Content uploaded by Tanvir Shahzad
Author content
All content in this area was uploaded by Tanvir Shahzad on Apr 21, 2015
Content may be subject to copyright.
This article was downloaded by: [Tanvir Shahzad]
On: 21 April 2015, At: 05:46
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Archives of Agronomy and Soil Science
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/gags20
Using expert knowledge data to validate crop models
on local situation data
Faisal Mahmooda, Jacques Weryb, Sabir Hussaina, Tanvir Shahzada, Muhammed Arslan
Ashrafc, Olivier Therondd & Hatem Belhouchettee
a Department of Environmental Sciences & Engineering, Government College University,
Faisalabad, Pakistan
b SupAgro Montpellier UMR-System, Montpellier, France
c Department of Botany, Government College University, Faisalabad, Pakistan
d INRA, UMR 1248 AGIR, 31326 Castanet-Tolosan, France
e CIHEAM-IAMM, Montpellier, France
Accepted author version posted online: 21 Apr 2015.
To cite this article: Faisal Mahmood, Jacques Wery, Sabir Hussain, Tanvir Shahzad, Muhammed Arslan Ashraf, Olivier
Therond & Hatem Belhouchette (2015): Using expert knowledge data to validate crop models on local situation data,
Archives of Agronomy and Soil Science
To link to this article: http://dx.doi.org/10.1080/03650340.2015.1043528
Disclaimer: This is a version of an unedited manuscript that has been accepted for publication. As a service
to authors and researchers we are providing this version of the accepted manuscript (AM). Copyediting,
typesetting, and review of the resulting proof will be undertaken on this manuscript before final publication
of the Version of Record (VoR). During production and pre-press, errors may be discovered which could
affect the content, and all legal disclaimers that apply to the journal relate to this version also.
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Accepted Manuscript
Publisher: Taylor & Francis
Journal: Archives of Agronomy and Soil Science
DOI: 10.1080/03650340.2015.1043528
Using expert knowledge data to validate crop models on local situation data
Faisal Mahmooda*, Jacques Weryb, Sabir Hussaina, Tanvir Shahzada, Muhammed Arslan
Ashrafc, Olivier Therondd, Hatem Belhouchettee
a Department of Environmental Sciences & Engineering, Government College University,
Faisalabad, Pakistan; b SupAgro Montpellier UMR-System, Montpellier, France;
c Department of Botany, Government College University, Faisalabad, Pakistan; d INRA, UMR
1248 AGIR, 31326 Castanet-Tolosan, France, e CIHEAM-IAMM, Montpellier, France
Corresponding author. E-mail: fslagronomy@hotmail.com
Abstract
Cropping system models are widely used tools for simulating the growth and development of
crops at field scale. However, it is often difficult to satisfy their detailed input and output data
requirements for a proper evaluation of model. In this study, expert knowledge data were used
as alternative source to fulfill these data requirements. The model was first calibrated for
major crops of the studied area and then evaluated for the same crops by using expert
knowledge data. Results showed that the model accurately simulated above-ground biomass
and grain yield with a relative root mean square error (RRMSE) of 20 and 17%, respectively.
On the other hand, simulated results were less satisfactory for N uptake and cumulated
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
evapotranspiration with RRMSE of 27% and 31%, respectively. The model simulated
cumulative variables more accurately than dynamic variables. The results of this study
suggest that expert knowledge can be used to get data for intermediate variables rarely
measured in experiments used for calibration (green LAI, actual evapotranspiration, rooting
depth) in typical crop management conditions in the region. This approach enables a global
and dynamic evaluation of cropping system models when experimental data is unavailable for
large heterogeneous areas in a region.
Keywords: APES model, cropping system model, dynamic model evaluation, Midi-Pyrénées
region
Introduction
Cropping system models are useful tools for simulating the growth and development of crops
at field scale under diverse conditions of soil, climate and management (e.g. EPIC: Williams
et al. (1989); APSIM: McCown et al. (1996); DSSAT: Jones et al. (2003); CropSyst: Stöckle
et al. (2003); Corre-Hellou et al. (2009). For such purposes, there is need to evaluate the
models for their ability to simulate the key variables of crop phenology, growth, yield and
water and N balances using experimental data in the range of cropping conditions
representative of the targeted use. In order to evaluate the model's performance in simulating
the particular biophysical conditions, calibration and evaluation should be done on two
different sets of independent experimental data (Odum 1983; Shugart 1984; Jorgensen 1986;
Power 1993). However, it is often difficult to find an independent set of data for model
evaluation i.e. the data that have not been used for calibration (Stöckle et al. 2003). The main
reason is that the observed data needed for model evaluation mainly require destructive
observation that are usually time consuming, costly and performed under limited soil, crop
management and climate conditions.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
In order to assess model performance in a large range of cropping conditions, two
steps are usually followed (Oreskes et al. 1994; Belhouchette et al. 2008, 2011). Firstly, the
crop model is calibrated with several dynamic and cumulative variables (e.g. yield, biomass,
LAI, N-leaching) but under limited cropping conditions and then the crop model is validated
for a wider range of cropping systems but usually only for crop yield which is the common
variable measured in all crop experiments (Bouman et al. 1996; Jagtap & Jones 2002; Van
Ittersum et al. 2003; Faivre et al. 2004; Therond et al. 2011). In many studies dealing with
cropping systems analysis at regional scale i.e. covering a large range of crop, management,
soil, and climate conditions (see for example Belhouchette et al. 2011), the input-output data
used to describe cropping systems are obtained through farmer surveys or existing regional
databases (Middelkoop & Janssen 1991; Faivre et al. 2004; Clavel et al. 2011; Therond et al.
2011). These sources of information have many drawbacks; e.g. (i) they lack detailed
information on soil climate conditions such as limited or unlimited water and nitrogen
conditions as well as management practices (e.g. dates and rates of irrigation and fertilization)
(Launay & Guérif 2005); (ii) surveys require long and costly data collection (Biarnés et al.
2004); (iii) they do not take into account the interactive effects of soil, climate and
management on output data and (iv) the cropping system performance is generally described
only with cumulative variables, especially yield.
