Field Crops Research 130 (2012) 145–154

Contents lists available at SciVerse ScienceDirect

Field Crops Research

journal homepage: www.elsevier.com/locate/fcr

Monitoring regional wheat yield in Southern Spain using the GRAMI model and
satellite imagery
F.L.M. Padilla a,∗ , S.J. Maas b , M.P. González-Dugo a , F. Mansilla a , N. Rajan c , P. Gavilán a , J. Domínguez a
a
IFAPA, Centro Alameda del Obispo, Córdoba, Spain
b
Texas Tech University, Lubbock, TX, USA
c
Texas Agrilife Research and Extension Center, Vernon, TX, USA

a r t i c l e i n f o a b s t r a c t

Article history: The worldwide increase in the demand for food and the limited available resources to produce it make
Received 7 February 2011 it necessary to develop tools which allow estimation of crop production, thereby helping to manage the
Received in revised form 8 February 2012 way food is produced, stored and distributed. GRAMI, a model developed to simulate the growth and yield
Accepted 20 February 2012
of grain crops and capable of using remotely sensed information, was applied to the semiarid region of
Southern Spain. The aim of this study was to demonstrate a methodology for using the GRAMI model,
Keywords:
along with satellite remote sensing data, to estimate regional crop yields, and to assess the accuracy
Yield
of the resulting yield estimates. Crop-specific model parameters (light-use efficiency, crop phenologi-
Wheat
GRAMI
cal stage and yield partitioning factor) were evaluated using information collected from 29 durum and
Landsat bread wheat experimental plots in order to verify the performance of the model in this region spectral
radiometry measurements were taken for each plot throughout the growing season to obtain experimen-
tal relationships between the normalized difference vegetation index (NDVI) and leaf area index (LAI).
This relationship was used to estimate crop LAI for the within-season calibration of GRAMI from satellite
remote sensing data. Forty-nine commercial wheat fields were chosen in 2008 and 2009 to validate the
model. Information from meteorological stations was used in running the model, and satellite image
data were used along with the LAI–NDVI relationship to provide estimates of crop LAI for within-season
calibration of the model. Yield data for comparison with model estimates were obtained for each field
from the farmers. For the validation study, the average amount that the yield of an individual field was
over- or under-estimated was 884 and 852 kg ha−1 for the 2008 and 2009 seasons, respectively. The abso-
lute errors between the average estimated and average observed yield values were 5.44% and 6.86% for
the 2008 and 2009 seasons, respectively. For each of the 2 years, the average estimated yield was not
significantly different from its corresponding average observed yield. Based primarily on its ability to
accurately estimate the average yield for a set of fields and its reliance on readily available weather and
remote sensing data, the GRAMI model, verified for a region and calibrated using satellite image data,
appears to be a practical and appropriate option for operationally monitoring regional crop yields with a
reasonable degree of accuracy.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction
Recent studies show that the global demand for food will
increase for at least another 40 years. It is estimated that the global
population will reach 9 billion people by the middle of this cen-
tury (Charles et al., 2010). To determine the extent to which the
demand for food can be met, it is necessary to monitor crop yields
in the agricultural production regions around the world. Crop yield
estimates can indicate the degree to which regional production can
∗ Corresponding author. Tel.: +34 957016173; fax: +34 957016043.
E-mail addresses: franciscol.munoz@juntadeandalucia.es, currolu@hotmail.com
(F.L.M. Padilla).
meet or exceed regional demand, which is the motivation for com-
modity trade that fuels a large portion of international commerce.
They can also indicate situations of regional crop failure that might
necessitate the mustering of humanitarian aid to avert famine con-
ditions for populations that cannot afford to import additional food
supplies. For these reasons, the development of methods for mon-
itoring regional crop yields has been a priority among researchers
for a number of decades.
Early programs such as LACIE (Large Area Crop Inventory Exper-
iment) and AgRISTARS, initiated by the U.S. government, sought to
address the challenge of crop yield monitoring through the use of
mathematical crop growth models and satellite remote sensing.
These efforts demonstrated the potential of these approaches and
established the foundations of the physical relationship between

