Field Crops Research 112 (2009) 131–140

Modeling boll maturation period, seed growth, protein, and oil content ofcotton (Gossypium hirsutum L.) in China

Wenfeng Li a, Zhiguo Zhou a, Yali Meng a,*, Naiyin Xu a, Michel Fok b

a Key Laboratory of Crop Growth Regulation, Ministry of Agriculture, Nanjing Agricultural University, Nanjing 210095, Jiangsu Province, PR Chinab The Center of International Cooperation on Agronomic Research for Development (CIRAD), TA 72/09, 34398 Montpellier Cedex 5, France

A R T I C L E I N F O

Article history:

Received 27 November 2008

Received in revised form 17 February 2009

Accepted 20 February 2009

Keywords:

Cottonseed

Crop model

Boll maturation period

Biomass accumulation

Protein content

Oil content

A B S T R A C T

The simulation of cottonseed (Gossypium hirsutum L.) growth is still an area of great uncertainty,

especially in the process of cottonseed quality formation. A simple process-based model was developed

to predict cotton boll maturation period and simulate cottonseed biomass accumulation, protein, and oil

content. The cotton boll maturation period module took solar radiation and N nutrition factors into

account in addition to temperature and variety maturity profile. Based on the hypothesis that the

accumulation of biomass, oil, and protein are mainly sink-determined, the model was developed by

considering parameters of cultivar characteristics, weather (temperature and solar radiation), and crop

management variables (precisely N supply). The subtending leaf N concentration of cotton boll was

simulated by a new semi-empirical model, and worked as the direct indicator of the N nutrition effect on

cottonseed growth and development. The model was calibrated using data obtained in experiment

conducted in Nanjing (the lower reaches of Yangtze River Valley) in 2005 and 2006. The model was then

tested using two field experimental data sets. One was obtained in Nanjing, China in 2007, and the other

in the Yellow River Valley (Xuzhou and Anyang) and the lower reaches of Yangtze River Valley (Huaian),

China in 2005. The simulated values of boll maturation period by the model were very consistent with

the observed values, with root mean square error (RMSE) lower than 3 days. The RMSE of cottonseed dry

weight, protein content, and oil content predictions were 8.9 mg seed�1, 2.19%, and 2.71%, respectively.

The result showed that the model is sufficiently robust to predict the cotton boll maturation period,

cottonseed dry weight, and quality in wide range of conditions. It is not only a necessary component of

cotton growth model, but also provides a good platform for further study in modeling cottonseed protein

and oil yield.

� 2009 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Field Crops Research

journa l homepage: www.e lsev ier .com/ locate / f c r

1. Introduction

Simulation models are increasingly used for the assessment ofcrop productivity and the impact on the environment that mayresult from given combinations of crop characteristics, weather,soil, and crop management. The study of crop growth modelstarted from 1960s and cotton (Gossypium hirsutum L.) was one ofthe earliest crops studied. Since the model SIMCOT (Duncan,1973), a lot of systemic cotton growth models had appeared.GOSSYM (Fye et al., 1984; Reddy and Baker, 1988, 1990; Booneet al., 1993) was one of the most famous models. It simulated thedynamics of development, growth, and yield production of cottonplants. Meanwhile, KUTUN developed by Mutsaers (1984) inHolland was also a more mechanistic model. The studies inAustralia paid more attention to field management, such as

* Corresponding author. Tel.: +86 25 84396813; fax: +86 25 84396813.

E-mail address: giscott@njau.edu.cn (Y. Meng).

0378-4290/$ – see front matter � 2009 Elsevier B.V. All rights reserved.

doi:10.1016/j.fcr.2009.02.009

SIRATAC (Hearn and daRoza, 1985) and OZCOT (Wells and Hearn,1992; Hearn, 1994). With the development of eco-physiologystudies in cotton, COTTAM (Jackson et al., 1988; Larson et al.,1996) was developed with more comprehensive inputs includingclimate condition, soil parameters, and management variables.Linking physiological and architectural models enhanced thecotton models in function, and the COTONS model developed in1998 was a good example (Jallas et al., 2000; Hanan and Hearn,2003). Moreover, many other models at that time showed theirparticular feature, such as COTCO2 (Wall et al., 1994), COTGROW(Pan et al., 1996), and the model developed by Reddy (1994). Nowmany researchers are still engaging in the area, with newadvances made (Ko et al., 2005; Yang et al., 2008; Ravula et al.,2008).

Cotton is not only the most important fibre crop in the worldbut also the second best potential source for plant proteins aftersoybean, the fifth oil-producing plant after soybean, palm-tree,colza, and sunflower (Sawan et al., 1988; Ahmad et al., 2007).Therefore, there is a need to understand the accumulating

W. Li et al. / Field Crops Research 112 (2009) 131–140132

processes of oil and protein in cottonseeds. Cotton modeling hasbeen well studied, but studies on simulating cotton seed growthhave so far remained limited. To our knowledge, there is noreport or communication regarding the modeling of theformation of cottonseed protein and oil. Both linear (Thakeret al., 1989) and cubic polynomial (Rabadia et al., 1999) modelscould be used to delineate cottonseed biomass accumulation.The nitrogen accumulation and the change of oil contentafter flowering had also been quantified (King and Leffler,1979) by simple equations. However, none of them couldexpress the changes in the accumulation under diverseenvironment conditions, nor explain the physiological mechan-isms involved.

