grown at 23 mC compared
to 17 mC for 8 weeks starting at anthesis of the first truss.
For the first 60 d after anthesis of the first truss, sensitivity
analysis showed a decrease in crop growth rate with 7 % per
K (Table 2), which is in agreement with simulation results of
De Koning (1994). The discrepancy between measurements
of Van den Boogaart and Schellekens (1990) and simulation
results (Table 2 ; De Koning, 1994) is mainly caused by
increased flower and\or fruit abortion at higher temperatures (Van den Boogaart and Schellekens, 1990), resulting in
increased partitioning to the vegetative plant parts, which
was ignored in the simulations. Therefore, a temperature
rise of 1 K decreased LAI by only 2 % in the measurements,
whereas simulation predicted a decrease of 7 %.
In this work emphasis is on the interaction between dry
matter production and dry matter partitioning in tomato. It
is a major limitation of the model that flower and\or fruit
abortion is not simulated, as reduced fruit set at low
assimilate supply (Bertin, 1995) is an important feedback
mechanism in the tomato plant (Fig. 1 ; De Koning, 1994).
Note that model validation in this paper is limited to crops
grown in The Netherlands.
The modular (explanatory) structure and the thorough
sensitivity analysis and validation, also at submodel level,
makes TOMSIM a valuable tool. It enables the study and
analysis of the tomato crop, a very complex system, as a
whole. Crop growth models, like TOMSIM, may contribute
significantly to computerized management support in
glasshouse cultivation (Lentz, 1998) and may be used for
optimization in glasshouse climate control (Challa, 1990 ;
Jones et al., 1991). The present work has shown that it may
be necessary to measure (or estimate) LAI, e.g. every week
or every 2 weeks and supply this value to the model as an
input if TOMSIM is to be used for glasshouse climate
control. This will be especially important at low LAI [young
crop, and crop in summer showing short leaf syndrome
(Nederhoff, De Koning and Rijsdijk, 1992)]. Correct
simulation of LAI is hindered in two ways : (a) prediction of
SLA is very difficult, as it is influenced by many factors (see
Model Description) ; and (b) small discrepancies in the

422

HeuelinkA Dynamic Simulation Model for Tomato Crop Growth and Deelopment

prediction of crop growth are amplified (positive feedback,

Fig. 1).
A C K N O W L E D G E M E N TS
I thank A. A. Rijsdijk of the Glasshouse Crops Research
Station at Naaldwijk, for the use of his data collected at
commercial nurseries and Prof. Dr. H. Challa, Dr A. N. M.
de Koning and Dr L. F. M. Marcelis for critically reading
the manuscript.
LITERATURE CITED
Acock B, Reynolds JF. 1989. The rationale for adopting a modular
generic structure for crop simulators. Acta Horticulturae 284 :
391396.
Amthor JS. 1989. Respiration and crop productiity. Berlin : Springer
Verlag, 69104.
Bertin N. 1995. Competition for assimilates and fruit position affect
fruit set in indeterminate greenhouse tomato. Annals of Botany 75 :
5565.
Bertin N, Gary C. 1993. Tomato fruit-set : a case study for validation
of the model TOMGRO. Acta Horticulturae 328 : 185193.
Bertin N, Heuvelink E. 1993. Dry-matter production in a tomato crop :
comparison of two simulation models. Journal of Horticultural
Science 68 : 9951011.
Breuer JJG, Van de Braak NJ. 1989. Reference year for Dutch
greenhouses. Acta Horticulturae 248 : 101108.
Bruggink GT. 1992. A comparative analysis of the influence of light on
growth of young tomato and carnation plants. Scientia Horticulturae 51 : 7181.
Challa H. 1990. Crop growth models for greenhouse climate control.
In : Rabbinge R, Goudriaan J, Van Keulen H, Penning de Vries
FWT, Lan Laar HH, eds. Theoretical production ecology :
reflections and perspecties. Simulation Monographs 34. Wageningen, The Netherlands : Pudoc, 125145.
Dayan E, Van Keulen H, Jones JW, Zipori I, Shmuel D, Challa H. 1993.
Development, calibration and validation of a greenhouse tomato
growth model : II. Field calibration and validation. Agricultural
Systems 43 : 165183.
De Koning ANM. 1994. Deelopment and dry matter distribution in
glasshouse tomato : a quantitatie approach. PhD Thesis, Wageningen Agricultural University, Wageningen, The Netherlands.
Gijzen H. 1992. Simulation of photosynthesis and dry matter production
of greenhouse crops. Simulation Reports CABO-TT, 28, CABO,
Wageningen, The Netherlands.
Gijzen H, Goudriaan J. 1989. A flexible and explanatory model of light
distribution and photosynthesis in row crops. Agricultural and
Forest Meteorology 48 : 120.
Harssema H. 1977. Room temperature and growth of young tomato
plants. Mededelingen Landbouwhogeschool 77(19).
Heuvelink E. 1995 a. Dry matter production in a tomato crop :
measurements and simulation. Annals of Botany 75 : 369379.
Heuvelink E. 1995 b. Growth, development and yield of a tomato crop :
periodic destructive measurements in a greenhouse. Scientia
Horticulturae 61 : 7799.