0378-4290/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.fcr.2012.02.025
146 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154
spectral measurements and the biophysical properties of soil and
vegetation. Since then, the number of sensing systems has greatly
increased, improving the quantity and quality of available remote
sensing data.
The use of vegetation indices (VI) obtained from remote sens-
ing measurements has been successfully used to estimate crop
yield at the plot scale through statistical relationships between
VI and yield. However, healthy development of the crop (without
biotic stresses) was required to obtain satisfactory results (Ma et al.,
2001). Numerous research results involving yield estimation based
on empirical relationships between VI obtained from radiometer
measurements and observed crop yield have been reported for
various crops, including cotton (Zhao et al., 2007), rainfed and irri-
gated wheat (Aparicio et al., 2000), and soybeans (Ma et al., 2001).
Regional-scale VI can be obtained from sensors on board satellites.
For example, the average regional yields of maize (Mkhabela and
Mashinini, 2005) and wheat (Labus et al., 2002) have been suc-
cessfully estimated using satellite image data. However, poorer
results have been obtained for cotton (Dalezios et al., 2001). To
improve their model’s estimation ability, various researchers have
included additional information besides VI, such as rainfall, soil
temperature and soil moisture, achieving improved predictions for
wheat (Balaghi et al., 2008; Wall et al., 2007) and corn and soy-
beans (Prasad et al., 2006). A primary drawback to using empirical
yield–VI relationships is that they are invariably site-specific and
cannot be directly applied to situations other than those for which
they were developed (i.e., other crops and regions). Another prob-
lem with empirical relationships is that their accuracy tends to be
a function of the size of the data set used in their development. To
build an accurate empirical model, a set of yield and VI observa-
tions must be obtained that includes the entire range of growing
conditions that can be experienced by the crop in that region. Oth-
erwise, the accuracy of the model may be adequate under “average”
growing conditions but poor for above- or below-average growing
conditions. Building an accurate empirical model may require data
from numerous years. Such extensive data sets may not exist for
some crops and regions.
A potential solution that has been suggested for this problem
is the use of crop growth simulation models. These models have a
more physical basis than strictly empirical relationships and, once
developed for a crop, should be more universally applicable. The
biomass model of Monteith (1972) allows the estimation of the
increase in crop biomass over time by accumulating the product of
APAR (absorbed photosynthetically active radiation) and a param-
eter representing light-use efficiency (i.e., the amount of biomass
produced per unit of APAR). It can include a harvest index that can
be used for estimating crop yield (Lobell et al., 2003; Zwart and
Bastiaanssen, 2007; Padilla et al., 2010). Asrar et al. (1989) modified
this model by introducing a stress index, as the original model did
not consider the possibility of water deficiency during the growing
season. The parameter fPAR (the fraction of PAR absorbed by the
plant canopy) is needed for computing APAR. The ability to esti-
mate fPAR from remotely sensed spectral measurements has been
demonstrated (Asrar et al., 1984). The Monteith biomass estima-
tion model has been applied over large areas (Bastiaanssen and Ali,
2003; Lobell et al., 2003; Zwart and Bastiaanssen, 2007).
Complex crop growth models, such as DSSAT (Jones et al., 1998,
2003), CropSyst (Stöckle et al., 1994, 2003) and STICS (Brisson et al.,
1998, 2003), have been widely used over the last 15 years for many
applications. Many of these efforts have been designed to study
management options at different study sites, including fertilizer,
irrigation, pest management and site-specific farming (Jones et al.,
2003). Some of these models have also been applied at the regional
scale using remote sensing (Weiss et al., 2001). Their ability to
estimate crop yield under experimental plot conditions has been
good. However, these models require a large number of inputs and
parameters that make their use difficult at the field or regional
scales in areas where detailed information is unavailable.
An alternative is to use less-complicated models that require
fewer and less detailed inputs and parameters. An example is
GRAMI, a model originally developed to simulate the growth and
yield of grain crops, including wheat (Maas, 1992, 1993a,b). While
GRAMI uses a less complex approach for modelling crop growth
similar to the Monteith biomass models, it compensates for this lack
of complexity by including the capability to use remotely sensed
information. Previous studies (Maas, 1988a,b) have shown that it
was possible to improve the performance of a crop model by using
information derived from remote sensing. By means of an iter-
ative numerical procedure built into the model, GRAMI can use
infrequent remote sensing observations of LAI or ground cover to
adjust the values of certain model parameters to minimize the error
between remote sensing and model estimates of crop growth. This
procedure is called “within-season calibration”. Studies using field
data (Maas, 1988b, 1993c) have demonstrated that the inclusion
of remotely sensed information can improve model estimates of
crop yield. In terms of practical applicability, a model like GRAMI
has distinct advantages over the other yield estimation techniques
described above. Unlike the empirical models, GRAMI is already
formulated so it does not need to be “developed” for the region to
which it is to be applied. Therefore, extensive historic data sets from
the region are not needed. Unlike the more detailed crop growth
simulation models, GRAMI uses inputs (weather observations and
operational satellite imagery) that are readily available from most
agricultural regions. The combination of a relatively simple model
and remote sensing data provides a tool of potential use in regional
crop monitoring programs.
The overall objectives of this study were to demonstrate a
methodology for using the GRAMI model to provide estimates of
regional crop yield and to evaluate the level of accuracy that could
be expected from this operation. There were three specific objec-
tives of this study. The first was to establish a relationship between
LAI and ground-derived NDVI and perform a comparison between
satellite-derived and ground-derived NDVI from wheat fields in
Spain. The second was to evaluate growth- and phenology-related
parameters in the GRAMI model using agronomic data from experi-
mental field plot studies in the region, and to verify the functioning
of the model by comparing simulated and observed wheat yields for
these studies. The third specific objective was to apply the verified
model to regional yield assessment in Spain using satellite remote
sensing data. Based on the results of this study, the suitability of
using GRAMI for operationally monitoring regional crop yields is
assessed.
2. Materials and methods
2.1. Model description
The GRAMI model (Maas, 1992, 1993a,b) was used in this study
to estimate wheat yield using remotely sensed data at the field
scale. In the model, the four processes involved in simulating daily
wheat crop growth are: (1) calculation of growing degree-days
(GDD), (2) absorption of incident PAR by leaves, (3) production of
above-ground new dry mass by the leaf canopy, and (4) determina-
tion of LAI partitioning of new dry mass to leaves, stems, and grain.
The daily increase in accumulated growing degree-days (D) is
calculated as follows
D = Max[T − Tb , 0] (1)
where T is the average daily air temperature (◦ C) and Tb is a base
temperature (◦ C) specific to a crop species. For wheat, Tb = 0 ◦ C. The
value of D is zero when T is less than Tb .
 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154
The daily increase in above-ground dry mass (M, g) is calcu-
lated as follows
M = εQ
where ε is the light use efficiency (g MJ−1
PAR) and Q is the daily
total absorbed PAR in MJ m−2 absorbed by the crop canopy. Q is
evaluated as follows
Q = ˇR(1 − e−k LAI )
where R is the incident daily total solar irradiance in MJ m−2 ,
ˇ is
the fraction of total solar irradiance that is PAR, and k is a light
extinction coefficient specific for a given crop (Charles-Edwards
et al., 1986). The value of ˇ is approximately 0.45 (Monteith and
Unsworth, 1990).
The daily increase in LAI due to new growth (L) is calculated
as
L = M P1 S
where M is the daily increase in aboveground dry biomass from
Eq. (2), S is the specific leaf area (SLA, m2 g−1 ) of the leaf tissues,
and P1 is the fraction of M partitioned to new leaves. The value of
P1 is calculated using the following equation
P1 = Max[1 − aebD , 0]
where a and b are parameters that control the magnitude and shape
of the function and D is the accumulated growing degree-days.
Leaf senescence is modelled by assuming that leaves appearing
on a given day have a specific lifespan in degree-days. The lifespan
associated with L for a given day is calculated using the empirical
function
J = c + dD
where c and d are parameters that control the magnitude and slope
of the function.
The number of growing degree-days between the various phe-
nological stages in the development of the wheat plants is an input
and allows estimating the dates on which emergence, flowering
and maturity occur. A yield partitioning factor is used between
flowering and maturity to estimate yield as a percentage of total
dry mass production.
As described previously, the model can use within-season cali-
bration based on observations of crop growth (LAI or ground cover)
obtained during the growing season. In this study, within-season
calibration (Maas, 1992, 1993b) was used to systematically manip-
ulate the values of selected model parameters (a, b, c, d) and the
initial value of LAI at crop emergence (L0 ) until values of crop
LAI simulated by the model matched corresponding values of LAI
obtained from remote sensing observations of the crop. Observed
crop growth used in the within-season calibration was estimated
from ground-based spectroradiometer measurements or satellite
imagery.
For inputs, the GRAMI model requires daily observations of aver-
age air temperature and total solar irradiance. It does not need
information on soil physical characteristics, soil water content, or
soil fertility. The planting date must be supplied, but information
on seeding rate or plant population density is not required. GRAMI
does not even need observations of precipitation or irrigation. One
might question how such a model can make reasonably accurate
simulations of crop growth and yield. In crop models that simu-
late biomass growth from absorbed PAR (Monteith-type models
and complex models like DSSAT), the most critical computational
activity is the simulation of leaf canopy growth. If the leaf canopy
can be accurately simulated, then the biomass growth probably
will also be simulated accurately. Since leaf canopy growth is very
sensitive to water stress and soil fertility, most models need to
147
explicitly incorporate soil characteristics, precipitation, irrigation,
and fertilization in order to accurately simulate crop growth and
yield. However, a different simulation strategy is used in GRAMI.
(2)
Through the use of within-season calibration based on observed
values of crop LAI, the effects of factors that affect canopy growth
(like water stress, nitrogen nutrition, or plant population density)
are “absorbed” into the simulation when the LAI simulation is “fit”
to the observed LAI data (i.e., when model parameter values are
(3) manipulated so that the resulting LAI simulation minimizes the
error between the simulated and observed LAI values) and implic-
itly affect the simulated crop growth and yield without the need for
them to be explicitly modelled. This is the strength of the GRAMI
model – the use of “extra” information (observed LAI) allows a sim-
ple crop growth model to be more accurate than it normally would
be.
2.2. Experimental plot and field data
(4)
2.2.1. Experimental plot data
Data for verifying the GRAMI model were obtained from two
existing experimental sites planted to wheat during the 2009
growing season (Fig. 1). The first site was located near Cañete
de las Torres (Córdoba). Fourteen durum (Triticum durum) and 14
(5) bread wheat cultivars (Triticum aestivum) (Table 1) were sown on
December 23, 2008, in 12-m2 experimental plots (six rows, 20 cm
apart). Measurements of the following factors were made every 15
days during the growing season: crop height, PAR, LAI and spectral
reflectance. Mean values were obtained by averaging three point
measurements made at each of three different locations in each
plot. In addition, the Zadoks growth stage (Zadoks et al., 1974)
was determined weekly for each cultivar. An AccuPAR Model LP-
80 (Decagon Devices, Inc., Pullman, WA) ceptometer was used to
(6)
estimate PAR and LAI at each site. At the end of the growing season
(July 2, 2009), the plots were manually harvested and grain yield
(kg ha−1 ) was determined for each plot.
The second experimental site consisted of a rainfed wheat field
(T. aestivum) with an area of 2 ha located at the IFAPA Alameda
del Obispo Experimental Farm (Córdoba). The field was sown on
December 19, 2008, with the wheat cultivar ‘Lubrican’. Field infor-
mation similar to that described for the first site was collected twice
a month but, in this case, 20 measurements distributed across the
field were made to produce field-representative average values for
each measured quantity. In addition, above-ground biomass was
measured by swathing three representative 0.25 m2 areas within
the field. Dry weights of these samples were obtained by drying
the samples in an oven at 80 ◦ C for 2 days. Finally, the remainder of
each plot was mechanically harvested on July 3, 2009.
Within-season calibration of GRAMI required remotely sensed
estimates of crop LAI. Field measurements of wheat canopy
reflectance were collected from the plots at the experimental site
located near Cañete de las Torres (Córdoba). Canopy reflectance
measurements were made using a hand-held field spectrora-
diometer (ASD FieldSpec, Analytical Spectral Devices, Boulder, CO).
Table 2 shows the days on which measurements were made. The
spectral range of the instrument was between 325 and 1075 nm
with a sampling interval of 1.6 nm, which included the visible and
near-infrared (NIR) portions of the spectrum used in computing
vegetation indices. The measurements were taken at midday under
cloudless conditions. Reflectance was calculated as the ratio of the
reflected and the incident irradiance, where the incident irradi-
ance was periodically measured using a white Spectralon reference
panel (Labsphere, North Sutton, NH). The normalized difference
vegetation index (NDVI) (Rouse et al., 1974), soil adjusted vege-
tation index (SAVI) (Huete, 1988), and the enhanced vegetation
index (EVI) (Huete et al., 2008) were calculated from the reflectance
measurements in the visible and NIR spectral bands. Blue, red and
148 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154