Modelling of boll maturation period serves at the basis of themodelling of cottonseed growth and quality formation. Previousstudies usually predicted crop development related to thermaltime mainly determined by temperatures (Hearn and daRoza,1985; Jamieson et al., 1998). Recently, Physiological Develop-ment Time (PDT) has been widely used in many crops such aswheat, rice, and cotton (Cao and Moss, 1997; Meng et al., 2003;Zhang et al., 2003), mainly taking into account variety maturityprofile, temperature conditions, and photoperiod. With regard tocotton boll development, some studies have showed that theboll maturation period is also influenced by solar radiation andnitrogen nutrition (Chen and Yu, 2001; Reddy et al., 2004), inaddition to genotype and temperature conditions. Abundantsunshine accelerates boll development as a subsidiary factor,especially when temperature conditions are sub-optimal (Chenand Yu, 2001; Ma et al., 2005). Excessive nitrogen nutrition hasbeen found to delay cotton boll maturation (Tang et al., 2003;Reddy et al., 2004). So far, studies on modeling boll developmenthave not integrated the factors of solar radiation and Nnutrition.

This paper presents the results of a first integrated model oncottonseed growth and development encompassing: (1) a moduleof cotton boll maturation period model taking into account theeffects of cultivar characteristic, weather conditions, and cropmanagement variables (precisely N supply); (2) module ofcottonseed growth model dealing with cottonseed biomassaccumulation in daily step; (3) module of cottonseed proteincontent and cottonseed oil content model based respectively on N

accumulation and fat synthesis.

2. Materials and methods

2.1. Field data

2.1.1. Experiments for boll maturation period model module

Field experiments were conducted at Jiangsu AgriculturalSciences Academy, Nanjing, Jiangsu, China (328020N, 1188500E) in2006 and 2007. The soil at the experimental site was a yellow-brown loam with 25 mg kg�1 organic matter, 12 g kg�1 total N,85.1 mg kg�1 available N, 13.0 mg kg�1 available P, and 91.6 mg kg�1

available K contained in 20 cm depth of the soil profile. Two cultivars,Kemian 1(The growth period is 135 days) and NuCOTN 33B (Thegrowth period is 120 days), were planted in 2006; Three cultivars,

Table 1Soil types and soil nutrient contents at the experimental sites in 2005.

Site Soil type Total N content

(g kg�1)

A

c

Nanjing Yellow brown Loam 10.3 4

Xuzhou Alluvial soil 11.7 4

Huaian Alluvial soil 9.7 4

Anyang Sand loam 9.4 3

Kemian 1, NuCOTN 33B and Dexia 1(The growth period is 100 days),were planted in 2007. Based on the research result that 240 kg ha�1 isthe optimum N rate in Nanjing (Xue et al., 2006), three N rates (0, 240,and 480 kg ha�1) were applied in two equal amounts, beforetransplanting and at initial flowering stage. Cotton seeds wereplanted on 25 April in 2006 and 2007. When the seedlings had threetrue leaves, individual healthy and uniform plants were transplantedto the field at row spacing 0.9 m and plant spacing 0.3 m. Theexperiments were designed as randomized complete blocks withthree replicates. Each plot was 6.3 m wide and 8.4 m long. In theexperimentsthebollflowering dates,opening dates,and bollpositionwere recorded.

2.1.2. Experiments for cottonseed growth and quality formation

model modules

To study the cottonseed biomass accumulation, protein and oilcontent, a series of field experiments were carried out in Nanjing(328020N, 1188500E), Huaian (338180N, 1198050E), and Xuzhou(348120N, 1178360E) in Jiangsu Province, and Anyang (368040N,1148130E) in Henan Province, China in 2005. The soil type and soilnutrient contents were listed out in Table 1. Two cultivars, Kemian1 and NuCOTN 33B, were planted on 25 April and 25 May, at rowspacing 0.9 m and plant spacing 0.3 m. Three N rates (0, 240, and480 kg ha�1) were applied in two equal amounts, before trans-planting and at the initial flowering stage. The experiments weredesigned as randomized complete blocks with three replicates inNanjing and four replicates in other locations. Crop managementwas in line with local cultivation practices followed for uplandcotton production in the region.

The white flowers at the first and second nodes of cottonplants were tagged on 15 July, 25 July, 10 August, 25 August, and10 September. 10–15 pairs of boll tagged and its subtending leafwere picked each time at the 5th, 10th, 17th, 24th, 31th, 38th,45th, and 52th day after flowering. The shells, seeds, and fiberwere separated and then dried. The oil contents of cottonseedswere measured with Soxhlet extraction (Luque De Castro andGarcia-Ayuso, 1998), and the N concentrations of cottonseedsand subtending leaves were measured with Kjeldahl method(Feil et al., 2005). Cottonseed protein content = 6.25 � N con-centration.

2.2. Statistical analysis

Field data collected in Nanjing in 2005 and 2006 were used todevelop model and calibrate model parameters. The model wastested using independent field data collected in Nanjing in 2007and in Xuzhou, Huaian and Anyang in 2005. Simulated values werecompared with the observed values using the root mean squareerror (RMSE) calculated as:

RMSE ¼Xn

j¼1

ðY j � X jÞ2

n

24

35

0:5

where Yj = simulated value on the ith day, Xj = observed valueon the ith day, n = number of pairs of simulated and observedvalue.

vailable N

ontent (mg kg�1)

Available P

content (mg kg�1)

Available K

content (mg kg�1)

0.13 28.95 79.17

7.83 29.00 77.54

1.33 27.7 75.7

9.28 23.57 71.19

W. Li et al. / Field Crops Research 112 (2009) 131–140 133

3. Results

3.1. Model development

This model was built to incorporate the current understanding ofthe basic eco-physiological processes involved in cottonseed growthand development. It included four modules: cotton boll maturationperiod module (BMPM), cottonseed biomass accumulation module(CSBAM), cottonseed protein content module (CSPCM), and cotton-seed oil content module (CSOCM). It was developed by thehypothesis that the biomass accumulation, nitrogen accumulation,and fat synthesis are mainly sink-determined.