On the other hand, there are regional experts who have detailed knowledge on the
crop growing conditions in the region, their behaviour during crop cycle and their
performances (Clavel et al. 2011). This knowledge, gained during years of field experiments,
field’s surveys and interactions with farmers, is often not expressed in terms of input-output
variables of a crop model (El Hajj et al. 2009). However, it is usually used for
recommendations and extension services. The present study was focused on analyzing
whether it is possible to elicit expert knowledge in a model-compatible format, both for
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
cumulative and dynamic variables, in order to use this “expert dataset” to evaluate a cropping
system model for crop yield and externalities at regional level in the Midi-Pyrénées region of
France. A specific protocol for this expert knowledge elicitation and its use for model
evaluation has been developed and tested in the Midi-Pyrénées region.
Materials and methods
Description of the study area
The study area was in the Midi-Pyrénées region, in the south-west of France. It is a French
region with the highest number of farms with an agricultural area of about 2 540 000 ha and
mainly devoted to livestock and arable crops. In this study, we considered only the arable
zone (Gers Department), which accounts for approximately 40% of the cultivated area of the
region (Belhouchette et al. 2011). In this zone, a wide range of agronomic conditions
including crops, soils, crop management and weather (rainfall and temperature) can be
observed. The main cultivated crops are cereals (durum wheat, soft wheat, maize and barley),
legumes (soybean, peas and fababean) and oilseeds (sunflower and rape). There are mainly
two soil types (loam and clay loam), which can be sub-divided into different types depending
on the soil texture and depth. Major crops that are usually grown on loamy soil are irrigated
maize combined with durum wheat, sunflower and peas. On clay loam, usually grown major
crops are durum and soft wheat combined with sunflower (Nolot & Debaeke 2003).
The arable zone has a temperate climate with temperatures increasing from south-east
to north-west, while rainfall and evapotranspiration increase from east to west. The arable
zone represents only 9% of the total cultivated area as irrigable; consequently rainfed annual
grain crops are predominant in the region (Belhouchette et al. 2011). Long-term
meteorological data indicate that the region is characterized by irregular and variable seasonal
and yearly rainfall. The mean annual rainfall for 1996-2002 period was 701 mm/yr (σ = 91
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
mm/yr). As a consequence, crop yields vary from year to year depending on weather, soil
type, and water and nitrogen management (Belhouchette et al. 2011).
APES model
The APES (Agricultural Production and Externalities Simulator) model (Donatelli et al. 2010)
was used for this study. It is a modular model developed within the SEAMLESS project (Van
Ittersum et al. 2008), as a part of a modeling chain enabling ex-ante impact assessment of
agricultural and environmental policies and technological innovations in different EU regions.
It is a multi-year, multi-crop and daily time step simulation model used for estimating the
biophysical behaviour of land-bound agricultural activities at field scale featuring a wide
range of climate, soil and agro-technical management options (Donatelli et al. 2010). It can
simulate soil-water budget, soil-plant nitrogen budget, crop phenology, crop canopy
development, root growth, biomass production and partitioning, crop yield, soil erosion and
soil carbon budget. It is a flexible and generic model which facilitates the adjustment of
model structure depending on the simulation goal (Adam et al. 2012), and can be adapted to
different environments and management practices for a wide range of biophysical conditions
(e.g. soil, rainfall), type of crops, land use or agro-management systems (cereal, legume crops,
and oil crops) as described in this study.
Experimental data
The model was calibrated by using experimental data collected for the main crops including
durum wheat (Triticum durum), sunflower (Helianthus annuus), maize (Zea mays) and pea
(Pisum sativum) cultivated in the arable zone of the Midi-Pyrénées region. The data of
cropping system experiment conducted by INRA in the region Auzeville near Toulouse
(latitude 43° 32’ N, longitude 1° 28’ E) between 1996 and 2002 (Nolot & Debaeke 2003) was
used for this purpose. Soil samples were taken up to 1.5 m in depth before sowing in order to
determine the initial soil moisture, soil mineral nitrogen (NO3-N) and organic matter content.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Soil texture in percentage of sand, silt and clay was measured for two layers i.e. 0-30 cm and
for maximum depth of the soil. Two types of soils were identified; clay loam and loam, with
organic matter content ranging from 0.8 to 1.41% for the 0-30 cm soil layer (Table 1).
Volumetric water content at permanent wilting point (PWP) and field capacity (FC), and
maximum bulk density (BD) were estimated from soil texture using the pedotransfer
functions provided by SoilPAR (Acutis & Donatelli 2003) (Table 1). The weather data i.e.
daily maximum and minimum temperature, precipitation, wind speed, global solar radiation
and maximum and minimum relative air humidity were recorded at the experimental site.
Management practices were described such as sowing and harvesting dates as well as the
dates and amounts of irrigation and fertilization. The key phenological stages such as
flowering, grain filling and physiological maturity were noted and grain yield, above ground
biomass and above ground N uptake were measured at harvest (Table 2).
Procedure for model evaluation
The model was evaluated (for cumulative and dynamic variables) for four crops (durum
wheat, sunflower, peas and maize grain), under typical combinations of soil and management
of the Midi-Pyrénées region, for specific type of climatic years. A 3-step evaluation protocol
was defined and applied in interaction with four experts from the region (Figure 1). These
experts had the skills and expertise of given crops behaviour based on surveys and various
experiments conducted under the actual field conditions. The work with the experts was
accomplished during a one-day workshop. It is assumed that, if experts are properly chosen,
then this local knowledge is accurate and reliable enough to be used as a reference to which
the model simulation were compared.
Step 1: Input data
Selection of representative agricultural activities in the region. In order to represent the
cropping system diversity of the arable zone (Gers department), the representative activities
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
(i.e. combinations of crop, crop management and soil, as defined by Belhouchette et al. 2011)
were selected by the experts. Initially three crop families, i.e. cereals, oil crops and protein
crops were identified. Then, the major crops of each family in the region were selected, which
included the maize and durum wheat as cereal, sunflower as oil crop and peas as protein
crops. Each expert identified activities for which he had enough confidence to characterize
them in terms of crop management and then of crop behavior (yield, biomass, etc.). Overall,
twelve representative agricultural activities were selected (Table 3). These activities included
four crops (maize, durum wheat, sunflower and peas), two soil types (loam and clay loam)
and, except for maize, two climatic conditions (wet and dry year). The maize crop, being an
irrigated crop, is cultivated under potential and limited water conditions. Durum wheat and
sunflower are cultivated on both soil types, while peas and maize are cultivated only on loam
soil (Table 3).