Fig. 1. Geographic locations of the biomass experimental plots (upper-left), 28 cultivar experimental plots (upper-right) and 49 commercial fields monitored during the
growing season 2009 (bottom).

Table 1
Yield of durum and bread wheat cultivars (kg ha−1 ) and associated phenology group. Group A includes earlier-maturing cultivars and Group B includes later-maturing
cultivars.

Durum wheat Bread wheat

Cultivar Grain yield (kg ha−1 ) Phenology group Cultivar Grain yield (kg ha−1 ) Phenology group

Ancalei 3808 A Abderramán 3758 B

Beldur 3717 A Artur Nick 3625 B
Calcas 3917 A Badiel 4317 A
Calero 3883 B Cartaya 4675 A
Core 5050 A Catedral 3508 B
Don Juan 4367 A Gades 3467 A
Euroduro 4342 A Galeón 3450 A
Próspero 4383 A Gazul 4225 B
Ramírez 5667 A Osado 3417 B
Saragolla 3817 A Palesio 4008 B
Sculptur 4617 B Salama 5200 B
Simeto 4617 A Sensas 3975 B
Vitrón 2508 A Tejada 4492 A
Vitrosol 4267 A Victorino 4708 A
Average 4211 4059

Table 2
Observation dates and sources of spectral data for farmers’ fields and the two experimental plots.

