Field Crops Research 94 (2005) 126–148
www.elsevier.com/locate/fcr
Assessment of durum wheat yield using visible and
near-infrared reflectance spectra of canopies
J.P. Ferrioa, D. Villegasb, J. Zarcob, N. Apariciob, J.L. Arausc, C. Royob,*
a
Departament de Producció Vegetal i Ciencia Forestal, Universitat de Lleida, Rovira Roure 191, 25198 Lleida, Spain
b
IRTA, Àrea de Conreus Extensius, Centre UdL-IRTA, Rovira Roure 191, 25198 Lleida, Spain
c
Unitat de Fisiologı́a Vegetal, Facultat de Biologia, Universitat de Barcelona, Diagonal 645, 08028 Barcelona, Spain
Received 29 July 2004; received in revised form 30 November 2004; accepted 20 December 2004
Abstract
The estimation of grain yield before harvesting could be a very useful tool for breeding programs and productivity
forecasting. Canopy reflectance indices have been used for yield estimation, but with limited success. This work was carried out
to study the suitability of the visible and near-infrared reflectance spectrum of the canopy for the assessment of grain yield in a
set of durum wheat genotypes. Five field experiments, each one including 25 genotypes, were conducted in low, medium and
high productivity environments, with average yields of 2.5, 4.5 and 7 t/ha. Spectral reflectance measurements between 400 and
1000 nm were made at anthesis and milk-grain stages. Partial least squares regression (PLSR) was used in the construction of
models that were tested by simple regression between genotype means of predicted and observed grain yields. The empirical
models for the estimation of grain yield showed generally stronger and more robust assessment of grain yield than previously
assayed spectral indices. For the best model, correlation coefficients between genotype means of predicted and measured yield
within each of the five environments ranged from 0.53 to 0.76. We concluded that, although the models did not provide an
accurate quantification of grain yield, they could still be used to rank genotypes for breeding purposes. The most reliable ranking
of genotypes was attained using measurements made at milk-grain stage on medium to high productivity environments.
# 2004 Elsevier B.V. All rights reserved.
Keywords: Canopy reflectance; PLSR; Anthesis; Milk-grain; Multivariate analysis; Breeding spectroradiometry
1. Introduction
Empirical breeding, based on selection of yield per
se, has been very effective in raising wheat produc* Corresponding author. Tel.: +34 973 70 25 83;
fax: +34 973 23 83 01.
E-mail address: [email protected] (C. Royo).
tions in the past. Nevertheless, the assessment of grain
yield requires the harvest of the experimental plots,
which is expensive and time-consuming when it has to
be done in the large sets of genotypes that are usually
managed by plant breeders. The estimation of plots
productivity in a non-destructive way before harvesting would be a very useful tool for selection, mainly in
the early generations of a breeding program.
0378-4290/$ – see front matter # 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.fcr.2004.12.002
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
The measurement of the spectral signature of crop
canopies at visible and near-infrared (VIS/NIR)
regions of the electromagnetic spectrum has shown
to be useful to monitor crop growth conditions (Bauer,
1975; Walburg et al., 1982). Spectral reflectance
measurements have been successfully used to estimate
biomass, leaf area index, photosynthesis and/or yield
in several species of trees (Gamon et al., 1995;
Richardson et al., 2001), rice (Vaesen et al., 2001),
barley (Bort et al., 2002), bread wheat (Filella et al.,
1995) and durum wheat (Aparicio et al., 2000, 2002,
2004; Royo et al., 2003).
To manage the information given by the spectrum,
vegetation indices, defined as simple operations
between reflectance values at given wavelengths, are
often used (Field et al., 1994). Some indices are
related to the photosynthetic active biomass, such as
the normalized difference vegetation index (NDVI),
or the simple ratio (SR, see Peñuelas et al., 1997a),
both being widely used. Vegetation indices have
been used to estimate biomass (Aparicio et al., 2002)
and yield (Aparicio et al., 2000) of durum wheat, but
phenotypic correlation coefficients found are usually
weak and largely dependent on the range of variation
of the tested material (Royo et al., 2003). Thus,
studies based on comparison among several species
have shown much promising results (Peñuelas et al.,
1995a, 1997b; Gamon et al., 1997) than those
obtained when comparing genotypes of a given
species under a single environment (Royo et al.,
2003), unless a gradient of variation is introduced in
the environment due to, e.g. fertilization (Serrano
et al., 2000; Vaesen et al., 2001) or salinity (Peñuelas
et al., 1997a). To overcome this problem, some
authors have tried to integrate the information given
by each index separately by analyzing all them
together, obtaining interesting relationships between
a combination of indices and chlorophyll concentrations (Filella et al., 1995). Raun et al. (2001)
proposed to add NDVI values collected at Feekes
growth stages 4 and 5 and divide the result by the
GDD between readings to obtain an indication of the
potential grain yield of winter wheat, although in a
wide range of growing conditions and planting
times.
In such context, it becomes useful to study the VIS/
NIR spectrum and try to relate all its intrinsic
information into a model to estimate genotype
127
variability (i.e. within a species and environment) in
important traits such as yield. However, since the
potential chemical and physiological basis of the link
between grain yield and the whole reflectance spectra
of the canopy is not fully understood, empirical
calibration is needed to model grain yield from raw
spectral data. For empirical calibration using spectral
data, multivariate approaches, like partial least squares
regression (PLSR), are the most recommended (Beebe
and Kowalski, 1987; Martens and Naes, 1991). PLSR
decomposes the variability of the spectrum matrix into
a number of factors that are not optimal for describing
this matrix, but are rotated to simultaneously describe
the variable to regress. Therefore, it is especially
suitable for this approach, as the main sources of
variation in reflectance spectrum are different from
those directly associated with grain yield.