RTEðTÞ ¼

0; T > Tc;or; T < Tb

T � Tb

To � Tb

� �1þTo�T=To�Tb

� Tc � T

Tc � To

� �Tc�To=To�Tb

; Tb � T � To

T � Tb

To � Tb

� �� Tc � T

Tc � To

� �Tc�To=To�Tb

; To < T � Tc

8>>>>><>>>>>:

(5)

3.1.1. Cotton boll maturation period module (BMPM)

This module predicts the developmental phases from floweringto maturity of cotton boll. The number of days having optimalclimatic conditions needed to complete a development phase isdefined here as physiological developmental time (PDT), whichdetermines the maximal development rates. Physiological effect(PE) stands for the ratio of boll developmental rate in actualconditions to which in the optimum condition, which is calculatedas a result of parameters of varietal maturity, weather, and cropmanagement variables (precisely nitrogen supply).

PEi ¼ VE� CETSi � ENbdi (1)

PDT ¼Xn

i¼1

PEi (2)

where VE is a variety early maturity parameter, CETS means thecomprehensive effect of temperature and solar radiation, and ENbdstands for the effect of nitrogen on boll development. In this paper,the subscript i always represents the days after flowering of theboll observed. VE stands for the relative development rate of thetested cultivar in the optimum condition, which can be estimatedaccording to the ratio of the boll maturation period of the givenearly matured cultivar here to which of the tested cultivar.According to the observed values in present experiments, 35 dayswas assumed as the boll maturation period of the given cultivarunder optimum growth condition.

The comprehensive effect of temperature and solar radiation(CETS) can be calculated as:

CETSi ¼ DTEi þ SREi � DTEi � SREi (3)

ELBPi ¼

1� 0:3� 1� 24� PDT

24

� �2 !

1� 0:3� 1� GDD� 380

220

� ��

1� 0:5� GDD� 600

600� 1�

�

1� 0:5� 1� 24� PDT

24

� �2 !

8>>>>>>>>>>>>>>><>>>>>>>>>>>>>>>:

where DTE means the daily temperature effect, SRE means thesolar radiation effect. SRE is a linear function of solar radiation (SR,MJ m�2) and can be calculated as:

SREi ¼ 0:4� SRi

30(4)

Temperature is considered as the primary environmental factorin cotton boll development (Roussopoulos et al., 1998). Accordingto the non-linear response curves of boll development totemperature (Plant et al., 1998), the relative temperature effect(RTE) is calculated by the Eq. (5), and the daily temperature effect(DTE) is calculated by Eq. (6).

DTE ¼ 0:5� RTEðTavgÞ � 0:25� RTEðTmaxÞ � 0:25

� RTEðTminÞÞ (6)

where Tb, To, and Tc are the cardinal temperatures values (base,optimum, and ceiling temperature) for development. They areretained as 15, 30, and 35 8C, respectively (Chen et al., 2006). Tavg,Tmax, and Tmin mean the daily average, maximum, and minimumair temperature, respectively.

The effect of nitrogen on boll development (ENbd) is defined bythe actual subtending leaf nitrogen concentration (LNC, %) inrelation to critical leaf N concentration (CLNC, %):

ENbdi ¼ 1þ 0:3� CLNCi�1 � LNCi�1

CLNCi�1(7)

According to the field experiment conducted in Nanjing in 2005,LNC decreases gradually following a negative exponential functionwith boll development (Fig. 1). CLNC is quantified according to themaximum LNC (Nmax, %), minimum LNC (Nmin, %), and bolldevelopment (indicated by PDT). Two semi-empirical formulaswere developed respectively to indicate the change of LNC andCLNC with PDT:

LNCCi ¼ ðNmax � NminÞ � expð�0:02� PDTÞ þ Nmin (8)

LNCi ¼ ELBPi � ESN� ðNmax � NminÞ � expð�0:015� PDTÞ

þ Nmin (9)

where ELBP means the effect of boll position on LNC, and ESNmeans the effect of soil N supply on LNC. According to theresponses of LNC to the boll position and soil N supply, theresponse functions were developed as follows:

; GDD<380

1� 24� PDT

24

� �2 !

; 380 � GDD<600

24� PDT

24

�2!; 600 � GDD � 1200

; GDD>1200

(10)

Fig. 1. Observed (symbols) and simulated (lines) subtending leaf N concentrations

of cotton bolls of cv. Kemian 1 and cv. NuCOTN 33B at different flowering dates

W. Li et al. / Field Crops Research 112 (2009) 131–140134

ESN ¼ 1� expð�0:015� ðNsoil � NAP� NRÞÞ (11)

where GDD is degree days accumulated above the threshold on15 8C from the first flowering date in the plant to present floweringdate; Nsoil is the seasonal soil N supply of no fertilizer field(kg ha�1), NAP is the N fertilizer rate(kg ha�1), and NR is the N

fertilizer use efficiency.

3.1.2. Cottonseed biomass accumulation module (CSBAM)

In general, crop growth models simulated biomass accumula-tion driven by photosynthesis, and simulated the partitioning inorgans with allocation coefficient. The source-sink relationship forcottonseed growth is difficult to quantify, so the present modulesimulates the cottonseed biomass accumulation by the way oflogistic function and assumes that the potential growth is mainlysink-determined. The potential growth rate (GRP g day�1) iscalculated as follows:

GRPi ¼ RGRi � DWi�1 (12)

RGRi ¼ r � 1� DWi

DWmax

� �(13)

where RGR (g seed�1 day�1) is potential relative growth rate, andDW (g seed�1) is the dry weight per seed. The initial cottonseed dryweight (DW(0), g seed�1), the maximum dry weight (DWmax,g seed�1), and r are cultivar-specific parameters.

Considering the effect of weather and crop managementmeasures (precisely N supply) on cottonseed growth, the actual

(from 15 July to 10 September) in Nanjing in 2005.