Crop management data. The experts also provided average crop management data, i.e. sowing
dates along with the dates and amounts of fertilization and irrigation for each individual
activity (Table 3). The management data were specified by taking into account the soil type
(clay loam, loam), year type for rainfed crops (wet, dry) and water condition (limited and
unlimited) for maize. Experts identified recent real year corresponding to the two year types
(wet, dry). Experts considered a year wet and dry, when the average rainfall ranged
respectively from 700 to 920 mm/year and 530 to 600 mm/year respectively.
Initial soil conditions, soil and climate data. The initial water and nitrogen conditions are
difficult to measure and rarely assessed in experiments and farmers fields. They depend
mainly on the previous crop in the rotation cycle, the capacity of the soil to retain water and
nitrogen and the rainfall pattern during the intercropping period (Leenhardt et al. 2006). It was
difficult to get data for initial soil conditions for several years; therefore, we assumed that
experts with experience of experimentation for many years (10 to 15 years) know the initial
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
soil conditions. Consequently the initial soil conditions were fixed for each crop by soil type
by the experts and deficiencies in the data give by experts were fulfilled by choosing the value
used for calibration (Table 3). The weather data of the climatic year specified by each expert
were provided by the local meteorological station. We used daily values of rainfall, minimum
and maximum temperature, minimum and maximum relative air humidity, solar radiation and
wind speed.
Step 2: Output data
For each activity (crop by soil, climate and management type), each expert was also asked to
fill specific tables and graphs for an imposed set of output variables of the model:
• cumulative variables in the form of table: above ground biomass, grain yield,
above ground N uptake and cumulated evapotranspiration at harvest (Table 3).
• dynamic variables in the form of curves: above ground biomass, leaf area index,
above ground N uptake, cumulated evapotranspiration and rooting depth (see an
example in Figure 2).
They had to provide these data before we run the APES model for the specific
situation, in order to ensure that the expert-elicited data are independent of the simulated data.
The experts were allowed to communicate during the identification of the activities and the
description of the input data, in order to ensure complementarity and consistency among the
situations described as well as exchange of information on soils and climate. But they were
not allowed to communicate during the description of the output variables, in order to avoid
interactions in the way they expressed their results. After this process, we then asked each
expert to describe what was the approach he considered to draw the curves for the dynamic
variables? We found that all the four experts used a similar approach that: i) determines the
number of days required to reach a specific stage of crop cycle, i.e. flowering, grain filling
and physiological maturity; ii) determines the value of the cumulative variable at each
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
phenological stage and mark it; iii) plots the curve between these mark points and the origin
with a sigmoid type of curve as shown in figure 2. We noted that some dynamic curves were
missing as the experts felt that they were unable to plot the curves for some dynamic
variables, i.e. green LAI, cumulated evapotranspiration, and root depth for durum wheat
(activities 3 and 4) and all the curves for the peas crop (activities 5 and 6).
Step 3: Criteria for model evaluation
The method for global and dynamic model evaluation was also determined through discussion
with the experts. The calibrated APES model was run for each activity by using input data
given in step1. Simulated output data were then compared with the experts’ output data both
for cumulative and dynamic variables. A global evaluation of the model, conducted on
cumulative variables, was achieved by using relative root mean square error (RRMSE)
(Loague & Green 1991). The over and underestimation of simulated variables, compared to
the expert values, were calculated by coefficient of residual mass (CRM).
Where:
Si: simulated data, Oi: observed data
Wherever possible, simulated and expert data were evaluated by using other data
provided by the literature for biophysical and management conditions similar to the simulated
activity. The dynamic evaluation of the model was done with dynamic variables i.e. above
ground biomass, green LAI, actual cumulated evapotranspiration, above ground N uptake and
root depth at key phenological stages i.e. flowering, grain filling and physiological maturity.
Correlation between simulated and expert data at these stages was assessed by relative root
mean square error (RRMSE) (Loague & Green 1991).
()
==
−
n
i
n
i
SiOi
11
=
n
i
Oi
1
CRM= × 100
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Results and discussion
Model calibration
APES model was calibrated for phenological stages of flowering, grain filling and
physiological maturity by using the experimental data. Then the calibration was done for light
interception, biomass production and nitrogen parameters to match the observed and
simulated data of above ground biomass (AGB), grain yield and N uptake. Some of these
parameters were fixed as default values which were derived from previous studies (Adam et al.
2012, 2013; Therond et al. 2011) (data not shown). The values of all parameters were adjusted
within a reasonable range of variation based on previous research and expert knowledge
(Donatelli et al. 2002). In order to ensure a good correlation between observed and simulated
data sets, the adjustment process was stopped when further modification of crop parameters
values generated little or no change on the basis of the relative root mean square error
(RRMSE) (Loague and Green 1991).
Where:
Si: simulated data, Oi: observed data, Ō: average observed data and n: observation number.
Statistical analysis shows that the model accurately simulated the crop phenology with
a RRMSE ranging from 3 to 19% for all phenological stages of flowering, grain filling and
physiological maturity, except for maize, for which the model simulated the crop flowering
with less accuracy (RRMSE of 39%), (data not shown). The calibrated model also
successfully simulated the above ground biomass, grain yield and N uptake for all crops of the
experiment, with the RRMSE ranging from 6 to 15% for biomass, 8 to 14% for grain yield
RRMSE =
−
=
n
iSiOi
n1
2
)(
1
Ōi
× 100
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
and 6 to 14% for N uptake. For the latter variable (N uptake), sunflower and pea were poorly
simulated, with an RRMSE of 28% (Table 4).
Model evaluation for cumulative and dynamic variables
Above ground biomass (AGB) and grain yield
The model accurately predicted the cumulative AGB and grain yield for all activities with a
RRMSE of 20 and 17% for AGB and grain yield, respectively (Figure 3 (a) & (b)). All the
data pairs are close to the 1:1 line and the activities with over or under-estimation are same for
grain yield and above ground biomass. The model also well predicted the dynamic AGB in
most of the activities with a RRMSE of 32, 19 and 18% for flowering, grain filling and
physiological maturity respectively (Figure 4). Statistical analysis shows that model-predicted
values were closer to the expert-given values of grain yield and above ground biomass. The
consistency of over and under-estimation of grain yield and above ground biomass for similar
data pairs might be due to the fact that experts declared, after filling the table, that for the
given activity, they were generally estimated grain yield and then above ground biomass by
using typical crop harvest index.