Commercial Fields Experimental plots

2008 growing season 2009 growing season Cañete de las Torres (2009) Córdoba (2009)

DAY Sensor DAY Sensor DAY Sensor DAY Sensor

350 ETM+ (L7) 345 TM (L5) 49 ASDa 43 ASDa

65 ETM+ (L7) 11 TM (L5) 72 ASDa 56 ASDa
105 TM (L5) 43 TM (L5) 85 ASDa 71 ASDa
122 Awifs 67 ETM+ (L7) 98 ASDa 78 ASDa
177 ETM+ (L7) 123 TM (L5) 112 ASDa 96 ASDa
147 ETM+ (L7) 124 ASDa 113 ASDa
138 ASDa 125 ASDa
168 ASDa 139 ASDa
175 ASDa
a
Field measurements with a hand-held ASD FieldSpec spectroradiometer.
 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154
NIR reflectance values required for VI calculation were computed
by averaging the reflectance values within the ranges of wave-
length corresponding to Landsat-5 TM Band 1 (blue), Band 3 (red),
and Band 4 (NIR). An average value for each vegetation index was
obtained for each measurement day and cultivar. To evaluate LAI
from remote sensing data, relationships were developed based on
these measurements. The spectral radiometry measurements taken
on each experimental plot over the course of the growing season
were used to calculate values of the NDVI, SAVI, and EVI vegetation
indices. Corresponding values of AccuPAR-measured LAI were then
regressed against these VI values to produce empirical non-linear
relationships between LAI and VI. The ability of these indices to
estimate LAI was evaluated based on the data in this study.
Similar spectroradiometer data were collected from the 2 ha
field located at the IFAPA Alameda del Obispo Experimental Farm
(Córdoba). In this case, the data were used to calculate values of
NDVI for validating corresponding estimates of NDVI derived from
satellite observations of that field (see Section 2.4).
Meteorological data to run the model (average daily air tem-
perature and daily total solar irradiance) were obtained from two
weather stations which are part of a regional weather station net-
work (RIA). The first was near Cañete de las Torres and the second
was located 100 m from the wheat experimental plots located
inside the IFAPA Alameda del Obispo Experimental Farm.
2.2.2. Commercial field data
To validate the GRAMI model, 49 commercial wheat fields
(Fig. 1) large enough to be clearly resolved by the Landsat-5 TM
and Landat-7 ETM+ sensors were selected in 2008 and 2009 in
the semiarid wheat-growing region in Southern Spain (Genil-Cabra
Irrigation District). All 2008 and 2009 cloudless Landsat-5 and
Landsat-7 images (a total of 10) and one partially cloudy Landsat-
7 image were acquired for this region and processed. In addition,
one 2008 IRS-Awifs image was also acquired and processed to
fill in a gap in the Landsat acquisitions resulting from overcast
conditions at the study site. Geometric and atmospheric correc-
tions were applied to the image data. Geometric correction using
a large number of ground control points was performed based
on a high-resolution orthographic photo taken in 2004 as a ref-
erence. Each image was subsequently corrected for atmospheric
transmittance to obtain surface reflectance values using the MOD-
TRAN 4 radiative transfer model (Berk et al., 1998). NDVI maps
were constructed using the red and NIR reflectance data for each
remote sensing image, and average values were extracted from
each map to assemble a temporal set of NDVI values for each
study field. Empirical relationships between LAI and NDVI devel-
oped from the experimental plot data were used to estimate LAI
for each commercial field from the satellite data. Table 2 lists the
images used in the study for both years. For each year, the yield
for each field was supplied by the farmers to provide data to val-
idate the corresponding yield estimates produced by the GRAMI
model.
2.3. Crop-specific parameter evaluation
Certain parameters in the GRAMI model (light-use efficiency,
degree-days to phenological stages, and yield partitioning factor)
are crop-specific. Their values were evaluated using information
from the durum and bread wheat experimental plots to ver-
ify that they were appropriate for the wheat cultivars common
to this region. Light-use efficiency (ε) was evaluated using the
biomass data collected from the experimental farm. This crop-
specific parameter was evaluated by running the model for the
Córdoba field experiment with different values for ε and determin-
ing which value produced the smallest error between simulated
and observed above-ground dry mass (AGDM). The number of
149
growing degree days to reach emergence, flowering and matu-
rity was estimated using the phenological information observed
for each cultivar. This involved dividing the 28 cultivars at the
Cañete de las Torres experiment into groups with similar pheno-
logical development according to the observed Zadoks stage. The
crop-specific yield partitioning factor in the GRAMI model was esti-
mated separately for each phenological group using data from the
experimental plot study. Once appropriate values for ε and crop
phenology had been determined, the value of the partitioning factor
was evaluated by running the model with different values for the
factor and determining which produced the smallest error between
simulated and observed yield. The evaluated values of ε, crop phe-
nology and partitioning factor remained constant during the model
validation stage.
2.4. Validation of satellite-derived and ground-derived NDVI
A comparison between satellite-derived and ground-derived
NDVI was performed to assess if satellite imagery could be suc-
cessfully used to calibrate the GRAMI model at the field scale. NDVI
values derived from the acquired Landsat-5 and Landsat-7 images
were compared to corresponding NDVI values derived from the
ground radiometric data collected at the site located at the IFAPA
Alameda del Obispo Experimental Farm (Córdoba) (see Table 2).
The change over time in ground-derived NDVI was obtained by lin-
ear interpolation of the values of NDVI between the observation
dates. Interpolated values of ground-derived NDVI were identified
for the dates with satellite image acquisitions for comparison with
corresponding satellite-derived values. Comparison of the NDVI
data from the two sources was performed using linear regression
analysis.
2.5. Model validation
Once the performance of the model had been verified against
the experimental plot data and crop-specific parameters and phe-
nological development evaluated, yield estimations for 2008 and
2009 were made for the commercial wheat fields. The meteorolog-
ical inputs for the model were supplied by the Santaella weather
station (RIA). Values of LAI used for within-season calibration of
the model simulations for each field were estimated by applying
the empirical LAI–VI relationship to the NDVI data derived from the
Landsat-5 and Landsat-7 imagery acquired for the study region dur-
ing the growing season. Planting dates for the farmers’ fields were
not recorded, so they were estimated based on knowledge of farm-
ers’ planting practices in the region. Planting dates typically range
between November 10 and December 10 in this region. In addition,
most farmers usually plant wheat just before a predicted heavy rain.
Thus, the planting date for each field for the 2 years of the study
was established to be 2 days prior to the occurrence of substantial
rainfall in the weather records within the typical range of planting
dates. As a result, November 19 was established as the planting date
for all the commercial fields for the 2008 season, while December
1 was established as the planting date for all the commercial fields
for the 2009 season. The model was run for each field using the
weather data and remotely sensed estimates of LAI, and the result-
ing yield estimates were compared to the yield data supplied by the
farmers. Most of them reported their field average yield based on
total grain weight measured at market. The mean estimated (mod-
elled) yield for the 49 commercial wheat fields was statistically
compared to the corresponding mean observed (farmer-reported)
yield for each year to assess the accuracy of the methodology in
monitoring regional crop yields.
150 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154

a 6 1
4.91x
y = 0.04e
2 y = 0.98x + 0.02
r = 0.91 2
0.8 r = 0.99
4
LAI

Satellite-derived NDVI
0.6
2

0.4
0
0.0 0.2 0.4 0.6 0.8 1.0
NDVI 0.2
b 6
7.35x
y = 0.05e
2
0
r = 0.9 0 0.2 0.4 0.6 0.8 1
4
Ground-derived NDVI
LAI

Fig. 3. Satellite-derived normalized difference vegetation index (NDVI) versus

2 ground-derived NDVI at the Córdoba experiment. The thin solid diagonal line rep-
resents the 1:1 line, while the dark line segment represents the linear regression
through the points.