NPstoi¼

0;

ðDWi � NCmax � DWi�1 � NCi�

ðDWi � NCmax � DWi�1 � NCi�

8>><>>:

growth rate (GR, g seed�1 day�1) is calculated as:

GRi ¼ GRPi � CETSi � ENgri (14)

where CETS stands for the comprehensive effect of temperatureand solar radiation, written by Eq. (3). ENgr means the effect ofnitrogen on cottonseed growth, calculated as:

ENgri ¼LNCi

CLNCi

� �0:5

; LNCi <CLNCi

1; LNCi�CLNCi

8<: (15)

Cottonseed dry weight (DW) is calculated as follows:

DWi ¼ DWð0Þ þXi

i¼1

GRi (16)

Then the value of cottonseed index (CSI, g per 100 seeds) isobtained as 100 times of DW.

3.1.3. Cottonseed protein content module (CSPCM)

Cottonseed protein consists of the structural and storageprotein. The structural protein mainly forms at the initial stageand the storage protein forms following. The field data shows asimilar pattern of cottonseed protein content change duringcottonseed growth between different cultivars or different flower-ing dates (Fig. 2). The change process is divided into two phasesduring cottonseed growth. At initial stage, the cottonseed proteincontent (PC, %) is comparatively high and declines gradually withPDT, when the nitrogen accumulation is mainly used for structuralprotein. The decline in cottonseed N concentration is a conse-quence of N dilution. At about 12 PDT, PC declines to the minimum.Then nitrogen accumulation enters into the second phase when PCis increasing gradually and the nitrogen is mainly used for storageprotein.

Total nitrogen accumulation (NA g seed�1) is calculated as

NAi ¼ DWð0Þ � NCð0Þ þXi

i¼1

ðNstriþ Nstoi

Þ (17)

where NC(0) (%) means the initial N concentration of cottonseed, Nstr

(g seed�1 day�1) means daily accumulation of structural N, and Nsto

(g seed�1 day�1) means daily accumulation of storage N. Nstr issimulated as a result of the cottonseed growth rate (GR, g day�1) andthe N concentration (NCstr, %) in daily cumulative biomass (Eq. (18)).The NCstr is calculated as a function of PDT and N supply (Eq. (19)).

Nstri¼ GRi � NCstri

(18)

NCstri¼ 0:016þ 0:024� 1� PDTi

12

� �2 !

� ENpri; PDT<12

0:016; PDT�12

8><>:

(19)

where ENpr stands for the effect of N supply on cottonseed N

uptake, which is a result of LNC and CLNC:

ENpri ¼LNCi

CLNCi

� �2

(20)

In the second phase, the potential accumulation rate of storageN (NPsto, g seed�1) is determined by the demand of protein and theboll development. NPsto and the actual accumulation rate ofstorage N (Nsto) are calculated by:

PDT<9

1Þ � 0:104� PDTi � 9

16

� �2

; 9 � PDT � 25

1Þ � 0:104; PDT>25

(21)

Fig. 3. Observed cottonseed oil content of cv. Kemian 1 and cv. NuCOTN 33B at

different flowering dates (from 15 July to 10 September) in Nanjing in 2005.

Fig. 2. Observed cottonseed protein content of cv. Kemian 1 and cv. NuCOTN 33B at

different flowering dates (from 15 July to 10 September) in Nanjing in 2005.

W. Li et al. / Field Crops Research 112 (2009) 131–140 135

Nstoi¼ NPstoi

� CETSi � ENpri � EBP (22)

where NCmax (%) means the maximum N concentration ofcottonseed, and NC (%) means the actual N concentration. N

accumulation is closely linked to biomass accumulation, so actualstorage N accumulation is also influenced by the impact factorswhich were used in CSBAM. EBP means the effect of boll position,which can be calculated by degree days accumulated from thefirst flowering date in the plant to present flowering date. EBPlinearly increase from 0.9 to 1 with the increasing of degree daysfrom 0 to 400.

Cottonseed N concentration (NC) is the ratio of N accumulationto cottonseed biomass. The protein content (PC) is calculated as6.25 times of N concentration.

3.1.4. Cottonseed oil content module (CSOCM)

Cottonseed oil content is mainly under genetic control, but itcan be affected to some extent by environmental conditions. Thematerials required for fat synthesis mainly derived from carbohy-drates accumulated in cottonseed. Nitrogen is an essential nutrientfor the synthesis of fat which requires both N and carbon skeletons(Egelkraut et al., 2004). The change process of cottonseed oilcontent is divided into two phases during the cottonseed growth(Fig. 3). In the initial stage, oil content decline slightly, as aconsequence of the rapid biomass accumulation and slow fatsynthesis. With the acceleration in fat synthesis, the oil contentstops declining at about 12–15 PDT and then increases graduallyfollowing logistic curve.

In the module, it is assumed that daily potential oil accumula-tion (DPOA) is mainly determined by the fat synthesis capability(FSC) and oil demand. The oil demand can be expressed by sinkactivity (SA). Daily actual oil accumulation (DAOA) is influenced by

temperature and nitrogen, and also associated with boll position.They are written:

DPOAi ¼ SAi � FSCi (23)

DAOAi ¼ DPOAi � EToil � ENoili� EBP (24)

where EToil means the effect of temperature, ENoil the effect ofnitrogen, and EBP the effect of boll position on fat synthesis.

With the increasing of oil content, SA decreases gradually tillthe time when oil content gets the maximum. FSC also changeswith oil accumulation and seed development. SA and FSC can bewritten as

SAi ¼DWi � OCm� TOAi�1

DWi � OCm� DWð0Þ � OCð0Þ(25)

FSCi¼0; PDT<6;or;PDT>36

TOAi�1�p�SINPDT� 6

30�p

� �; 6 � PDT � 36

8<: (26)

where OCm means the maximum oil content (%), OC(0) means theinitial oil content(%), and TOA means the total oil accumulation perseed (g seed�1). OCm, OC(0) and p are cultivar-specific parameters.