Above ground N uptake
Statistical analysis shows that, for all the activities, the model predicted the cumulative N
uptake less accurately (RRMSE = 27%) than for biomass and grain yield. The model under-
predicted the N uptake with an average CRM value of 16% (Figure 3 (c)). Similar to
cumulative N uptake, APES model also predicted the dynamic N uptake with less accuracy as
shown in figure 5. Overall, the model under-predicted the N uptake for all activities with a
CRM value of 20% for all phenological stages. The discrepancy between simulated and
expert given N uptake was much higher for the pea and maize crops (activities 5, 6 and 11, 12
respectively). The simulated data for pea crop shows that total N uptake ranged from 150 to
180 kg ha-1, which is lower than the average value of 250 kg ha-1 provided by experts. In this
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
study, initial modeling solution of APES was used, which had the tendency of
underestimation of crop nitrogen uptake. These results were also confirmed by Adam et al.
(2012), where APES model also underestimated the crop nitrogen uptake in maize crop due to
fraction of soil nitrogen absorbed by the microbes. Similarly, in another study, Adam et al,
(2013) reported the lower contribution of N2 fixation in total crop nitrogen uptake in pea crop
under water stress conditions. These results also confirmed the findings of Wery (1996) that,
under potential growth conditions, N fixation contribute more to total crop N uptake as
compared to water limited conditions, which also reduced the above ground biomass by 20-
40% and consequently the N fixation and total crop N uptake.
Actual cumulated evapotranspiration (ETC)
Statistical analysis showed a discrepancy between simulation and expert data for cumulative
ETC with a higher value of RRMSE of 31%. Most of the data pairs are far away from the 1:1
line. Overall the model under-predicted the ETC with a CRM value of 15% (Figure 3 (d)). In
case of dynamic evaluation, a large difference was also observed between simulated and
experts' ETC curves as shown in figure 6. Similar to cumulative evaluation, for all
phenological stages, the model also under-predicted the ETC with a CRM value of 32%, but
this under-prediction was higher at flowering (CRM of 38%) than at grain filling (CRM of
33%) and physiological maturity (CRM of 25%). For all crops, the bad prediction of
cumulated evapotraspiration was due to the model, as for similar environmental and soil
conditions, the ETC values reported in literature are similar to the values given by the experts.
For example, Nolot & Debaeke (2003) reported average ETC of 560 mm for durum wheat,
715 mm for maize, 586 mm for sunflower and 562 mm for pea. Other studies also showed the
same range of ETC for almost similar biophysical conditions, e.g. 444 to 570 mm for durum
wheat, 483 to 660 mm for maize and 470 mm for sunflower (Claude Mailhol et al. 1997;
Ruget et al. 2002; Utset et al. 2004). For all crops, except for pea, these amounts are close to
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
the value given by the experts. The underestimation of ETC by the APES model is consistent
with the observation done by Adam et al. (2009) that the plant available water is usually
underestimated in the APES model.
Green Leaf area index (LAI)
The statistical analysis showed that, for all the activities, the model predicted the LAI
dynamic with a RRMSE of 17 and 20% respectively for flowering and grain filling stages.
Most of the data pairs are close to the 1:1 line with a higher correlation at grain filling stage
than at flowering (Figure. 7).
Rooting depth
The statistical analysis showed that the model predicted the rooting depth at all phenological
stages with a RRMSE of 23, 16 and 15% respectively for flowering, grain filling and
physiological maturity (Figure 8).
Global evaluation of cumulative variables for all activities showed that the model
simulated wet climatic conditions relatively more accurately than dry ones. For all cumulative
variables, the difference (CRM) between simulated and experts given variable values were
higher for dry conditions as compared to wet conditions. Adam et al. (2009) also found that
APES is less sensitive to water stress conditions for above ground biomass and also its
possible residual effect on N uptake.
Statistical analysis shows that same trend was observed for simulated and expert given
dynamic curves of above ground biomass, LAI and rooting depth as compared to N uptake
and cumulated evapotranspiration. For N uptake and ETC the simulated results were
consistent with the simulated results of cumulative variables. Moreover, the better simulation
of later stages (physiological maturity) of crop development than the earlier ones (flowering
and grain filling) might be due to the fact that, for drawing the dynamic curves for all
variables, the experts kept in their mind the cumulative variable value at main phenological
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
stages, which obviously resulted in better simulation of later stages (physiological maturity)
of crop development than the earlier ones.
Methodology analysis
We found only a single seven-year dataset for four major crops in one location of the region
(INRA experimental station in Toulouse, France) with several key crop variables measured at
harvest and at some phenological stages (Nolot & Debaeke 2003). These data were precious
to conduct the necessary work of parameterization of crop specific parameters (Wallach et al.
2002). Nowadays, for model-based assessment of agricultural systems in a region, it is a
common practice to use the simplified regional data sets for the evaluation of the models
(Leenhardt et al. 2006; Therond et al. 2011). However, these regional data mainly focus on
yield and phenology, while local experts with a good knowledge of regional crops, soils and
farms can provide not only the detailed input (climate, management) and output data but also
the intermediate variables’ data of shoot and root growth, and water and nitrogen balances.
In some recent studies, author used the expert knowledge data for different purposes.
For example Clavel et al. (2011) used the expert knowledge to get the data of cropping
systems distribution in a given region. Qualitative information’s are usually required to
establish the relationship between cropping systems and their location factors. Use of farm
typology is an important way to determine relationship between crops and farm types and area
of different farm types. However, use of farm typology lacking the data, regarding different
combinations of crops, farms, soils and irrigation practices. Clavel et al. (2011) showed that,
these data can be obtained from experts. In another study, Lamanda et al. (2012) used the
expert knowledge to develop the conceptual modelling of an agro-ecosystem (CAM) protocol
in order to define the crop management practices for improving the production of coffee in
Guinea (West Africa), determining the factors responsible for increasing yield variability of
cotton crop in Mali (West Africa) and factors explaining the decline of individual Syrah
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
grapevines in southern France. Similarly, Adam et al. (2010) developed a procedure for
combining different modelling approaches based on expert knowledge data. They used the
expert knowledge data of crop phenology, integrated with software technology, for
developing a method to improve the flexibility in crop models. In later both studies qualitative
expert knowledge data was used to built the models. However, in this study, we used the
expert’s knowledge to get the detailed input and output data for evaluation of the model. The
initial soil water and nitrogen status was an exception but this type of data is rarely available
in regional databases (Leenhardt et al. 2006; Therond et al. 2011). Moreover it was easier and
less time consuming to get a complete data set for contrasted soil, crop, management and
climatic conditions during a one day meeting with local experts than from experimental
network not aimed for model evaluation.