The objective of this work was to study the
suitability of the VIS/NIR reflectance spectrum of the
canopy to assess grain yield of a set of durum wheat
genotypes. Additional objectives were to determine
the influence of environment, through definition of
which environment was more adequate to apply this
technique, as well as the predictive value of a model
calibrated within a given environment when applied to
another one. Finally, we studied the influence of the
phenological stage in which the spectra were taken on
the ability to assess grain yield.
2. Materials and methods
2.1. Experimental setup
Five field experiments were carried out at three
sites of northeastern Spain in 1998 and 1999 (see
details in Table 1). Each experiment consisted of 25
durum wheat genotypes sown in a randomized
complete block design with four replicates, in plots
of 12 m2 (six rows, 20 cm apart). The genotypes
included four commercial Spanish cultivars (Altaraos, Jabato, Mexa and Vitrón) and 21 advanced lines
of the CIMMYT/ICARDA durum wheat breeding
program (Awalbit, Bicrecham-1, Chacan, Chahra-1,
Haurani, Korifla, Krs/Haucan, Lagost-3, Lahn/
Haucan, Massara-1, Moulchahba-1, Mousabil-2,
Omlahn-3, Omrabi-3, Omruf-3, Quadalete//Erp/
Mal, Sebah, Stojocri-3, Waha, Zeina-1 and Zeina-2).
128
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Table 1
Description and main agronomical characteristics of the sites where trials were performed, including mean yield values achieved
Coordinates
Altitude (m above
the sea level)
Soil type (USDA)
Soil texture
Sowing date
Total rainfall received
by the crop (mm)
Total irrigation (mm)
Mean temperature (8C)
Yield (average standard
error, kg ha1)
Productivity classification
Trial code
El Canós
Gimenells
Palau d’Anglesola
418410 N, 18130 E
440
418400 N, 08200 E
200
418390 N, 18120 E
200
Fluventic-Xerochrept
Loamy-fine
Calcixerolic-Xerochrept
Fine-loamy
Aquic-Xerofluvent
Fine-loamy
1998
17 November 1997
183
1999
3 November 1998
256
1998
23 November 1997
285
1999
19 November 1998
293
1999
10 November 1998
255
0
9.8
2531 56
0
10.0
3820 111
100
10.3
5202 181
100
11.6
4052 77
150
9.6
7009 73
Low
LR (low
productivity,
rainfed 1998)
Medium
MR (medium
productivity,
rainfed 1999)
Medium
MI1 (medium
productivity,
irrigated 1998)
Medium
MI2 (medium
productivity,
irrigated 1999)
High
HI (high
productivity,
irrigated 1999)
The genotypes were chosen to represent a wide range of
genetic variability in terms of agronomical characteristics. Seed rate was adjusted to 550 viable seeds m2.
Soil analyses were done prior to sowing, and
appropriate fertilization was provided according to
the common agronomical practices at each site. Weeds
and diseases were controlled, when necessary, using
chemicals.
The experiments were classified according to
their productivity as: low, medium and high,
corresponding to average yields of 2500, 4500 and
7000 kg ha1, respectively. Low productivity class
consisted on one experiment conducted in 1998
under rainfed conditions (LR, see Table 1). Medium
productivity class included three experiments, one of
which was conducted in 1999 under rainfed
conditions and two with supplementary irrigation
in 1998 and 1999 (MR, MI1, MI2, respectively).
High productivity class consisted of one experiment
conducted in 1999 under irrigation (HI). All irrigated
experiments were flooded-irrigated and 50 mm were
applied 2–3 times at monthly intervals (see Fig. 1).
The inclusion of MR environment on medium
productivity class was made because of the relatively
high yields obtained in this experiment, mainly due
to the rains fell in March, April and May, when most
of the yield components are determined in these
environments.
2.2. Data recorded
Canopy reflectance was measured with a narrowbandwidth visible-near-infrared portable field spectroradiometer fitted with an 188 field-of-view optic
(FieldSpec UV/VNIR, Analytical Spectral Devices,
Boulder, CO, USA) as described in Aparicio et al.
(2000). The instrument detects 512 continuous bands
(with a sampling interval of 1.4 nm) from 350 to
1050 nm wavelengths, thereby covering the visible
and near-infrared portion of the spectrum. Measurements were taken with the sensor placed on a vertical
rod to take reading from a nadir position, with the
sensor raised 2 m above the ground. The measurements were made at midday under cloudless
conditions. Three readings (1–2 s each), each being
the average of five scans, were made on three
different portions of each plot. The reflectance
spectrum was calculated in real time as the ratio
between the reflected and incident spectra of the
canopy. The incident spectrum was obtained every
five plots (every minute approximately), from the
light reflected by a white reference panel with a very
close to Lambertian surface (Spectralon, Labsphere,
North Sutton, NH).
Spectral reflectance measurements were made at
mid-anthesis and milk-grain stages, corresponding to
stages 65 and 75 of the Zadoks’ scale (Zadoks et al.,
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
129
Fig. 1. Monthly minimum (*) and maximum (*) temperatures, rainfall (white bars) and irrigation (striped bars) for each trial: LR, low
productivity under rainfed conditions (year 1998) MR, medium productivity under rainfed conditions (1999); MI1, medium productivity under
irrigation (1998); MI2, medium productivity under irrigation (1999); and HI, high productivity under irrigation (1999). See more details about
the trials in Table 1.
1974), respectively. Plots were harvested mechanically at ripening, and grain yield (kg ha1) was
determined on a plot basis of 12 m2 and is reported at a
10% moisture level.