0
was used in the estimation of wheat yields for the commercial
0.0 0.2 0.4 0.6 0.8 1.0
wheat fields to estimate LAI for calibrating the GRAMI model from
SAVI NDVI values extracted from satellite image data
c 6
6.59x 3.2. Comparison of ground-derived and satellite-derived NDVI
y = 0.06e
2
r = 0.88 Caution must be exercised in applying Eq. (7), which was devel-
4
oped from field spectroradiometer data, to the satellite data, due
LAI

to the differences in the sensors (Jackson and Huete, 1991). Fig. 3

shows field-averaged values of NDVI obtained from spectrora-
2
diometer observations on the 2-ha wheat field located at the IFAPA
Alameda del Obispo Experimental Farm (Córdoba) plotted versus
corresponding values of NDVI derived from concurrent Landsat
0 observations made on that field. While the sample size is limited
0.0 0.2 0.4 0.6 0.8 1.0 to 6 observations, the range of NDVI values in this sample covers
EVI the entire range of NDVI (essentially from bare soil to full canopy
cover) used to develop Eq. (7). Results presented in Fig. 3 show
Fig. 2. Relationship between (a) normalized difference vegetation index (NDVI), (b) a strong relationship between satellite-derived NDVI and ground-
soil adjusted vegetation index and (c) enhanced vegetation index (EVI) and leaf area derived NDVI, with a coefficient of determination (r2 ) of 0.99 and an
index (LAI) data from the 28 wheat cultivars.
RMSE of 0.015. Similar results (r2 ) were reported by González-Dugo
and Mateos (2008) where spectral radiometry measurements were
taken over cotton and sugarbeet. Therefore, we are reasonably con-
fident that the use of Eq. (7) to estimate LAI from satellite data was
3. Results and discussion
appropriate.

3.1. LAI estimation from remote sensing data

3.3. Phenological stages

Fig. 2 shows values of LAI obtained from measurements made

Substantial differences in phenological development were
in the experimental plots at the site located near Cañete de las Tor-
found among the wheat cultivars in the experimental plot study.
res (Córdoba) plotted versus corresponding values of the three VI
From the analysis of phenological data, it appeared that the culti-
(NDVI, SAVI and EVI) used in this study determined from spectro-
vars could be divided into two groups (Groups “A” and “B”) with
radiometer measurements made in those plots. Also shown is the
similar phenology. The change over time in LAI for three culti-
exponential function fit to each set of data. The estimation ability of
vars of each group is presented in Fig. 4, where general differences
each nonlinear relationship was assessed from ordinary regression
between the two groups can be seen. The maximum LAI was
and root mean square error (RMSE) analysis between estimated
reached on approximately DAY 112 for cultivars in Group A (see
and measured LAI values (data not shown). RMSE values of 0.33,
Fig. 4), while cultivars in Group B exhibited maximum LAI approxi-
0.47 and 0.58 were obtained for the relationships involving NDVI,
mately 12 days after Group A. In addition, Group B cultivars ripened,
SAVI and EVI, respectively. The RMSE associated with NDVI was
on average, 10 days later than those in Group A. Table 1 shows
substantially less than the RMSE associated with the two other VI.
the phenological group that each cultivar in the experimental plot
Based on these results, NDVI was selected as the best of the three
study was assigned to based on these observations. Average dates
VI to estimate LAI from remote sensing data. This relationship
of crop emergence, flowering, and maturity, along with the calcu-
lated growing degree-days (base temperature = 0 ◦ C) from planting
LAI = 0.04e4.91(NDVI) (7) to these dates, are summarized in Table 3 for the two groups.
 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154 151

a 3.5 a 1400

3 1200
MEASURED LAI

Ancalei Measured AGDM

2.5 1000

AGDM (g m-2)
Vitrosol Simulated AGDM
2 Ramirez
800
1.5 600
1
400
0.5
200
0
0
0 20 40 60 80 100 120 140 160
43 56 71 78 96 113 125 139 175
DAY DAY
3.5 b 1400
b
y=x

Simulated AGDM (g m-2)

3 1200
2
r = 0.99
MEASURED LAI

Osado
2.5 1000
Gazul
2 Sculptur 800

1.5 600

1 400

0.5 200

0 0
0 20 40 60 80 100 120 140 160 0 200 400 600 800 1000 1200 1400
DAY Measured AGDM (g m-2)

Fig. 4. Measured LAI for three cultivars of phenological Group A (a) and three of Fig. 5. Change over time in measured and simulated aboveground dry mass at
phenological Group B (b). The arrow indicates the time when maximum LAI is Córdoba experiment (a) and regression between the measured and simulated above-
reached. ground dry mass at the same experiment (b).

Table 3 3.4. Light-use efficiency (ε)