On the other hand, it is very important to identify the most widely-acknowledged,
experienced and skilled experts from the study area. The interpretation and transformation of
qualitative information into quantitative, called 'defuzzification problem' (Alcamo 2008) is an
important challenge when using experts’ knowledge data. For example, the meaning of wet
and dry year was found to be different for different experts. The experts considered a year wet
and dry, when the average rainfall ranged respectively from 700 to 920 mm/year and 530 to
600 mm/year. It is well known that such assumptions on rainfall variation can affect crop
growth and development differently. The key aspect of the approach is the assumption that the
data produced with expert knowledge can be considered as a reference for the model
assessment. To ensure quality of expert data the protocol of elicitation of expert knowledge in
a compatible form with the model input and output, included the following elements:
•the experts need to be as much as possible confident with the situations for which they
provide the dynamic and cumulative variables. This is why, during the workshop, we
left experts select the crop, the climatic year, the soil and the crop management of the
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
first situation they described. In a second phase we asked them to complete the
datasets with contrasted situation representative of the region. The experts were asked
to provide, on a 1 to 5 scale, the confidence level for each variable in each situation.
This information was difficult to use as most of the answers were in the 3-4 range and
no curve or table value was provided by the experts when they were not confident
enough with the variable, e.g. as in case of pea crop for dynamic curves.
•in order to ensure independence of the evaluation dataset, compared to the model and
also possibility of imposing the opinions and decisions of one expert on another
(Koehler & Koontz (2008), the experts were not allowed to use the model or to see
simulation results with the model prior to the workshop when they provided the
variables for the various situations.
•before simulation of the situations described by the experts, we verified in the
literature, when data on similar situations were available, that the values given by the
experts were similar. As shown in the results we identified only two cases (N uptake
and cumulated evapotranspiration in pea crop) where the expert provided the wrong
data.
Conclusion
This methodology is valuable for global and dynamic evaluation of cropping system models
in case of unavailability of experimental data for typical crop-soil-management-climate
conditions of a region. Often several regional experts are able to provide detailed knowledge
on the crop growing conditions, their behaviour and performances during the crop cycle. This
expert knowledge, which is generally used for recommendations and extension services, can
be elicited into model compatible format and can be used as input-output variables of a crop
model. Local experts with a good knowledge of regional crops, soils and farms can not only
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
provide the data for key cumulative variables but also the important intermediate variables
(green LAI, actual evapotranspiration, and rooting depth) rarely measured in experiments.
The results of the study show that, if experts are properly chosen, their local knowledge can
be used as a reference for evaluation of the crop models. It is commonly accepted that more
experts (depending upon the number of crops and growing conditions) are better for collective
representation of the systems under studies (Mermet 1991; Becu 2006; Le Bars & Le Grusse
2008). But, it was challenging to identify the reliable, skilled and experienced experts having
detailed knowledge of growing conditions for all crops cultivated in the region. Therefore, we
could get only four experts for providing the input and output data for four main crops (durum
wheat, sunflower, maize and peas). Although, it was important to consider all the crops and
their growing situation in the region, here, our objective was not to build the consensual
representation of all the crops and their biophysical conditions in the region, rather to
demonstrate the possibility of integrating the expert knowledge data for global and dynamic
evaluation of crop models in case of non-availability of detailed input and output
experimental data for large heterogeneous area of a region. In order to cover all the crops and
to get good representativeness of all the cropping system in a region, the most widely-
acknowledged, experienced and skilled experts can be identified from the study area.
Acknowledgement
The authors are grateful to Dr. Philippe Debaeke (INRA Toulouse) and Mr. Bernard Lacroix
for providing the expert knowledge data. We are also grateful to Dr. Myriam Adam (PPS
Wageningen, Netherlands) who provided consistent support for parameterization of pea crop
and Dr. Martin K. Van Ittersum, Kamel Louhichi and Marie-Hélène Jeuffroy for their
valuable comments during different developmental stages of this publication.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
References:
Acutis M, Donatelli M. 2003. Soil PAR 2.00: Software to estimate soil hydrological
parameters and functions. Eur J Agron. 18 (34):373-377.
Adam M, Belhouchette H, Corbeels M, Ewert F, Perrin A, Casellas E, Celette F, Wery J.
2012. Protocol to support model selection and evaluation in a modular crop modelling
framework: An application for simulating crop response to nitrogen supply. Comput
Electron Agric. 86:43-54.
Adam M, Ewert F, Leffelaar PA, Corbeels M, Van Keulen H, Wery J. 2010. CROSPAL,
software that uses agronomic expert knowledge to assist modules selection for crop
growth simulation. Environ Model Softw. 25 (8):946-955
Adam M, Wery J, Leffelaar PA, Ewert F, Corbeels M, Van Keulen H. 2013. A systematic
approach for re-assembly of crop models: An example to simulate pea growth from
wheat growth. Ecol Mod. 250:258-268
Adam M, Belhouchette H, Casellas E, Therond O, Wery J. 2009. APES an agricultural
production and externalities simulator evaluated for two main crops in Midi-Pyrenee.
Paper presented at: AgSAP Conference; Egmond aan Zee, Netherlands.
Alcamo J. 2008. Environmental Futures: The Practice of Environmental Scenario Analysis.
Amsterdam: Elsevier.
Becu N. 2006. Identification et modélisation des représentations des acteurs locaux pour la
gestion des bassins versants [Identification and modeling representations of local
stakeholders for watershed management] [dissertation]. Montpellier (France):
University of Montpellier II.
Belhouchette H, Braudeau E, Hachicha M, Donatelli M, Mohtar R, Wery J. 2008. Integrating
spatial soil organization data with a regional agricultural management simulation
model: a case study in northern Tunisia. Am Soci Agric Biol Eng. 51(3):1099-1109
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Belhouchette H, Louhichi K, Thérond O, Mouratiadou I, Wery J, Van Ittersum MK, Flichman
G. 2011. Assessing the impact of the Nitrate Directive on farming systems using a bio-
economic modelling chain. Agric Syst. 104 (2):135–45.