2.3. Model construction and validation
Spectra were imported into the program
Unscrambler, version 6.0 (CAMO Ltd., New Market,
130
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
United Kingdom) that included the PLSR algorithms
used in the construction of models. Only spectral
bands between 400 and 1000 nm were included in
the models, as the spectra outside these limits were
noisier and less sensitive. To correct baseline shifts,
which are often associated with structural effects,
spectra were mean-centered, but not scaled, using
the standard normal variate (SNV) algorithm
(Barnes et al., 1989). We used all the samples
(plots) within each of the trials to calibrate the
models. The models were named according to the
calibration environment (LR, MR, MI1, MI2, HI),
but adding a number to indicate the phenological
stage of measurement (6 for anthesis, 7 for milkgrain). We also assayed models calibrated with
genotype means within each trial, but they showed
poorer performance. The number of PLSR factors
used in each model was determined by full crossvalidation (Wold, 1978). It consists in building as
many models as plots comprised on each trial, each
one calibrated leaving out data of one plot from the
same trial to be validated. The optimum number of
factors was determined by minimizing the root mean
standard error (RMSE):
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
ðYref Yest Þ2
RMSE ¼
N1
where N is the number of samples, Yref are the
observed values of grain yield, and Yest the values
estimated from spectral models. However, RMSE did
not allow comparing the predictive ability within trials
with different grain yield variability. Thus, we also
calculated the relative RMSE (rRMSE), defined as the
ratio between RMSE and the standard error of grain
yield within each trial. The lower the ratio, the greater
the ability of the model to detect differences in grain
yield within the trial. The resulting regression coefficients for all the models used in this study are plotted
in Appendix A.
The robustness of the models was further tested
by applying them to the spectra acquired in different
experiments. As this work was focused on the
prediction of grain yield in breeding programs, we
assessed the relationship between genotype means of
predicted and measured values by simple regression. We took cross-validation values to estimate the
validation performance of the model within the
trial used for calibration. Finally, we calculated
broad-sense heritabilities (H2) of measured grain
yield and the values estimated by the models, using
variance components obtained from the MIXED
procedure of SAS statistical package (SAS Institute
Inc., 1987). This would indicate to what extent the
models were able to track genotypic variability in
grain yield.
3. Results
3.1. Spectral features related with grain yield and
main wavelengths included in the models
Plots with higher grain yield (GY) showed lower
SNV reflectance in the green-red region (500–700 nm)
than low-yielding plots, as exemplified in Fig. 2. In
contrast, the reflectance in the near-infrared
(>700 nm) was higher, with the notable exception
of the bands between 950 and 1000 nm. The slope for
the increase of reflectance from red to near-infrared
(red edge, lRE) was higher in high-yielding than in
low-yielding plots, and shifted towards the near-
Fig. 2. Mean SNV spectra of the 10 plots of lower and higher GY
from the medium-yielding rainfed trial at the milk-grain stage
(7MR). Main spectral regions are outlined (see text for further
details). Chl + Car, absorption bands shared by chlorophylls and
carotenoids; Chl, spectral region with chlorophyll absorption not
shared by carotenoids; Chl (lRE), wavelength of maximum slope in
the increase of reflectance from red to near-infrared (red edge);
brown pigments, wavelength region absorbed by brown pigments;
water, near-infrared region affected by tissue water content. Arrows
indicate the sense for increasing content of the referred compounds.
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
131
Fig. 3. Regression coefficients for each wavelength, resulting from the models calibrated in three different trials at two developmental stages. In
order to make comparisons easier, coefficients have been re-scaled, dividing them by the maximum absolute value for each model. Main spectral
regions are outlined (see text for further details). 6LR, 6MR, 6HI, stand for models developed from anthesis data in low-yielding, mediumyielding (rainfed) and high-yielding trials; 7LR, 7MR, 7HI, models developed from milk-grain data in the same trials; Chl + Car, absorption
bands shared by chlorophylls and carotenoids; Chl, spectral region with chlorophyll absorption not shared by carotenoids; Chl (lRE), wavelength
of maximum slope in the increase of reflectance from red to near-infrared (red edge); brown pigments, wavelength region absorbed by brown
pigments; water, near-infrared region affected by tissue water content. Arrows indicate the sense for increasing content of the referred
compounds.
infrared region. Finally, we found that reflectance in
high-yielding plots was generally higher in the blue
range, between 400 and 500 nm.
Regression coefficients (i.e. the matrix product of
the weight given to each PLSR factor and its
corresponding wavelength loadings) obtained from
PLSR calibration (Fig. 3) were coherent with these
spectral patterns. Although there were differences
among environments and/or growth stages in the
relative contribution to the models of each wavelength, we found several common features. All the
models showed high negative coefficients in the left
side of the red edge region (700–750 nm). We also
found consistent negative coefficients for the wavelengths between 950 and 1000 nm. Some models also
included positive coefficients around 800–900 nm
(6LR, 6HI, 7LR, 7HI), whereas between 400 and
500 nm we found either positive (6HI, 7LR, 7HI) or
negative (6LR, 7MR) coefficients.
3.2. Models construction and calibration
performance
Data of each individual plot were used for model
construction. The results showed that PLSR reflectance models explained between 20 and 81% of within
trial grain yield variability (Table 2). Although
estimated yield was always significantly correlated
with observed grain yield, this relationship was
stronger for low- and medium-yield environments.
Calibration RMSE ranged from 253 to 997 kg ha1,
whereas rRMSE varied between 4.4 and 8.9. Generally, RMSE and rRMSE were greater in medium and
high-yield trials. Cross-validation performance was
somewhat poorer than calibration results, but still
showed a significant relationship between predicted
and measured values (r2 from 0.16 to 0.76). Models
performance within the calibration environment was
generally similar regardless of the stage of measure-
132
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Table 2
Main statistical parameters of the models
Model
N
PLSR factors
Calibration
Cross-validation
r2
RMSE
rRMSE
r2
RMSE
rRMSE
Anthesis
6LR
6MR
6MI1
6MI2
6HI
100
100
100
100
68
7
3
4
3
2
0.79***
0.76***
0.74***
0.46***
0.23***
253
543
940
562
613
4.5
4.9
5.2
7.3
8.4
0.71***
0.74***
0.67***
0.41***
0.16***
300
568
1049
590
644
5.4
5.1
5.8
7.7
8.8
Milk-grain
7LR
7MR
7MI1
7MI2
7HI
100
100
100
98
99
2
5
1
4
2
0.61***
0.81***
0.71***
0.61***
0.20***
349
484
997
474
650
6.2
4.4
5.5
6.2
8.9
0.58***
0.76***
0.69***
0.42***
0.16***
364
547
1021
578
671
6.5
4.9
5.6
7.5
9.2
N, number of samples; r2, determination coefficient of the regression line between predicted and measured values; RMSE, root mean standard
error (kg ha1); rRMSE, ratio between RMSE and the standard error of grain yield within each trial.