Fat synthesis is a complex biological process strongly influ-enced by temperature. The temperature response function here isdifferent with the one for cottonseed growth:

EToilðTÞ ¼

0; T <10;or; T >35T � 10

30; 10 � T <20

T � 15

30� 15; 20 � T <30

T � 30

35� 30; 30 � T � 35

8>>>>>>><>>>>>>>:

(27)

EToili¼ 0:5� EToilðTavgÞ þ 0:25� ðEToilðTmaxÞ þ EToilðTminÞÞ (28)

Fig. 4. Comparison between simulated and observed boll maturation period of cv. Dexia 1, cv. Kemian 1, and cv. NuCOTN 33B in Nanjing in 2007.

W. Li et al. / Field Crops Research 112 (2009) 131–140136

where EToil(T) means instantaneous effect of temperature, andEToil means daily relative effect of temperature. The nitrogeneffect on oil synthesis (ENoil) is calculated by the LNC in relation toCLNC:

ENoili¼

LNCi

0:85� CLNCi; LNCi � 0:85� CLNCi

1� LNCi

CLNCi�0:85

� �� 1

0:85; 0:85� CLNCi < LNCi

8>><>>: (29)

Fig. 5. Comparison between simulated (lines) and observed (symbols) cottonseed biomas

sowing dates (25 April (a) and 25 May (b)) under three N rates (^, - - - 0 kg N ha�1; &, –

Then the total oil accumulation (TOA, g seed�1) can be writtenas:

TOAi ¼ DWð0Þ � OCð0Þ þXi

i¼1

DAOAi (30)

The oil content (OC, %) is the ratio of the TOA to cottonseedbiomass.

s accumulation of cv. Kemian 1 and cv. NuCOTN 33B at different flowering dates and

– – 240 kg N ha�1; and ~, — 480 kg N ha�1) in Xuzhou, Anyang, and Huaian, China.

Table 2Parameter values used in the model.

Parameter Explanation Value Unit

Kemian 1 NuCoTN 33B

LNCmax The maximum leaf N concentration 4.853 4.853 (%)

LNCmin The minimum leaf N concentration 2.1 2.1 (%)

DWm The maximum cottonseed dry weight 0.12 0.13 (g seed�1)

DW(0) The initial cottonseed dry weight 0.005 0.005 (g seed�1)

r Reproductive potential parameter for cottonseed 0.17 0.155 (day�1)

NCm The maximum cottonseed N concentration 4.5 4.2 (%)

NC(0) The initial cottonseed N concentration 4.0 4.0 (%)

OCm The maximum cottonseed oil content 18.0 22.0 (%)

OC(0) The initial cottonseed oil content 4.0 4.0 (%)

p Reproductive potential parameter for cottonseed oil 0.5 0.5 (day�1)

W. Li et al. / Field Crops Research 112 (2009) 131–140 137

3.2. Model calibration

Values for the model parameters were determined from theliterature or by calibration on experimental data. The parametersfor BMPM were calibrated with the data from the field experimentconducted in Nanjing in 2006, and the other modules with the datafrom the field experiment conducted in Nanjing in 2005. Based onthe observed data of boll maturation period in present study, thevarietal early maturity parameters (VE) were estimated as 0.972for cv. Dexia 1, 0.933 for cv. Kemian 1, and 0.919 for cv. NuCOTN33B. Table 2 listed out the parameter values for CSBAM, CSPCM,

Fig. 6. Comparison between simulated (lines) and observed (symbols) cottonseed prot

different sowing dates (25 April (a) and 25 May (b)) under three N rates (^, - - - 0 kg N ha

China.

and CSOCM. LNCmax and LNCmin were set by the reference of Panet al. (1996), and the others were set directly by the field data or bythe model calibration using field data. In the study, only two orthree cultivars were used for the model calibration. Adjusting thecultivar-specific parameters in the model is needed to make themodel suitable for more other cultivars.

3.3. Model validation

The accuracy of the boll maturation period module was testedusing the independent data from a field experiment conducted

ein content of cv. Kemian 1 and cv. NuCOTN 33B at different flowering dates and�1; &, – – – 240 kg N ha�1; and ~, — 480 kg N ha�1) in Xuzhou, Anyang, and Huaian,

W. Li et al. / Field Crops Research 112 (2009) 131–140138

with different cultivars and different N rates in Nanjing in 2007.The observed boll maturation period (BMP) widely ranged from 40to 65 days for different flowering dates. Results showed goodcoherence between the predicted BMP values and the observedBMP values with RMSE of 2.25 days for cv. Dexia 1, 2.61 days for cv.Kemian 1, and 2.75 days for cv. NuCOTN 33B (Fig. 4).

CSBAM, CSPCM, and CSOCM were tested using the data setsfrom the field experiments in three locations (Huaian, Xuzhou, andAnyang) belonging to two cotton growing areas in China. Cotton-seed biomass showed a wide range of values in accumulation rateunder diverse environmental conditions. Cottonseed growth ratein Xuzhou was higher than other sites, in cv. Kemian 1 was higherthan cv. NuCOTN 33B (Fig. 5). CSBAM showed good consistencywith the RMSE of 9.5 mg seed�1 for cv. Kemian 1, 8.2 mg seed�1 forcv. NuCOTN 33B, and 8.9 mg seed�1 on average.