Heuvelink E. 1995 c. Effect of temperature on biomass allocation in

tomato (Lycopersicon esculentum). Physiologia Plantarum 94 :
447452.
Heuvelink E. 1996 a. Dry matter partitioning in tomato : validation of
a dynamic simulation model. Annals of Botany 77 : 7180.
Heuvelink E. 1996 b. Tomato growth and yield : quantitatie analysis and
synthesis. PhD Thesis, Wageningen Agricultural University,
Wageningen, The Netherlands.
Heuvelink E, Bertin N. 1994. Dry-matter distribution in a tomato crop :
comparison of two simulation models. Journal of Horticultural
Science 69 : 885903.
Heuvelink E, Buiskool RPM. 1995. Influence of sink-source interaction
on dry matter production in tomato. Annals of Botany 75 :
381389.
Heuvelink E, Marcelis LFM. 1989. Dry matter distribution in tomato
and cucumber. Acta Horticulturae 260 : 149157.
Heuvelink E, Batta LGG, Damen THJ. 1995. Radiance transmission of
a multispan Venlo-type glasshouse : validation of a model.
Agricultural and Forest Meteorology 74 : 4159.
Jones JW, Dayan E, Allen LH, Van Keulen H, Challa H. 1991. A
dynamic tomato growth and yield model (TOMGRO). Transactions of the American Society of Agricultural Engineers 34 :
663672.
Kano A, Van Bavel CHM. 1988. Design and test of a simulation model
of tomato growth and yield in a greenhouse. Journal of the
Japanese Society for Horticultural Science 56 : 408416.
Lentz W. 1998. Model applications in horticulture : a review. Scientia
Horticulturae 74 : 151174.
McCarl BA. 1984. Model validation : an overview with some emphasis
on risk models. Reiew of Marketing and Agricultural Economics
52 : 153173.
Madsen E. 1973. Effect of CO -concentration on the morphological,
#
histological and cytological changes in tomato plants. Acta
Agriculturae Scandinaica 23 : 241246.
Marcelis LFM, Heuvelink E, Goudriaan J. 1998. Modelling biomass
production and yield of horticultural crops : a review. Scientia
Horticulturae 74 : 83111.
Nederhoff EM, De Koning ANM, Rijsdijk AA. 1992. Leaf deformation
and fruit production of glasshouse grown tomato (Lycopersicon
esculentum Mill.) as affected by CO , plant density and pruning.
#
Journal of Horticultural Science 67 : 411420.
Rijsdijk AA, De Ruiter HW, Bergman G. 1993. Gewasgroeigegeens,
erzameld in de praktijk in het teeltseizoen 19911992 bij tomaat
komkommer, paprika en aubergine. Glasshouse Crops Research
Station, Internal Report 5, Naaldwijk, The Netherlands.
Rosenberg NJ, Blad BL, Verma SB. 1983. Microclimate : the biological
enironment. 2nd edn. New York, USA : John Wiley & Sons,
4445.
Spitters CJT, Van Keulen H, Van Kraalingen DWG. 1989. A simple and
universal crop growth simulator : SUCROS 87. In : Rabbinge R,
Ward SA, Van Laar HH, eds. Simulation and systems management
in crop protection. Wageningen, The Netherlands : Pudoc, 147181.
Van den Boogaart J, Schellekens F. 1990. Inloed an temperatuur op
ontwikkeling, groei en produktie an tomaat : temperatuurproef
307\oorjaar 1990. Internal Report 44, Glasshouse Crops Research
Station, Naaldwijk, The Netherlands, 30pp.