***
P < 0.001.
ment. Nevertheless, we found somewhat better fit in
medium-yield trials with the models calibrated with
spectra acquired at milk-grain stage, whereas for the
drier environment (LR) the best predictions were
found in anthesis.
3.3. General trends in models performance when
comparing genotype yields
The usefulness of models to discriminate between
the yield of different genotypes was evaluated using
mean data of each genotype within each trial. Fig. 4
shows the relationship between predicted and
harvested grain yield within the calibration trial
(cross-validation values). The strongest and most
consistent relationships were found within mediumyielding trials, specially at milk-grain stage (Fig. 4c
and d). Within the low-yielding trial (LR) the
relationship was only significant at anthesis, whereas
the opposite was the case for the high-yielding trial
(HI). In all cases, slope and intercept did not differ
significantly from 1 and 0, respectively. Both RMSE
and rRMSE for genotype means were generally
smaller than for predicted plot values (data not
shown). Nevertheless, the quantification of grain
yield was still poor, with RMSE values being from
three- to five-fold greater than the standard error
across genotype means.
The robustness of the models was further tested by
applying each one to all the trials assayed (see Fig. 5).
The ability of PLSR models derived from VIS/NIR
spectra for predicting grain yield of durum wheat
varied with both developmental stage and trial where
measures were taken. Fig. 5a shows that the five
models constructed with anthesis spectra had similar
predictive ability across the five trials. In contrast, we
found greater differences in the response across trials
among milk-grain models (Fig. 5b). PLSR models had
generally better performance for predicting grain yield
of durum wheat grown in trials with medium
productivity potential. Indeed, most of the models,
either from anthesis or milk-grain spectra, showed
greater correlations between predicted and estimated
mean genotype yield within the three trials of
medium-yield potential (average correlation coefficient of both stages, r = 0.65 0.09) than within the
low (r = 0.41 0.13) and high (r = 0.42 0.17)
productivity trials (Fig. 5). On the other hand, the
yield of the low productivity trial was best predicted
by models constructed with anthesis spectra, whereas
in the high productivity trial the best predictions were
attained using milk-grain models.
The broad-sense heritability (H2) of grain yield
observed by harvesting was 0.33. The H2 of estimated
GY by PLSR was different across models, ranging
from 0.1 to 0.25 and from 0.01 to 0.20 for milk-grain
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
133
Fig. 4. Relationships between genotype means of measured (harvested) and predicted grain yield using data from the same trials where have
been calibrated. Predicted values are those resulting from the cross-validation procedure: (a) and (b) anthesis models; (c) and (d) milk-grain
models. 6LR, 7LR, models calibrated within the low-yielding, rainfed trial; 6MR, 7MR, models calibrated within the medium-yielding, rainfed
trial; 6MI1, 6MI2, 7MI1, 7MI2, models calibrated within medium-yielding, irrigated trials; 6HI, 7HI, models calibrated within the high-yielding,
irrigated trial.
and anthesis models, respectively (Fig. 6). In spite of
the lack of a clear tendency, H2 showed higher and
more steady values in the models from milk-grain
spectra (0.18 0.06) than in those from anthesis
(0.14 0.08). In addition, within milk-grain model,
the medium productivity environments had higher
values (0.21 0.05) than low and high productivity
environments.
4. Discussion
4.1. Physiological background of the relationship
between reflectance models and GY
Despite being an empirical approach, the models
obtained by PLSR calibration were able to integrate
physiological information from several spectral bands
in order to estimate GY (Fig. 3). Indeed, among the
highest regression coefficients, we found wavelengths
previously included in spectral indices of green
biomass and LAI (near-infrared/red, Peñuelas et al.,
1997a), chlorophyll content (550–680 nm, Haboudane
et al., 2002), water content (970 nm, Peñuelas et al.,
1996) and carotenoids (430–445 nm, Peñuelas et al.,
1995b). All models showed negative coefficients in
the left side of the red edge region (lRE), where
reflectance is reduced when the red/near-infrared
slope increases and shifts towards the right (see
Fig. 2). Filella and Peñuelas (1994) showed that this
slope and its position were closely related to
chlorophyll content, biomass and water status. The
positive coefficients given by some models (6LR, 6HI,
7LR, 7HI) in the near-infrared region (750–950 nm),
associated with the relative content of brown pigments
(Peñuelas and Filella, 1998), might also be related
with green biomass and water status. In addition,
variability in water status was considered in all the
134
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Fig. 6. Broad-sense heritabilities (calculated from data of all trials)
for observed grain yield and predicted grain yield using PLSR
models developed either from anthesis or milk-grain spectra. LR,
models calibrated within the low-yielding, rainfed trial; MR, models
calibrated within the medium-yielding, rainfed trial; MI1, MI2,
models calibrated within medium-yielding, irrigated trials; HI,
models calibrated within the high-yielding, irrigated trial.