Observed dates of phenological stages. Numbers in parentheses are number of grow-
ing degree days (◦ C) to reach the phenological stage.
The value of the light use efficiency parameter that minimized
Group Phenological stage the estimation error between simulated and observed above-
ground dry mass (AGDM) was 2.5 g MJ−1 PAR (RMSE = 56.76 g m−2 ).
Planting Emergence Flowering Maturity
Fig. 5a presents the GRAMI simulation of AGDM with ε = 2.5 g MJ−1
A 23 December 5 January 27 April 6 June
PAR compared to the observed values of AGDM measured for
(122) (1375) (2150)
B 23 December 7 January 6 May 15 June
that field. Fig. 5b shows simulated AGDM plotted versus observed
(133) (1531) (2374) AGDM from Fig. 5a. Simulated AGDM agreed with measured AGDM
with a coefficient of determination (r2 ) of 0.99. The value of ε
determined in this study is the same the value used in previous
GRAMI modelling efforts involving wheat (Maas, 1992, 1993a,c).
Information regarding wheat cultivars was not recorded for It is also in the range reported by other researchers, such as
the 49 commercial wheat fields from the Genil-Cabra Irrigation 2–2.4 g MJ−1 (Gallagher and Biscoe, 1978; Lobell et al., 2003),
District in Southern Spain used in validating the yield estimation 2.92 g MJ−1 (Asrar et al., 1984), 2.82–3.22 g MJ−1 (Garcia et al.,
methodology. Thus, it was not possible to explicitly identify the 1988), and 1.4–1.74 g MJ−1 (Gregory et al., 1992).
phenology group for individual fields in this data set. However, it
was possible to identify when the peak in the curve constructed 3.5. Yield partitioning factor
for each commercial field from the satellite-derived LAI values
occurred. Comparing the timing of these peaks to similar curves The crop-specific yield partitioning factor in the GRAMI model
constructed for the cultivars in the experimental plot study (see was estimated separately for each phenological group. A single
Fig. 4), it was determined that the commercial fields were likely value (0.8) for the partitioning factor minimized the estimation
planted to Group B cultivars. Thus, growing degree-day sums listed error between simulated and observed yield for both pheno-
in Table 3 for phenology Group B were used to model the dates logical groups. Using this value, mean estimated and observed
of crop emergence, flowering and maturity for the commercial yields for Group A were 4198 and 4120 kg ha−1 , respectively, while
fields. mean estimated and observed yields for Group B were 4022 and
152 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154

Table 4
a 8000
Crop-specific parameters used in this study for all simulations at regional scale.

Name Value Units

Estimated Yield (kg ha-1)

Crop-specific parameters
6000 ◦
Base temperature (Tb ) 0 C
Fraction of total solar irradiance that is PAR (ˇ) 0.45 –
Specific leaf area (S) 0.024 m2 g−1
Light extinction coefficient (k) 0.6 –
4000 Light use efficiency (ε) 2.5 g MJ−1 PAR
Yield partitioning factor 0.8 –
Phenological development (see Table 3) Group B –

2000
a
8000
0

Estimated Yield (kg ha-1)

0 2000 4000 6000 8000
6000
Observed Yield (kg ha-1)

b 8000
4000
Estimated Yield (kg ha-1)

6000
2000

4000
0
0 2000 4000 6000 8000

2000 Observed Yield (kg ha-1)

b
0 8000
Estimated Yield (kg ha-1)

0 2000 4000 6000 8000

Observed Yield (kg ha-1)
6000
Fig. 6. Wheat yield estimation of (a) 18 cultivars which belong to phenology Group
A and (b) 10 cultivars which belong to phenology Group B using the GRAMI model
at the Cañete de las Torres experiment. The thin solid diagonal line represents the
1:1 line.
4000

4021 kg ha−1 , respectively. For each group, the mean estimated and 2000
observed yields were not significantly different (t = 0.333 for Group
A and t = 0.002 for Group B with 34 and 18 df, respectively, ˛ = 0.05).
Use of 0.8 for the yield partitioning factor resulted in yield estima-
0
tion errors of 12.76 and 12.80% for Groups A and B, respectively,
with corresponding RMSE of 532 and 515 kg ha−1 . Results of these
0 2000 4000 6000 8000
simulations are presented in Fig. 6a and b, respectively, for Groups Observed Yield (kg ha-1)
A and B. The good results achieved by simulating crop yield under
experimental plot conditions can be partly explained by the fact Fig. 7. Wheat yield estimation for 49 fields in 2008 (a) and 2009 (b) using the GRAMI
that the planting date and phenological stages were known for this model and satellite imagery. The thin solid diagonal line represents the 1:1 line.