The change of protein content during the cottonseed growthwas obviously different at different flowering dates (Fig. 6). Thecottonseed protein content variation in most of the bolls floweringbefore 25 August appeared in V-shape curve, while for those bollswhich flowered later expressed little or no increasing trend at thelate stage of development. The CSPCM also worked well with RMSEof 2.05% for cv. Kemian 1, 2.33% for cv. NuCOTN 33B, and 2.19% onaverage.

Cottonseed oil content changed with a S-shaped curve duringthe cottonseed growth and the maximum oil contents weredifferent at different treatments (Fig. 7). The CSOCM could welldescribe the oil content with RMSE of 2.45% for cv. Kemian 1, 2.95%for cv. NuCOTN 33B, and 2.71% on average. Fig. 7 shows that theCSOCM did slightly poorly in the later stage of boll development for

Fig. 7. Comparison between simulated (lines) and observed (symbols) cottonseed oil co

dates (25 April (a) and 25 May (b)) under three N rates ( , - - - 0 kg N ha�1; &, – – – 2

bolls flowering on 10 September, where the simulated values werelower than the observed ones.

4. Discussion

Accurately predicting boll maturation period is the basis tosimulate cottonseed growth and quality. The growth-degree-daywas a common and simple method when one refers to temperatureonly but lacks accuracy. To improve the accuracy of prediction,cotton model COTCO2 (Wall et al., 1994) adapted physiology timeto predict cotton development considering more environmentalfactors. Complex calculation procedure and excessive number ofmodel parameters were its main defects. PDT driven bytemperature and variety maturity profile has been used inmodeling BMP (Zhang et al., 2003). Ma et al. (2005) furtherexploited this path by introducing sunshine hours in bollmaturation period model. Nevertheless PDT model remainsimperfect in considering the environment factors as sunshinehours, but cannot perfectly reflect the sunlight condition. In thisstudy, the solar radiation and N nutrition factors are taken intoaccount besides temperature and varietal maturity. Hence, thepresent model is more comprehensive and systemic. This modulepredicted BMP more precisely with a lower RMSE of 2.54 days onaverage.

Cottonseed index, protein and oil content are the main criteriaof cottonseed quality and they differ between cultivars (Gotmareet al., 2004; Mert et al., 2005). Environmental conditionsalso strongly influenced the seed quality (Turner et al., 1976).Some cultivar-specific parameters and response functions are

ntent of cv. Kemian 1 and cv. NuCOTN 33B at different flowering dates and sowing

40 kg N ha�1; and ~, — 480 kg N ha�1) in Xuzhou, Anyang, and Huaian, China.

W. Li et al. / Field Crops Research 112 (2009) 131–140 139

incorporated in the present model to illustrate the effects ofcultivars, weather, soil, and crop management measures (such as N

supply). This paper combined with the physiological knowledgeand developed a mechanistic simulation model of cottonseedquality formation. The model was based on the hypothesis that thepotential cottonseed biomass accumulation, N accumulation, andoil synthesis are sink-determined. The hypothesis of ‘‘sink-determination’’ has also been used in modeling organ growth ornutrients accumulation for many other crops, such as tomato fordry matter distribution (Heuvelink, 1996), wheat for proteincontent (Martre et al., 2006) and for nitrogen uptake (Jamieson andSemenov, 2000). Our study addressed the cottonseed potentialgrowth rate by logistic function which has been widely used inmodeling crop biomass accumulation (Bange and Milroy, 2004).The major features of the logistic method are simple structure,understandability, and stability in running. Logistic model was asuitable method in modeling cottonseed biomass accumulation.Integrated with response functions of environment conditions, themodel predicted cottonseed dry weight with RMSE of 8.9 mg.Combined with the physiological knowledge in cottonseed N

uptake, the present model simulated the change of cottonseedprotein content and maintained a high accuracy with RMSE of2.19%. The model simulated oil potential accumulation through oildemand and synthesis capability. The result in modeling oilcontent is validated with RMSE of 2.71%.

Adequate N supply can raise cottonseed biomass and proteincontent, and lower oil content (Sawan et al., 1985, 2001, 2006). Thesubtending leaf of cotton boll is the main source organ for the bollgrowth, and the N concentration in subtending leaf directlyinfluence the cottonseed growth and development. It is hencemore relevant to link the cottonseed growth and development tothe N concentration in subtending leaf rather than the whole plantN concentration or N rate. Our study simulated the effect of N

nutrition using the subtending leaf N concentration as a directindicator, and developed a semi-empirical model to illustrate thechange of actual and critical N concentration in subtending leafduring the cottonseed growth. The quantitative studies in critical N

concentration of crops were rich (Pan et al., 2006; Keating et al.,1999). Xue et al. (2006) also qualified the cotton plant critical N

concentration. But few reports concerned the subtending leafcritical N concentration for cottonseed growth and development.By using different response functions, the N nutrition effect wasconsidered in each module of the model. In comparison to previousstudies which have quantified the relationship of cottonseed N

concentration and N rates by linear equations (Egelkraut et al.,2004), the present model appears to be more reasonable.

The field data were obtained from four locations in 2005, andprovided a wide range of environmental conditions and cropcharacteristics for validation purposes. The validation led to goodconsistency between the simulated and the observed values,indicating that the model is robust. However, the simulated valuesof cottonseed oil content were inferior to observed ones in the latestage of development for late flowering bolls. This phenomenonoccurs in November when the meteorological conditions weremuch less favorable. This is an indication that the model cannotwork well to simulate cottonseed oil content under extremeenvironmental conditions. The verification of the model on abroader set of data remains needed, through additional fieldexperiments in more regions with more cultivars. Efforts to fine-tune the model should be concentrated on the formation ofcottonseed quality in more variable environmental conditions.