Biarnès A, Rio P, Hocheux A. 2004. Analyzing the determinants of spatial distribution of
weed control practices in a Languedoc vineyard catchment. Agron. 24:187–196.
Bouman BAM, Van Keulen H, Van Laar HH, Rabbinge R. 1996. The ‘School of de Wit’ crop
growth simulation models: a pedigree and historical overview. Agric Syst. 52:171–
198.
Claude MJ, Ayorinde A, Olufayo b, Ruelle P. 1997. Sorghum and sunflower
evapotranspiration and yield from simulated leaf area index. Agric Water Manage.
35:167-182
Clavel L, Soudais J, Baudet D, Leenhardt D. 2011. Integrating expert knowledge and
quantitative information for mapping cropping systems. Land Use Policy. 28 (1):57-
65.
Corre-Hellou G, Faure M, Launay M, Brisson N, Crozat Y. 2009. Adaptation of the STICS
intercrop model to simulate crop growth and N accumulation in pea–barley intercrops.
Field Crops Res. 113(1):72–81.
Donatelli M. 2002. Simulations with CropSyst of cropping systems in Southern France,
common agricultural policy strategy for Regions. Montpellier (France): Mediterranean
Agronomic Institute of Montpellier.
Donatelli M, Russell G, Rizzoli AE, Acutis M, Adam M, Athanasiadis IN, Balderacchi M,
Luca Bechini L, Belhouchette H, Bellocchi G, et al. 2010. Environmental and
agricultural modelling: Integrated approaches for policy impact assessment. Springer:
Dordrecht. Chapter 4, A Component-Based Framework for Simulating Agricultural
Production and Externalities; p. 63-108.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
El Hajj M, Bégué A, Guillaume S, Martiné JF. 2009. Integrating SPOT-5 time series, crop
growth modeling and expert knowledge for monitoring agricultural practices-The case
of sugarcane harvest on Reunion Island. Remote Sens Environ. 113(10):2052-2061.
Faivre R, Leenhardt D, Voltz M, Benoît M, Papy F, Dedieu G, Wallach D. 2004. Spatialising
crop models. Agron. 24:205–217.
Jagtap SS, Jones JW. 2002. Adaptation and evaluation of the CROPGRO-soybean model to
predict regional yield and production. Agric Ecosyst Environ. 93:73–85.
Jones JW, Hoogenboom G, Porter CH, Boote KJ, Batchelor WD, Hunt LA, Wilkens PW,
Singh U, Gijsman AJ, Ritchie JT. 2003. The DSSAT cropping system model. Eur J
Agron. 18(3-4):235-265.
Jorgensen SE. 1986. Fundamentals of Ecological Modelling. Elsevier, Amsterdam, 389 pp.
Koehler B, Koontz TM. 2008. Citizen participation in collaborative watershed partnerships.
Environ Manage. 41:143–154.
Lamanda N, Roux S, Delmotte S, Merot A, Rapidel B, Adam M, Wery J. 2012. A protocol for
the conceptualisation of an agro-ecosystem to guide data acquisition and analysis and
expert knowledge integration. Eur J Agron. 38:104-116.
Le Bars M, Le Grusse P. 2008. Use of a decision support system and a simulation game to
help collective decision-making in water management. Comput Electron Agric.
62:182–189.
Leenhardt D, Wallach D, Le Moigne P, Guérif M, Bruand A, Casterad MA. 2006. Working
with Dynamic Crop Models: Evaluation, Analysis, Parameterization, and
Applications. Amsterdam: Elsevier. Chapter 7, Using crop models for multiple fields;
p. 209-248
Launay M, Guerif M. 2005. Assimilating remote sensing data into a crop model to improve
predictive performance for spatial applications. Agric Ecosyst Environ. 111:321−339.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Loague K, Green RE. 1991. Statistical and graphical methods for evaluating solute transport
models: overview and application. J Contam Hydrol. 7:51-73.
Mermet L. 1991. Participation, strategies and ethics: roles of people in wetland management.
Landsc Urban Plan. 20:231–237.
McCown RL, Hammer GL, Hargreaves JNG, Holzworth DP, Freebairn DF. 1996. APSIM: a
novel software system for model development, model testing and simulation in
agricultural systems research. Agric Syst. 50:255–271.
Middelkoop H, Janssen LLF. 1991. Implementation of temporal relationships in knowledge
based classification of satellite images. Photogramm Eng Remote Sens. 57:937−945.
Nolot JML, Debaeke P. 2003. Principes et outils de conception, conduite et évaluation de
systèmes de culture [Principles and tools for design, conduct and evaluation of
cropping systems]. Cahiers Agric. 12:1-14. French.
Odum HT. 1983. Systems Ecology: An Introduction. John Wiley & Sons: NY.
Oreskes N, Shrader-Frechette K, Belitz K. 1994. Verifcation, validation and confirmation of
numerical models in the earth science. Science. 263:641-646.
Power M. 1993. The predictive validation of ecological and environmental models. Ecol
Model. 68:33-50.
Ruget F, Brisson N, Delécoller R, Faivrer O. 2002. Sensitivity analysis of a crop simulation
model, STICS, in order to choose the main parameters to be estimated. Agron.
22:133–158
Shugart HH. 1984. A Theory of Forest Dynamics. Springer- Verlag, New York, NY.
Stöckle CO, Donatelli M, Nelson R. 2003. CropSyst, a cropping systems simulation model.
Eur J Agron. 18:289-307.
Therond O, Hengsdijk H, Casellas E, Wallach D, Adam M, Belhouchette H, Oomen R,
Russell G, Ewert F, Bergez JE, et al. 2011. Using a cropping system model at regional
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
scale: Low-data approaches for crop management information and model calibration.
Agric Ecosyst Environ. 142 (1-2):85-94.
Utset A, Farré I, Martınez A, Cavero J. 2004. Comparing Penman–Monteith and Priestley–
Taylor approaches as reference-evapotranspiration inputs for modeling maize water-
use under Mediterranean conditions. Agric Water Manage. 66:205–219.