Fig. 5. Correlation coefficients between genotype means of measured and predicted grain yield for the models developed in each trial
when applied to all the trials assayed, plotted against average yield
of the application trial: (a) and (b) stand for models calibrated from
anthesis and milk-grain spectra, respectively. 6LR, 7LR, models
calibrated within the low-yielding, rainfed trial; 6MR, 7MR, models
calibrated within the medium-yielding, rainfed trial; 6MI1, 6MI2,
7MI1, 7MI2, models calibrated within medium-yielding, irrigated
trials; 6HI, 7HI, models calibrated within the high-yielding, irrigated trial.
models, by including negative coefficients around a
secondary peak of water absorption (970 nm). The
depletion of reflectance in this region has proven to be
a good indicator of plant water content at the canopy
level (Peñuelas et al., 1997b).
Other wavelengths accounted by the models were
directly associated with chlorophyll absorption (450,
550–670 nm). Under low chlorophyll content in the
canopy (e.g. due to drought or crop senescence)
sensitivity is greater in absorption peaks (450,
670 nm). In contrast, for canopies with greater
chlorophyll content, these absorption peaks become
saturated, and the most sensitive spectral bands are
placed around 550 nm (Peñuelas and Filella, 1998).
Fig. 3 shows that at anthesis the regression coefficients
at the maximum absorption peaks of chlorophyll shift
from negative values at 6LR to values close to zero at
6MR and positive at 6HI, while the opposite trend was
found for the 550 nm region. Given that the estimation
of canopy chlorophyll content by crop reflectance
depends on the product of green biomass and
chlorophyll concentration at the leaf level (Filella
et al., 1995), these results indicates that total
photosynthetic capacity at anthesis increased from
6LR to 6MR and 6HI, likely saturating the spectra at
the trial showing the highest biomass. On the other
hand, from Fig. 3 it may be inferred that at trial MR
senescence increased from anthesis to milk-grain
stage (6MR and 7MR). At milk-grain stage negative
values of the regression coefficients appeared around
670 nm, whereas they were positive in the blue
domain (400–500 nm) at 7LR (see Fig. 3). In the blue
region, both chlorophylls and carotenoids have high
absorbances (Peñuelas and Filella, 1998). Provided
negative coefficients in the region where only
chlorophylls absorb (500–700 nm), the positive
coefficients in the blue region might account for
variations in the ratio between carotenoids and
chlorophylls. This ratio is higher under stress and in
senescing leaves, and decreases with higher nutrient
availability (Filella et al., 1995). Thus, the pattern
followed by the regression coefficients at milk-grain
stage indicates that at 7LR the crop was far more
senescent than at 7MR and 7HI.
In summary, according to the regression coefficients for the different spectral regions, our empirical
models to estimate grain yield might account for three
major constraints in GY: (1) photosynthetic size of the
canopy (e.g. green biomass), through the nearinfrared/red ratios, (2) water status, through the
reflectance around 970 nm, and (3) nitrogen status,
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
as related with chlorophyll content and carotenoids to
chlorophylls ratio.
4.2. Impact of biomass and crop senescence on the
relationship between spectra and yield
The values of the r2 of the calibration and
validation models indicated that the amount of grain
yield variability within a trial (across plots) explained
by the models was greater in low and medium-yield
environments with spectra captured either at anthesis
or milk-grain stage (see Table 2). This result could be
explained by the differences between trials in total and
green biomass when spectra were captured, as
confirmed by Fig. 3. Thus, the large biomass on HI
trial probably saturated the spectra at red and infrared
wavelengths, giving poor predictive assessment of
yield both at anthesis or milk-grain. On the other hand,
these results could be partially attributed also to the
recorded large variability of yield between blocks in
the poorest environments, which on increasing the
range of data, improved the predictive value of the
models at the plot level.
Moreover, a trial growth stage interaction
appeared for the determination coefficients of calibration models, given that r2 was higher at low
productivity than at moderate productivity environments at anthesis, the opposite being true at milk-grain
stage. This result may be explained in terms of crop
senescence, since at anthesis biomass at MR and MI
were higher than at LR (see Section 4.1), probably
saturating the spectra at some wavelengths, and thus
the best appraisal of crop production was obtained at
LR. Contrarily, at milk-grain stage, at MR and MI the
crop had started to senesce, but still had a large
photosynthetic capacity, able to contribute to grain
yield. In contrast, at LR senescence was much
advanced, limiting the ability of canopy reflectance
to properly track changes in productivity.
4.3. Model usefulness for selection purposes
In general, when the relationships between
predicted and observed grain yield were studied using
mean values of genotypes for each trial, the best
relationships were found in medium to high-yielding
trials. On them, genotypes maximized their divergences in yield, giving more chance to the models to
135
detect genotype differences (see Figs. 4 and 5).
Nevertheless, even within the same trial used for
calibration, RMSE values were far greater than the
standard error of the data, and this error increased
significantly when applying the models to other trials.
The main consequence of such results is that spectral
models of grain yield did not provide an accurate
quantification of grain yield values. However, it should
be taken into account that when screening within a
large number of genotypes, breeders are more
interested on the ranking of their yields, than in the
accurate quantification of yield values. Provided the
strongly significant relationship found between measured and modeled grain yield, reflectance models of
grain yield could still be useful for breeding purposes.
Indeed, the relationships between grain yield and our
reflectance models were generally stronger than those
found for previously suggested selection tools for
yield improvement (Araus et al., 1998; Ferrio et al.,
2004; Royo et al., 2002).
4.4. PLS models versus vegetation indices
Despite requiring empirical calibration, our models
showed generally stronger and more robust relationships with grain yield than any of the previously
assayed spectral indices (Aparicio et al., 2000; Royo
et al., 2003). Indeed, both milk-grain and anthesis
models attained high and steady performances in
discriminating between genotype means (average
r = 0.66 0.1 and r = 0.64 0.1, respectively).