experiment, and similar management practices were applied to all

the plots.
3.6. Regional scale yield assessment (model validation)
Once the GRAMI model had been verified (i.e., crop-specific
parameters and phenological development evaluated) and a rela-
tionship for estimating crop LAI from satellite data had been
established, the yield estimation procedure was applied to an inde-
pendent data set, the two years of commercial wheat field data
from the Genil-Cabra Irrigation District in Southern Spain. GRAMI
was run for the commercial fields using the previously estab-
lished values for the crop-specific model parameters. As previously
described, the cultivars in all the commercial fields in both years
were assumed to be from phenology Group B. Table 4 shows the
value of the constant parameters used in this study for all simula-
tions at the regional scale.
Fig. 7a shows estimated (modelled) yields for the 49 commercial
wheat fields in the 2008 season plotted versus the corresponding
observed (farmer-reported) yields. Similar results are presented in
Fig. 7b for the 2009 season. Table 5 summarizes the average esti-
mated and observed yields for the 2 years. For both years, the points
in Fig. 7a and b tend to cluster along the 1:1 line. Scatter among the
points about the 1:1 line is substantial, with average absolute errors
(AAE, the average amount that the yield for an individual field is
over- or under-estimated) of 884 and 852 kg ha−1 for the 2008 and
2009 seasons, respectively. These error values represent 21% and
 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154
Table 5
Results of simulating yield at field scale during the 2008 and 2009 growing seasons.
Growing season Observations Planting date
2008 49 November 19th
2009 49 December 1st
a
Average values (kg ha−1 ) for the 49 fields.
18% of the average observed yield for the 49 fields in the study. They
are comparable to values reported from previous validations of the
GRAMI model for grain crops. Maas (1988b) reported an AAE repre-
senting 21% of the observed average yield for simulations of grain
sorghum in Central Texas. Maas (1993c) reported an AAE repre-
senting 21% of the observed average yield for simulations of winter
wheat at three locations in the U.S. Great Plains. A number of factors
can contribute to the magnitude of estimation errors for individual
fields in a data set. In this study, the use of a single planting date
for all the fields in a given year, along with the specification of a
single phenological group for all the fields, probably accounts for
some degree of the error among the individual yield simulations.
Also, some of this error is likely in the form of measurement error
within the farmer-reported observed yield values.
Of more relevance to assessing the utility of this methodology
for monitoring regional crop yield is the comparison of the average
estimated and observed yields. Regional food production for grain
crops can be determined as the product of the average crop yield
for the region and the land area in the regional devoted to the crop.
For accurate monitoring of crop production in these situations, the
average estimated yield should be a good estimator of the average
actual yield. In this study, the absolute errors between the average
estimated and average observed yield values were 5.44% and 6.86%
for the 2008 and 2009 seasons, respectively (see Table 5). For each
of the 2 years, the average estimated yield was not significantly dif-
ferent from its corresponding average observed yield (t = 0.980 for
2008 and t = −1.281 for 2009 with 96 df, ˛ = 0.05). A similar situation
was reported by Maas (1988b, 1993c) involving GRAMI estimates
of average winter wheat and grain sorghum yields. The absolute
errors between the average estimated and average observed yield
values reported for this study (5.44% and 6.86%) are similar to the
value (5.8%) reported by Maas (1993c) for GRAMI simulations of
winter wheat. They are somewhat greater than the absolute error
(2%) reported by Maas (1988a,b) for GRAMI simulations involving
grain sorghum.
4. Conclusions
A methodology has been developed for estimating regional crop
yields using the GRAMI model and satellite remote sensing. The
steps of this methodology are as follows:
(1) Verify the performance of the model for the crop and region.
This involves evaluating crop-specific model parameters and
crop phenology using existing agronomic data, possibly from
experimental plot studies.
(2) Establish a relationship between crop LAI and satellite-derived
VI. This relationship allows LAI data for within-season calibra-
tion of the model to be estimated using satellite observations
of fields in the region.
(3) Choose a sample of fields within the region for which the aver-
age observed yield can be considered representative of the
average yield of all fields in the region planted to the crop (i.e.,
the average regional yield).
(4) Run the verified model for this sample of fields using weather
data (daily average air temperature and daily total solar irradi-
ance) and satellite-derived LAI estimates. The resulting average
153
Observed yielda Estimated yielda Error (%)
4191 3963 5.44
4708 5031 6.86
modelled yield for the sample of fields should represent an
estimate of the average regional yield.
Results of modelling wheat yields over 2 years in a semiarid
region in Southern Spain (Genil-Cabra Irrigation District) using this
methodology showed reasonable agreement between estimated
and observed yields. Results of this study are in good agreement
with those of previous studies involving the GRAMI model. Based
on these results, we can offer the following conclusions regarding
the accuracy of the GRAMI model as applied to estimating yields
using this methodology:
(1) The average amount that the yield of an individual field in a
region is over- or under-estimated should be approximately
20% of the average observed yield for the region.
(2) Numerically, the average yield for the region should be esti-
mated to within 7% of the average observed yield for the region.
(3) Statistically, the average estimated yield for the region should
not be significantly different from the average observed yield
for the region (˛ = 0.05).
Based primarily on its ability to accurately estimate the average
yield for a set of fields and its reliance on readily available weather
and remote sensing data, we conclude that the GRAMI model, veri-
fied for a region and calibrated using satellite image data, appears to
be a practical and appropriate option for operationally monitoring
regional crop yields with a reasonable degree of accuracy.
Acknowledgments
The authors wish to recognize the support of CICE-Junta de
Andalucía through the project P06-AGR-2317. We also would like
to acknowledge the contributions of IFAPA (Instituto Andaluz de
Investigación y Formación Agraria y Pesquera) through access to
their experimental fields, which are established and maintained
by personnel of the RAEA network. The authors wish to thank Mr.
Berlanga and Mrs. Carmona (manager and technician of the Genil-
Cabra Irrigation District) for allowing us to collect yield data from
fields worked by their associated farmers.
References
Aparicio, N., Villegas, D., Casadesús, J., Araus, J.L., Royo, C., 2000. Spectral vegetation
indices as nondestructive tools for determining durum wheat yield. Agron. J. 92,
83–91.
Asrar, G., Fuchs, M., Kanemasu, E.T., Hatfield, J.L., 1984. Estimating absorbed pho-
tosynthetically active radiation and leaf area index from spectral reflectance in