5. Conclusion

The model we have developed are capable of predicting bollmaturation period and simulating cottonseed growth, protein and

oil content driven by the inputs of cultivar parameters, weather,and crop management variables (N fertilization). Our studyimproved cotton boll maturation model by quantifying theresponse to solar radiation and N supply. This model structurewas based on the processes of biomass accumulation, N uptake andfat synthesis in cottonseed. The responses of cottonseed growthand quality formation to weather or crop management werequantified respectively. The field data set conducted in wide rangeof climate conditions with different cultivars cotton planted ondifferent dates was used for model validation. The result showedthat the model is robust enough to accurately predict cotton bollmaturation period, cottonseed biomass, protein and oil content,with the RMSE of 2.54 days, 8.9 mg, 2.19% and 2.71% on average,respectively. Its main advantages lie upon comprehensiveintegration of inputs as compared to existing models, reasonablemodel algorithms, higher accuracy, and application to a diverserange of environment conditions. The model can be not only usedindependently to predict cottonseed quality formation, but alsoused to predict the cottonseed, protein and oil yield by integrationwith a cotton growth model.

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Nos. 30771277 and 30771279).

References

Ahmad, S., Anwar, F., Hussain, A.I., Ashraf, M., Awan, A.R., 2007. Dose soil salinityaffect yield and composition of cottonseed oil? J. Am. Oil Chem. Soc. 84, 845–851.

Bange, M.P., Milroy, S.P., 2004. Growth and dry matter partitioning of diverse cottongenotypes. Field Crop Res. 87, 73–87.

Boone, M.Y., Porter, D.O., McKinion, J.M., 1993. Calibration of GOSSYM: theory andpractice. Comput. Electron. Agric. 9, 193–203.

Cao, W., Moss, D.N., 1997. Modelling phasic development in wheat: a conceptualintegration of physiological components. J. Agric. Sci. 129, 163–172.

Luque De Castro, M.D., Garcia-Ayuso, L.E., 1998. Soxhlet extraction of solid materi-als: an outdated technique with a promising innovative future. Anal. Chim. Acta369, 1–10.

Chen, G., Yu, Y., 2001. Preliminary study on the temperature-light effects on bolldevelopment. Cotton Sci. 13, 63–64 in Chinese, with English abstract.

Chen, B., Cao, W., Zhou, Z., 2006. Simulation and validation of dry matter accumula-tion and distribution of cotton boll at different flowering stages. Sci. Agric. Sin.39, 487–493 in Chinese, with English abstract.

Duncan, W.G., 1973. SIMCOT: a simulation of cotton growth and yield. In: Murphy,C., Hesketh, J.D., Strain, B. (Eds.), Modelling the Growth of Trees. Oak RidgeNational Laboratory, Oak Ridge, pp. 115–118.

Egelkraut, T.M., Kissel, D.E., Cabrera, M.L., Gascho, G.J., Adkins, W., 2004. Nitrogenconcentration in cottonseed as an indicator of N availability. Nutr. Cycl. Agroe-cosyst. 68, 235–242.

Feil, B., Moser, S.B., Jampatong, S., Stamp, P., 2005. Mineral composition of the grainsof tropical maize varieties as affected by pre-anthesis drought and rate ofnitrogen fertilization. Crop Sci. 45, 516–523.

Fye, R.E., Reddy, V.R., Baker, D.N., 1984. The validation of GOSSYM: Part 1—Arizonaconditions. Agr. Syst. 14, 85–105.

Gotmare, V., Singh, P., Mayee, C.D., Deshpande, V., Bhagat, C., 2004. Geneticvariability for seed oil content and seed index in some wild species andperennial races of cotton. Plant Breed. 123, 207–208.

Hanan, J.S., Hearn, A.B., 2003. Linking physiological and architectural models ofcotton. Agric. Syst. 75, 47–77.

Hearn, A.B., 1994. OZCOT: a simulation model for cotton crop management. Agric.Syst. 44, 257–299.

Hearn, A.B., daRoza, G.D., 1985. A simple model for crop management 1 applicationsfor cotton 2 (Gossypium hirsutum L). Field Crops Res. 12, 49–69.

Heuvelink, E., 1996. Dry matter partitioning in tTomato: validation of a dynamicsimulation model. Ann. Bot. 77, 71–80.

Jackson, B.S., Arkin, G.F., Hearn, A.B. (Eds.), 1988. The cotton simulation model‘‘COTTAM’’: fruiting model calibration and testing. Trans. ASAE 31, 846–854.

Jallas, E., Martin, P., Sequeira, R., Turner, S., Cretenet, M.G.E., 2000. Virtual COTONS1,the firstborn of the next generation of simulation model. Virtual Worlds 1834,235–244.

Jamieson, P.D., Semenov, M.A., 2000. Modelling nitrogen uptake and redistributionin wheat. Field Crops Res. 68, 21–29.

Jamieson, P.D., Semenov, M.A., Brooking, I.R., Francis, G.S., 1998. Sirius: a mechan-istic model of wheat response to environmental variation. Eur. J. Agron. 8,161–179.

W. Li et al. / Field Crops Research 112 (2009) 131–140140

Keating, B.A., Robertson, M.J., Muchow, R.C., Huth, N.I., 1999. Modelling sugarcaneproduction systems I. Development and performance of the sugarcane module.Field Crops Res. 61, 253–271.

King, E.E., Leffler, H.R., 1979. Nature and patterns of proteins during cotton seeddevelopment. Plant Physiol. 63, 260–263.

Ko, J., Maas, S.J., Lascano, R.J., Wanjura, D., 2005. Modification of the GRAMI modelfor cotton. Agron. J. 97, 1374–1379.

Larson, J.A., Mapp, H.P., Verhalen, L.M., Banks, J.C., 1996. Adapting a cotton model fordecision analyses: a yield response evaluation. Agric. Syst. 50, 145–167.