Van Ittersum MK, Donatelli M. 2003. Modelling cropping systems—highlights of the
symposium and preface to the special issues. Eur J Agron. 18:187–394.
Van Ittersum MK, Ewert F, Heckelei T, Wery J, Alkan Olsson J, Andersen E, Bezlepkina I,
Brouwer F, Donatelli M, Flichman G, et al. 2008. Integrated assessment of agricultural
systems – A component-based framework for the European Union (SEAMLESS)
Agric Syst. 96:150-165.
Wallach D, Goffinet B, Tremblay M. 2002. Parameter estimation in crop models: exploring
the possibility of estimating linear combinations of parameters. Agron. 22:171–178.
Wery J. 1996. Production de graines et fixation d’azote par des cultures de légumineuses sous
contrainte hydrique. Formation doctoral sur les bases de la production végétale [Seed
production and nitrogen fixation by legumes under water stress conditions. Doctoral
training on the bases of crop production]. Montpellier 2 University of Sciences and
Literature, Languedoc-Roussillon, France. 65 pp. French.
Williams JR, Jones CA, Kiniry JR, Spanel DA. 1989. The EPIC Crop Growth Model. Trans
Asae. 32:497-511.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Table 1. Experimental (INRA Toulouse) soil data used for calibration of the model on 1
meter soil depth.
Soil Characteristics Soil types (average value for all layers)
Clay loam Loam
Sand (%) 26 35
Silt (%) 40 37
Clay (%) 34 28
Bulk density (g cm-3) 1.3 1.3
Permanent wilting point (m3 m-3) 0.19 0.16
Field capacity (m3 m-3) 0.34 0.30
Organic matter (%) 1.34 1.41
Maxi. Depth (m) 0.95 0.94
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Table 2. Experimental crop data (INRA Toulouse) used for calibration of the APES model
Note: N*: Total number of selected plots for each crop
Crops N*
Soil types
Management data Output data
Sowing date
Fertilization
(kg N ha-1)
Irrigation
(mm)
Above ground
biomass (t ha-1)
Grain yield
(t ha-1)
Above ground N
uptake (kg ha-1)
Average σ Average σ Average σ Average σ Average σ
Plots with
clay loam soil
Plots with
loam soil
Durum
wheat 7 4 3
4-21
November 119 56 Rainfed - 13.70 3.4 6 1.5 167 40
Maize 7 3 4 8-22 April 196 18 214 74 22.4 3 11 1 216 20
Sunflower 6 3 3 7-10 April 63 3 Rainfed - 8.7 1.5 3.4 0.8 101 35
Peas 7 4 3
24 November
and
1-10
December
0 - 50 3 7.5 0.7 3.4 0.6 196 14
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Table 3. Typical agricultural activities selected by experts in the arable zone of the Midi-Pyrénées region.
Selected activities Crop input data Crop output data
Expert
Crops Soil
types
Climate
condition
Real year
specified
by
experts
Activities
Initial soil
conditions Management Cumulative variables
Water
content
(m3 m-3)
Nitrogen
content
(kg N
ha-1)
Average
sowing
date
Fertilization
(kg N ha-1)
Irrigation
(mm)
Above
ground
biomass
(t ha-1)
Grain
yield
(t ha-1)
Above
ground N
uptake
(kg ha-1)
Actual
accumulated
evapotranspiration
(mm)
Durum
wheat
Clay
loam
Wet year 1996-1997 1 0.34 22 Ist Nov 200 Rainfed 15 7 220 400 1
Dry year 2004-2005 2 0.34 22
Ist Dec 150 Rainfed 10 4.5 160 300 1
Loam Wet year 1997-1998 3 0.34 22
11-Nov 220 Rainfed 14 5.5 200 450 2
Dry year 2002-2003 4 0.34 22
11-Nov 150 Rainfed 12 6 200 450 2
Peas Loam
Wet year 2001-2002 5 0.34 18
15-Dec 0 Rainfed 10 4.5 250 250 2
Dry year 2006-2007 6 0.34 18
15-Dec 0 90 10 5 250 245 2
Sunflower
Loam Wet year 2002 7 0.38 9
20-Apr 50 Rainfed 9 3 150 450 2
Dry year 2003 8 0.38 9
20-Apr 0 Rainfed 5.5 2.5 130 350 2
Clay
loam
Wet year 2008 9 0.38 9
15-Apr 60 Rainfed 8 3 125 450 3
Dry year 2009 10 0.38 9
15-Apr 60 Rainfed 7 2.5 115 400 3
Maize Loam
Unlimited
water
conditions
2009 11 0.23 15
20-Apr 200 240 22 11 260 580 4
Water
limited
conditions
2009 12 0.23 15
20-Apr 200 170 18 9 240 500 4
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Table 4. Statistical analysis of the calibration results for above ground biomass, grain yield
and above ground N uptake
Notes: N: Number of plots, Ô: average observed value Ŝ: average simulated value, RRMSE:
Relative root mean square error.
Crops N Variable
Ô
Ŝ RRMSE
(%)
Durum wheat 7
Above ground biomass (t ha-1) 13.70 15.33 15
Grain yield (t ha-1) 5.98 6.44 14
Above ground N uptake (kg ha-1) 167 184 14
Maize 7
Above ground biomass (t ha-1) 22.4 22.2 8
Grain yield (t ha-1) 11.08 10.64 8
Above ground N uptake (kg ha-1) 216 218 6
Sunflower 6
Above ground biomass (t ha-1) 8.7 8.2 13
Grain yield (t ha-1) 3.4 3.1 14
Above ground N uptake (kg ha-1) 101 108 28
Pea 7
Above ground biomass (t ha-1) 7.5 8.2 15
Grain yield (t ha-1) 3.4 3.5 13
Above ground N uptake (kg ha-1) 196 143 28
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 1. Methodology for calibration and validation of the model; using experimental and
expert knowledge data. First the model was calibrated classically by using experimental data.
Then the model was evaluated, which is divided into 3 steps, the global and dynamic
evaluation of model was achieved by using expert knowledge data.