wheat. Agron. J. 76, 300–306.
Asrar, G., Mineny, R.B., Kanemasu, E.T., 1989. Estimation of plant-canopy attributes
from spectral reflectance measurements. In: Asrar, G. (Ed.), Theory and Applica-
tions of Optical Remote Sensing. John Wiley and Sons, Chichester, pp. 252–292.
Balaghi, R., Tychan, B., Eerens, H., Jlibene, M., 2008. Empirical regression models
using NDVI, rainfall and temperature data for the early prediction of wheat grain
yields in Morocco. Int. J. Appl. Earth Observ. Geoinfo. 10, 438–452.
Bastiaanssen, W.G.M., Ali, S., 2003. A new crop yield forecasting model based on
satellite measurement applied across the Indus Basing. Agric. Ecosyst. Environ.
97, 321–340.
Berk, A., Bernstein, L.S., Anderson, G.P., Acharya, P.K., Robertson, D.C., Chetwynd, J.H.,
Adler-Golden, S.M., 1998. MODTRAN cloud and multiple scattering upgrades
with application to AVIRIS. Remote Sens. Environ. 65, 367–375.
Brisson, N., Gary, C., Justes, E., Roche, R., Mary, B., Ripoche, D., Zimmer, D., Sierra,
J., Bertuzzi, P., Burger, P., Bussière, F., Cabidoche, Y.M., Cellier, P., Debaeke, P.,
154 F.L.M. Padilla et al. / Field Crops Research 130 (2012) 145–154
Gaudillère, J.P., Hènault, C., Maraux, F., Seguin, B., Sinoquet, H., 2003. An overview
of the crop model STICS. Eur. J. Agron. 18, 309–332.
Brisson, N., Mary, B., Ripoche, B., Jeuffroy, D., Ruget, M.H., Nicoullaud, F., Gate, B.,
Devienne-Barret, P., Antonioletti, F., Durr, R., Richard, C., Beaudoin, G., Recous,
G., Tayot, S., Plenet, X., Cellier, D., Machet, P., Meynard, J.M., Dellécolle, J.M., 1998.
STICS: a generic model for the simulation of crops and their water and nitrogen
balance. I. Theory and parameterization applied to wheat and corn. Agronomie
(Paris) 18, 311–346.
Charles-Edwards, D.A., Doley, D., Rimmington, G.M., 1986. Modeling Plant and
Development. Academic Press, Orlando, FL.
Charles, J., Godfray, J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir,
J.F., Pretty, J., Robinson, S., Thomas, S.M., Toulmin, C., 2010. Food security: the
challenge of feeding 9 billion people. Science (Washington, DC) 327, 812–818.
Dalezios, N.R., Domenikiotis, C., Loukas, A., Tzortzios, S.T., Kalaitzidis, C., 2001. Cotton
yield estimation based on NOAA/AVHRR produced NDVI. Phys. Chem. Earth (B)
26, 247–251.
Gallagher, J.N., Biscoe, P.V., 1978. Radiation absorption, growth and yield of cereals.
J. Agric. Sci. (Cambridge) 91, 47–60.
Garcia, R., Kanemasu, E.T., Blad, B.L., Bauer, A.A., Hatfield, J.L., Major, D., Reginato,
R.J., Hubbard, K.G., 1988. Interception and use efficiency of light in winter wheat
under different nitrogen regimes. Agric. For. Meteorol. 44, 175–186.
González-Dugo, M.P., Mateos, L., 2008. Spectral vegetation indices for benchmarking
water productivity of irrigated cotton and sugarbeet crops. Agric. Water Manage.
95, 48–58.
Gregory, P.J., Tennant, D., Belford, R.K., 1992. Root and shoot growth and water and
light use efficiency of barley and wheat crop grown on a shallow duplex soil in
a Mediterranean-type environment. Aust. J. Agric. Res. 43, 555–573.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25,
295–309.
Huete, A.R., Didan, K., van Leeuwen, W., Miura, T., Glenn, E., 2008. MODIS vegetation
indices. In: Land Remote Sensing and Global Environmental Change: NASA’s
Earth Observing System and the Science of ASTER and MODIS.
Jackson, R.D., Huete, A.R., 1991. Interpreting vegetation indices. Prev. Vet. Med. 11,
185–200.
Jones, J.W., Tsuji, G.Y., Hoogenboom, G., Hunt, L.A., Thornton, P.K., Wilkens, P.W.,
Imamura, D.T., Bowe, W.T., Singh, U., 1998. Decision support system for
agrotechnology transfer; DSSAT v3. In: Tsuji, G.Y., Hoogenboom, G., Thornton,
P.K. (Eds.), Understanding Options for Agricultural Production. Kluwer Academic
Publisher, Dordrecht, The Netherlands, pp. 157–177.
Jones, J.W., Hoogenboom, G., Porter, C.H., Boote, K.J., Batchelor, W.D., Hunt, L.A.,
Wilkens, P.W., Singh, V., Gijsman, A.J., Ritchie, J.T., 2003. The DSSAT cropping
system model. Eur. J. Agron. 18, 235–265.
Labus, M.P., Nielsen, G.A., Lawrence, R.L., Engel, R., Long, D.S., 2002. Wheat yield
estimates using multi-temporal NDVI satellite imagery. Int. J. Remote Sens. 23,
4169–4180.
Lobell, D.B., Asner, G.P., Ortiz-Monasterio, J.I., Benning, T.L., 2003. Remote sensing of
regional crop production in the Yaqui Valley, Mexico: estimates and uncertain-
ties. Agric. Ecosyst. Environ. 94, 205–220.
Ma, B.L., Dwyer, L.M., Costa, C., Cober, E.R., Morrison, M.J., 2001. Early prediction of
soybean yield from canopy reflectance measurements. Agron. J. 93, 1227–1234.
Maas, S.J., 1988a. Use of remotely-sensed information in agricultural crop growth
models. Ecol. Modell. 41, 247–268.
Maas, S.J., 1988b. Using satellite data to improve model estimates of crop yield.
Agron. J. 80, 655–662.
Maas, S.J., 1992. GRAMI: A Crop Growth Model that can use Remotely Sensed Infor-
mation. Publication ARS-91, U.S. Department of Agriculture, Agricultural Res.
Service, Washington, DC.
Maas, S.J., 1993a. Parameterized model of gramineous crop growth. I. Leaf area and
dry mass simulation. Agron. J. 85, 348–353.
Maas, S.J., 1993b. Parameterized model of gramineous crop growth: II. Within-
season simulation calibration. Agron. J. 85, 354–358.
Maas, S.J., 1993c. Within-season calibration of modeled wheat growth using remote
sensing and field sampling. Agron. J. 85, 669–672.
Mkhabela, M.S., Mashinini, N.N., 2005. Early maize yield forecasting in the four agro-
ecological regions of Swaziland using NDVI data derived from NOAA’s-AVHRR.
Agric. For. Meteorol. 129, 1–9.
Monteith, J.L., 1972. Solar radiation and productivity in tropical ecosystems. J. Appl.
Ecol. 9, 747–766.
Monteith, J.L., Unsworth, M.H., 1990. Principles of Environmental Physics, 2nd ed.
Edward Arnold, New York.
Padilla, F.L.M., Gónzalez-Dugo, M.P., Mansilla, F., Gavilán, P., Domínguez, J., 2010.
Wheat yield monitoring in Southern Spain using a series of satellite images. In:
Proceeding of 3rd International Symposium Recent Advances in Quantitative
Remote Sensing, Valencia (Spain).
Prasad, A.K., Chai, L., Singh, R.P., Kafatos, M., 2006. Crop yield estimation model for
Iowa using remote sensing and surface parameters. Int. J. Appl. Earth Observ.
Geoinfo. 8, 26–33.
Rouse, J.W., Haas, R.W., Shell, J.A., Deering, D.W., Harlan, J.C., 1974. Monitoring the
Vernal Advancement and Retrogradation of Natural Vegetation. NASA/GSFC.
Type III, Final Report. Greenbelt, MA, USA.
Stöckle, C.O., Donatelli, M., Nelson, R., 2003. CropSyst a cropping system model. Eur.
J. Agron. 18, 289–307.
Stöckle, C.O., Martin, S., Campbell, G.S., 1994. CropSyst a cropping system model:
water/nitrogen budgets and crop yield. Agric. Syst. 46, 335–359.
Wall, L., Larocque, D., Léger, P.M., 2007. The early explanatory power of NDVI in crop
yield modelling. Int. J. Remote Sens. 1, 1–15.
Weiss, M., Troufleau, D., Baret, F., Chauki, H., Prevot, L., Olioso, A., Bruguier, N., Bris-
son, N., 2001. Coupling canopy functioning and radiative transfer models for
remote sensing data assimilation. Agric. For. Meteorol. 108, 113–128.
Zadoks, J.C., Chang, T.T., Konzak, C.F., 1974. Decimal code for growth stages of cereals.
Weed Res. 14, 415–421.
Zhao, D., Reddy, K.R., Kakani, V.G., Read, J.J., Koti, S., 2007. Canopy reflectance in cot-
ton for growth assessment and lint yield prediction. Eur. J. Agron. 26, 335–344.
Zwart, S.J., Bastiaanssen, W.G.M., 2007. SEBAL for detecting spatial variation of water
productivity and scope for improvement in eight irrigated wheat systems. Agric.
Water Manage. 89, 287–296.