Ma, F., Cao, W., Zhang, L., Zhu, Y., Li, S., Zhou, Z., Li, C., Xu, L., 2005. A physiologicaldevelopment time-based simulation model for cotton development stages andsquare and boll formation. Chin. J. Appl. Ecol. 16, 626–630 in Chinese, withEnglish abstract.

Martre, P., Jamieson, P.D., Semenov, M.A., Zyskowski, R.F., Porter, J.R., Triboi, E.,2006. Modelling protein content and composition in relation to crop nitrogendynamics for wheat. Eur. J. Agron. 25, 138–154.

Meng, Y., Cao, W., Zhou, Z., Liu, X., 2003. A process-based model for simulatingphasic development and phenology in rice. Sci. Agric. Sin. 36, 1362–1367 inChinese, with English abstract.

Mert, M., Aki, Y., Gen, O., 2005. Genotypic and phenotypic relationships of lint yield,fibre properties and seed content in a cross of two cotton genotypes. Acta Agric.Scand. B: Soil Plant Sci. 55, 76–80.

Mutsaers, H.J.W., 1984. KUTUN: a morphogenetic model for cotton (Gossypiumhirsutum L). Agric. Syst. 14, 229–257.

Pan, X., Han, X., Shi, Y., 1996. A cotton growth and development simulation modelfor culture management—COTGROW. Sci. Agric. Sin. 29, 94–96 in Chinese, withEnglish abstract.

Pan, J., Zhu, Y., Jiang, D., Dai, T., Li, Y., Cao, W., 2006. Modeling plant nitrogen uptakeand grain nitrogen accumulation in wheat. Field Crops Res. 97, 322–336.

Plant, R.E., Kerby, T.A., Zelinski, L.J., Munk, D.S., 1998. A qualitative simulation modelfor cotton growth and development. Comput. Electron. Agric. 20, 165–183.

Rabadia, V.S., Thaker, V.S., Singh, Y.D., 1999. Relationship between water contentand growth of seed and fibre of three cotton genotypes. J. Agron. Crop Sci. 183,255–261.

Ravula, P.P., Grisso, R.D., Cundiff, J.S., 2008. Cotton logistics as a model for a biomasstransportation system. Biomass Bioenerg. 32, 314–325.

Reddy, V.R., 1994. Modeling cotton growth and phenology in response to tempera-ture. Comput. Electron. Agric. 10, 63–73.

Reddy, V.R., Baker, D.N., 1988. Estimation of parameters for the cotton simulationmodel GOSSYM: cultivar differences. Agric. Syst. 26, 111–122.

Reddy, V.R., Baker, D.N., 1990. Application of GOSSYM to analysis of the effects ofweather on cotton yields. Agric. Syst. 32, 83–95.

Reddy, K.R., Koti, S., Davidonis, G.H., Reddy, V.R., 2004. Interactive effects of carbondioxide and nitrogen nutrition on cotton growth, development, yield, and fiberquality. Agron. J. 96, 1148–1157.

Roussopoulos, D., Liakatas, A., Whittington, W.J., 1998. Controlled-temperatureeffects on cotton growth and development. J. Agric. Sci. 130, 451–462.

Sawan, Z.M., El-Kasaby, A.T., Sallouma, B.M., 1985. Effect of plant density, nitrogenfertilization and growth regulators on cottonseed yield and seedling vigour. J.Agron. Crop Sci. 154, 120–128.

Sawan, Z.M., El-Farra, A.A., El-Latif, S.A., 1988. Cottonseed, protein and oil yields, andoil properties as affected by nitrogen and phosphorus fertilization and growthregulators. J. Agron. Crop Sci. 161, 50–56.

Sawan, Z.M., Hafez, S.A., Basyony, A.E., 2001. Effect of nitrogen fertilization andfoliar application of plant growth retardants and zinc on cottonseed, proteinand oil yields and oil properties of cotton. J. Agric. Crop Sci. 186, 183–191.

Sawan, Z.M., Saeb, A., Hafez, A.E.B., Alkassas, A.R., 2006. Cottonseed, protein, oilyields and oil properties as affected by nitrogen fertilization and foliar applica-tion of potassium and a plant growth retardant. World J. Agric. Sci. 2, 56–65.

Tang, X., Tian, C., Lu, Z., 2003. Relationship between the spread quantity ofnitrogenous fertilizer and the accumulated sunshine requirement of cottonin its growing stages. Arid Zone Res. 20, 226–229 in Chinese, with Englishabstract.

Thaker, V.S., Saroop, S., Vaishnav, P.P., Singh, Y.D., 1989. Genotyopic variations andinfluence of diurnal temperature on cotton fibre development. Field Crops Res.22, 129–141.

Turner, J.H., Ramey, H.H., Worley, J.S., 1976. Influence of environment on seedquality of four cotton cultivars. Crop Sci. 16, 407–409.

Wall, G.W., Amthor, J.S., Kimball, B.A., 1994. COTCO2: a cotton growth simulationmodel for global change. Agric. For. Meteorol. 70, 289–342.

Wells, A.T., Hearn, A.B., 1992. OZCOT: a cotton crop simulation model for manage-ment. Math. Comput. Simulat. 33, 433–438.

Xue, X., Zhou, Z., Zhang, L., Wang, Y., Guo, W., Chen, B., 2006. Development andapplication of critical nitrogen concentration Dilution Model for cotton afterflowering. Acta Ecol. Sin. 26, 1781–1791 in Chinese, with English abstract.

Yang, Y., Ouyang, Z., Yang, Y., Liu, X., 2008. Simulation of the effect of pruning andtopping on cotton growth using COTTON2K model. Field Crops Res. 106, 126–137.

Zhang, L., Cao, W., Zhang, S., Luo, W., 2003. Simulation model for cotton develop-ment stage based on physiological development time. Cotton Sci. 15, 97–103 inChinese, with English abstract.