Experts
(4)
APES
Evaluation
Step 3
Global
evaluation
Dynamic
evaluation
APES calibration (Crop
phenology, Biomass and N
parameters)
Crop management data
Climate data
Soil data
Experimental data
Crops (Durum wheat, Maize,
Sunflower and Peas)
Selection of
representative
activities
Step 1 (Input data)
Mangement,
initial soil
conditions, soil
and climate
Step 2 (Output data)
Model
generated
output data
Running of
calibrated
APES model
Experts
given output
data
Phase 1: Model calibration
Phase 2: Model evaluation
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 2. An example of expert drawn dynamic curve for activity 1 (durum wheat cultivated on clay
loam soil under wet climatic conditions)
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 3(a), (b) (c) (d). Correlation of simulated and expert given cumulative values of
above ground biomass, grain yield, above ground N uptake and actual cumulated
evapotranspiration (ETC) respectively for all activities. Activities 1, 2, 3 and 4 indicate the
durum wheat, 5 and 6 peas, 7, 8, 9 and 10 sunflower and 11 and 12 maize crops.
1
2
7
4
9
6
3
810
5
11
12
0
100
200
300
400
500
600
0 100 200 300 400 500 600
Simulated ETC (mm)
Expert given ETC (mm)
d
RRMSE = 31 %CRM = 15 %
1
2
3
4
6
89
10
11
12
0
5
10
15
20
25
0 5 10 15 20 25
Simulated AGB (t/ha)
Expert given AGB (t/ha)
aRRMSE = 20 %
7
5
CRM = 8 %
1
2
45
7
10
8
12 11
6
0
50
100
150
200
250
300
0 50 100 150 200 250 300
Simulated N uptake (kg/ha)
Expert given N uptake (kg/ha)
c
3
RRMSE = 27 %CRM = 16 %
9
1
23
4
6
7
8
10
11
12
0
2
4
6
8
10
12
024681012
Simulated grain yield (t/ha)
Expert given grain yield (t/ha)
bRRMSE = 17 %
5
CRM = 1 %
9
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 4. Correlation of simulated and expert given values for the dynamic of AGB across different phenological stages. Activities 1, 2, 3 and 4
indicate the durum wheat 7, 8, 9 and 10 sunflower and 11 and 12 maize crops.
1
2
3
4
7
8
9
10 11
12
0
5
10
15
20
25
0 5 10 15 20 25
Simulated AGB (t/ha)
Expert given AGB (t/ha)
Flowering
RRMSE = 32 %
1
2
3
4
7
8
9
10
11
12
0
5
10
15
20
25
0 5 10 15 20 25
Simulated AGB (t/ha)
Expert given AGB (t/ha)
Grain f illing RRMSE = 19 %
1
2
3
4
8
8
9
10
0
5
10
15
20
25
0 5 10 15 20 25
Simulated AGB (t/ha)
Exp ert giv en AGB (t /ha)
Physiological Maturity
RRMSE = 18 %
7
12
11
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
1
2
3, 4
7
8
9
10 11
12
0
25
50
75
100
125
150
175
200
0 25 50 75 100 125 150 175 200
Simulated N uptake (kg/ha)
Expert given N up take (kg /ha)
Flowering
RRMSE = 37 %
2
13,4
7
8
9
10
11
12
0
50
100
150
200
250
300
0 50 100 150 200 250 300
Simulated N uptake (kg/ha)
Expert giv en N uptake (kg/ha)
Grain filling
RRM SE = 22 %
1
2
4
8
7
10
12
11
0
50
100
150
200
250
300
0 50 100 150 200 250 300
Simulated N uptake (kg/ha)
Expert given N uptake (kg/ha)
Physiological Maturity
3
RRMSE = 27 %
9
Figure 5. Correlation of simulated and expert given values for the dynamic of N uptake across different phenological stages. Activities 1, 2, 3 and 4
indicate the durum wheat 7, 8, 9 and 10 sunflower and 11 and 12 maize crops
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
1
2
7
11
9
10
8, 12
0
50
100
150
200
250
300
350
400
0 50 100 150 200 250 300 350 400
Simulated ETC (mm)
Expert given ETC (mm)
Flowering
RRM SE = 36 %
1
2
7
8, 10 9
11
12
0
100
200
300
400
500
600
0 100 200 300 400 500 600
Simulated ETC (mm)
Expert given ETC (mm)
Grain filling
RRMSE = 41 %
1
2
7
8
9
11
12
0
100
200
300
400
500
600
0 100 200 300 400 500 600
Simulated ETC (mm)
Expert given ETC (mm)
Physiological Maturity
RRMSE = 28 %
10
Figure 6. Correlation of simulated and expert given values for the dynamic of ETC across different phenological stages. Activities 1 and 2 indicate the
durum wheat 7, 8, 9 and 10 sunflower and 11 and 12 maize crops.
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 7. Correlation of simulated and expert given values for the dynamic of LAI across different phenological stages. Activities 1 and 2
indicate the durum wheat 7, 8, 9 and 10 sunflower and 11 and 12 maize crops
1
2
7
8
9
10 11
12
0
1
2
3
4
5
6
0123456
Simulated LAI (m
2
/m
2
)
Expert given LAI (m
2
/m
2
)
Flowering
RRMSE = 17 %
1
2
7
8
9
10
11
12
0
1
2
3
4
5
6
0123456
Simulated LAI (m
2
/m
2
)
Expert given LAI (m
2
/m
2
)
Grain filling
RRMSE = 20 %
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
Accepted Manuscript
Figure 6. Correlation of simulated and experts given values for the dynamic of rooting depth across different phenological stages. Activities 1
and 2 indicate the durum wheat 7, 8, 9 and 10 sunflower and 11 and 12 maize crops.
1, 2
7
8
9
10
11, 12
0
0.3
0.6
0.9
1.2
1.5
0 0.3 0 .6 0.9 1.2 1 .5
Simulated rooting depth (m)
Expert given root depth (m)
Flowering
RRMSE = 23 %
1, 2
7, 8
9, 10
11, 12
0
0.3
0.6
0.9
1.2
1.5
0 0 .3 0.6 0 .9 1.2 1.5
Simulated rooting depth (m)
Expert given rooting depth (m)
Grain filling
RRMSE = 16 %
1, 2
7, 8
9, 10
11, 12
0
0.3
0.6
0.9
1.2
1.5
0 0.3 0 .6 0.9 1 .2 1.5
Simulated rooting depth (m)
Expert given rooting depth (m)
Physiological maturity
RRMSE = 15 %
Downloaded by [Tanvir Shahzad] at 05:46 21 April 2015