Aparicio et al. (2000) found similarly strong relationships between durum wheat yield and SR (r = 0.63),
and NDVI (r = 0.60), but only when measured at
anthesis for experiments under rainfed conditions,
whereas for irrigated ones the best correlations were
attained at maturity and were lower than those of
rainfed environments (r = 0.55). In a similar way,
Royo et al. (2003) found that the usefulness of
reflectance indices for yield prediction were highly
environmental-dependent. The most useful indices
were R680, WI and SR, which had coefficients of
correlation with yield of 0.11 to 0.68, 0.19 to 0.57,
and 0.05–0.59, respectively, that are lower and
unsteady than those found in this work. For our best
model, correlation coefficients between genotype
means of predicted and measured GY within each
of the five trials ranged from 0.53 to 0.76.
136
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
4.5. Best environments and growth stage for
measurements
The best performance of models for predicting the
grain yield of durum wheat grown in environments
with medium and high productivity was in concordance with that pointed out by Royo et al. (2003). They
found that the ability of some spectral reflectance
indices (i.e. water index) to predict yield of durum
wheat was higher in locations where genotypes
expressed their potential. In our results, the models
built with spectra of milk-grain ranked adequately the
yield of genotypes, not just from medium productivity,
but also from high productivity environments. This
could be useful for breeding programs, as selection for
yield potential is usually carried out at locations from
medium to high productivity that maximize the
heritability of target traits.
Regarding the optimal crop stage for measurements, the models developed from either milk-grain or
anthesis spectra showed similar responses as those
reported for some spectral indices. The yield of the
low productivity environment could be better ranked
by measures taken at anthesis, which was also found
by Aparicio et al. (2000) using the SR and NDVI
indices. In contrast, within the medium to highyielding trials, yield was better assessed by milk-grain
models. This agrees with previous works showing
better performance of vegetation indices when
measured during grain filling in mid-high-yielding
environments (Aparicio et al., 2000; Royo et al.,
2003). Whereas under low-watered environments
photosynthetic size of the canopy is already declining
at anthesis, it can still remain or even increase under
well-watered environments (Royo et al., 2004).
Therefore, reflectance spectra measured after anthesis
in the low-yielding environment did not provide
additional information, but added extra noise from
senescent leaves, whereas in the high-yielding
environment, measures taken at anthesis failed to
estimate the overall photosynthetic potential of the
canopy, probably due to the saturation of spectra for
leaf area index (LAI) values higher than 3 (Sellers,
1987). While Royo et al. (2003) using spectral indices
(e.g. R680, WI, SR) established that milk-grain was
the best stage for durum wheat yield appraisal, our
results indicated that both anthesis and milk-grain
spectra could be useful for yield assessment, depend-
ing on the productivity of the environment. However,
the fact that the higher heritabilities were obtained at
milk-grain, indicates that this is the best stage for yield
assessment in locations from medium to high
productivity, which are the most used for breeding
programs in Mediterranean environments (Ceccarelli,
1989).
5. Concluding remarks
From these results, we can conclude that our
empirical models, although not being accurate enough
to quantify grain yield, could still provide a qualitative
assessment of yield differences among genotypes. In
general, the models obtained were robust enough to
rank genotypes by their yield, even when applied to
environments different from those used for calibration. Thus, our approach might be useful at the early
generations of breeding programs, when yield trials
are less feasible, in order to discard poor-yielding
genotypes. The only limitation in this case would be
the plot size needed for capturing the spectra. By
integrating in the same model the miscellaneous
information provided by several spectral regions,
PLSR models developed from VIS/NIR spectra
showed generally stronger and more robust assessments of grain yield than previously assayed spectral
indices. Our results suggest that the most reliable
ranking of genotypes can be attained within medium
to high productivity environments. We also found that,
in such environments, the most recommended stage
for measurements was milk-grain. This technique
could be especially useful for breeding purposes, as
selection for yield is mostly performed on medium–
high-yielding trials. Nevertheless, these models
should be tested on a wider range of environments
and genotypes in order to further assess their
robustness.
Acknowledgments
This study was supported in part by the CICYT
(Spain) research projects AGF96-1137-C02-01 and
AGL-2002-04285. The skilled technical assistance of
the staff of the Àrea de Conreus Extensius is gratefully
acknowledged. We thank Dr. J. Puy, from the
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Departament de Quı́mica of Universitat de Lleida, for
his useful advises on multivariate calibration. We also
thank J. Casadesús, from the Camps Experimentals of
137
Universitat de Barcelona, for his technical tips on
spectroradiometry. J.P. Ferrio is a recipient of a Ph.D.
fellowship from the Generalitat de Catalunya.
Appendix A
Regression coefficients for each wavelength resulting from the models described in the paper. The coefficients
should be applied to SNV reflectance spectra
Wavelength
Anthesis models
6LR
398.760
400.195
401.630
403.064
404.499
405.934
407.369
408.804
410.238
411.673
413.108
414.543
415.978
417.412
418.847
420.282
421.717
423.152
424.586
426.021
427.456
428.891
430.326
431.760
433.195
434.630
436.065
437.500
438.934
440.369
441.804
443.239
444.674
446.108
447.543
300
372
458
393
465
406
385
513
467
530
465
480
493
509
534
436
432
488
480
438
418
474
493
474
469
445
501
434
448
459
391
464
606
514
496
6MR
40
47
40
36
39
42
39
29
31
31
33
34
29
29
32
31
26
24
23
27
28
23
21
22
18
20
17
19
16
18
19
18
16
18
19
Milk-grain models
6MI1
128
143
149
139
124
134
128
148
139
145
137
127
128
121
117
111
102
82
70
72
60
37
57
168
160
180
179
197
208
231
216
242
265
260
292
6MI2
40
43
23
65
22
67
55
16
34
39
49
44
22
50
36
45
40
36
26
29
54
29
34
31
33
45
16
26
28
35
16
17
14
16
24
6HI
38
37
37
38
35
35
36
33
34
36
33
33
32
31
32
31
30
30
30
28
28
28
27
27
26
25
25
26
24
24
23
24
23
23
23
7LR
42
41
40
40
41
39
40
38
39
39
38
38
37
37
37
37
36
35
36
34
34
33
32
33
32
32
30
30
30
30
29
28
28
27
27
7MR
130
124
124
149
133
148
152
150
162
175
149
150
182
196
189
186
199
202
211
203
213
241
216
225
221
229
249
260
247
250
272
260
265
273
278
7MI1
81
81
80
79
79
78
77
77
76
75
74
74
73
72
71
70
69
69
68
66
66
65
64
63
63
62
62
60
60
59
58
57
57
56
55
7MI2
173
169
198
171
185
148
94
105
142
117
88
80
62
83
43
30
21
37
23
28
28
40
46
47
31
21
38
28
11
10
3
0
9
27
13
7HI
39
38
38
37
38
38
38
37
36
36
35
34
34
33
32
31
31
30
29
29
28
28
27
27
26
26
25
25
25
24
24
23
22
22
22
138
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Appendix A (Continued )
Wavelength
Anthesis models
6LR
448.978
450.413
451.848
453.282
454.717
456.152
457.587
459.022
460.456
461.891
463.326
464.761
466.196
467.630
469.065
470.500
471.935
473.370
474.804
476.239
477.674
479.109
480.544
481.978
483.413
484.848
486.283
487.718
489.152
490.587
492.022
493.457
494.892
496.326
497.761
499.196
500.631
502.066
503.500
504.935
506.370
507.805
505
516
519
483
493
473
446
455
442
447
482
436
459
449
487
501
492
501
494
529
554
527
556
580
580
575
603
588
618
620
614
631
601
648
626
623
670
708
691
683
642
695
6MR
17
16
17
17
18
16
17
18
19
19
18
18
19
19
18
18
19
19
20
19
19
20
19
22
20
21
22
22
23
23
24
24
25
24
23
22
23
25
25
23
23
23
Milk-grain models
6MI1
283
290
283
285
287
292
287
290
296
290
274
263
253
255
251
261
254
255
250
253
255
244
239
235
185
190
204
233
220
224
229
253
264
234
238
248
250
248
282
294
327
330
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
10
18
13
12
20
9
15
6
18
12
11
11
5
18
5
4
6
9
31
1
15
1
8
18
4
11
8
23
7
2
6
8
15
7
1
7
3
7
9
4
6
6
23
22
22
22
22
22
21
20
21
21
21
21
20
20
20
20
21
20
19
19
20
19
19
19
19
19
19
19
18
18
18
18
17
17
17
17
15
15
14
14
13
12
26
25
26
25
24
24
24
24
23
23
22
22
22
22
21
21
21
21
20
19
19
19
18
18
17
18
16
17
16
15
15
14
13
12
13
12
11
10
10
10
9
8
271
272
277
270
273
276
274
262
276
270
266
271
259
272
265
260
264
263
261
267
275
257
261
262
270
269
257
245
241
245
235
236
227
232
230
222
217
194
200
184
179
173
54
53
52
51
51
50
49
49
48
47
46
45
45
44
44
43
42
41
41
40
39
39
38
38
37
36
35
35
34
34
32
32
31
30
29
28
27
27
25
24
23
22
30
22
20
23
31
20
23
39
18
22
31
32
37
43
20
44
36
39
54
49
54
72
67
60
68
71
63
63
50
43
57
53
53
39
63
58
65
50
63
69
74
68
21
21
21
21
20
20
19
19
19
19
19
19
19
19
19
18
18
18
18
18
17
17
17
17
16
16
16
16
15
15
15
15
14
14
13
13
13
12
11
10
9
8
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
139
Appendix A (Continued )
Wavelength
Anthesis models
6LR
509.240
510.674
512.109
513.544
514.979
516.414
517.848
519.283
520.718
522.153
523.588
525.022
526.457
527.892
529.327
530.762
532.196
533.631
535.066
536.501
537.936
539.370
540.805
542.240
543.675
545.110
546.544
547.979
549.414
550.849
552.284
553.718
555.153
556.588
558.023
559.458
560.892
562.327
563.762
565.197
566.632
568.066
712
643
653
654
689
496
705
586
567
513
471
393
355
345
327
222
218
123
136
105
29
16
24
31
100
166
145
143
177
213
237
311
315
304
352
366
325
318
275
274
256
237
6MR
23
21
18
18
17
15
12
9
7
5
0
2
6
7
8
13
14
15
17
17
20
22
22
24
25
26
26
28
28
27
28
30
27
26
25
24
23
20
20
17
14
13
Milk-grain models
6MI1
328
349
368
368
368
387
447
497
530
553
585
619
669
716
757
791
818
846
880
916
936
947
966
1001
1037
1064
1066
1071
1103
1102
1103
1088
1062
1041
1009
983
965
930
899
851
777
731
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
9
21
19
16
11
29
37
27
41
29
38
49
51
55
48
58
58
66
68
62
77
66
77
65
66
60
58
56
43
51
50
51
46
40
51
44
43
39
34
47
40
39
11
10
9
8
5
5
3
1
1
3
5
5
8
9
11
12
14
15
16
17
17
18
19
20
20
21
22
22
23
23
25
24
25
24
24
24
24
23
24
23
23
21
7
6
6
6
4
3
3
3
1
0
0
1
3
3
4
4
6
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
23
25
26
26
27
28
28
159
155
147
126
118
110
97
76
68
79
67
57
53
49
52
59
44
45
51
48
56
48
33
32
42
35
38
19
31
19
27
12
26
23
31
20
7
15
16
14
11
9
21
19
17
16
15
13
12
10
9
7
6
4
3
1
0
1
3
5
6
8
9
10
12
13
14
16
18
19
20
21
23
24
26
28
29
30
31
33
34
35
37
38
82
87
78
81
82
70
81
78
72
69
83
76
75
76
65
81
75
89
77
87
94
98
87
84
79
80
98
95
85
74
81
69
64
65
63
58
49
50
47
53
61
44
7
5
4
3
1
1
3
4
6
7
9
11
12
13
14
16
17
19
20
21
22
23
24
25
26
27
28
28
29
30
30
31
31
32
32
31
32
32
32
32
31
31
140
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Appendix A (Continued )
Wavelength
Anthesis models
6LR
569.501
570.936
572.371
573.806
575.240
576.675
578.110
579.545
580.980
582.414
583.849
585.284
586.719
588.154
589.588
591.023
592.458
593.893
595.328
596.762
598.197
599.632
601.067
602.502
603.936
605.371
606.806
608.241
609.676
611.110
612.545
613.980
615.415
616.850
618.284
619.719
621.154
622.589
624.024
625.458
626.893
628.328
629.763
195
171
253
166
151
105
136
160
143
120
164
88
84
10
71
57
94
121
190
172
182
230
249
184
263
179
199
210
194
186
187
210
209
249
301
311
222
227
353
99
298
328
355
6MR
11
7
5
4
1
3
5
6
6
9
11
12
13
13
15
16
16
15
17
18
17
16
18
19
19
20
21
22
25
25
25
26
29
30
31
31
29
33
33
33
30
33
34
Milk-grain models
6MI1
690
635
576
534
475
435
383
339
315
282
250
179
131
84
38
45
13
15
30
48
80
100
121
125
125
158
179
185
186
210
227
231
249
242
265
292
351
326
329
366
387
359
357
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
27
32
30
35
35
33
33
34
35
29
27
44
54
55
47
45
53
42
47
41
47
55
46
43
46
52
45
57
49
51
57
52
51
46
47
52
44
55
54
52
55
62
52
21
20
19
18
18
17
16
16
16
15
15
15
15
15
14
14
14
13
14
14
12
14
14
13
13
13
13
12
12
10
10
10
10
9
8
9
8
8
8
9
7
8
7
29
29
30
31
31
32
33
33
34
34
35
35
36
37
37
37
37
38
38
39
38
39
40
40
39
40
41
40
40
40
41
41
41
41
41
41
41
41
42
42
42
42
43
10
8
4
4
6
1
1
0
7
4
2
2
1
1
5
6
0
2
4
1
2
0
0
5
7
2
8
15
20
24
24
43
33
27
35
34
30
26
16
18
23
21
3
39
40
41
42
43
44
45
46
47
47
48
49
50
51
52
53
54
55
56
56
57
58
59
59
60
61
61
61
62
62
62
63
63
64
64
64
65
66
66
67
68
69
69
40
36
42
33
21
18
22
25
7
5
7
6
7
6
3
2
1
9
5
4
11
4
1
11
14
10
6
5
8
2
11
36
26
32
37
48
48
33
25
1
23
13
8
31
31
30
30
29
29
29
28
28
28
28
28
27
27
27
28
27
27
27
28
28
27
28
28
28
28
27
27
26
26
25
25
25
24
24
24
24
24
24
24
23
24
24
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
141
Appendix A (Continued )
Wavelength
Anthesis models
6LR
631.198
632.632
634.067
635.502
636.937
638.372
639.806
641.241
642.676
644.111
645.546
646.980
648.415
649.850
651.285
652.720
654.154
655.589
657.024
658.459
659.894
661.328
662.763
664.198
665.633
667.068
668.502
669.937
671.372
672.807
674.242
675.676
677.111
678.546
679.981
681.416
682.850
684.285
685.720
687.155
688.590
690.024
691.459
352
387
389
447
471
528
599
672
717
767
804
873
873
887
812
768
687
674
610
554
517
468
439
366
359
331
333
314
207
220
177
168
102
65
79
51
50
111
109
180
159
128
245
6MR
33
31
33
33
35
35
37
38
43
42
46
43
47
48
49
48
49
47
53
51
53
56
55
57
57
59
58
60
63
60
60
58
62
60
57
60
58
56
54
50
45
41
32
Milk-grain models
6MI1
381
413
374
421
437
441
472
492
515
522
567
568
593
600
619
605
595
601
584
574
576
582
587
600
597
585
587
581
602
582
587
598
592
584
594
576
583
569
547
538
524
514
483
6MI2
6HI
7LR
7MR
43
43
45
45
36
24
23
25
26
10
24
22
13
25
2
26
28
30
30
10
8
18
21
17
7
22
18
20
9
4
25
24
23
16
24
35
37
33
24
61
56
41
67
7
8
7
7
7
7
7
6
6
5
4
4
3
4
3
3
3
3
1
1
0
0
0
2
1
2
3
3
4
3
2
3
2
4
3
2
1
1
1
3
6
7
11
43
43
43
43
44
44
44
43
44
45
45
45
44
45
45
45
45
45
46
45
45
45
45
45
44
44
45
45
45
46
44
45
46
45
45
46
46
47
48
48
48
50
52
15
19
22
2
2
15
7
17
25
34
33
55
50
46
42
44
50
75
69
58
51
56
66
84
80
78
86
104
113
100
102
108
104
114
109
96
86
85
48
26
1
32
53
7MI1
70
70
71
71
71
72
72
72
72
73
72
73
73
73
73
74
74
75
74
73
73
72
72
72
71
71
71
71
71
71
71
72
72
73
74
76
78
80
84
88
91
96
100
7MI2
7HI
5
8
3
31
25
15
3
14
26
33
27
6
18
15
20
8
12
22
33
11
12
30
15
21
3
20
16
17
10
23
10
15
8
24
15
6
30
23
35
52
46
58
37
25
24
24
24
24
24
23
22
22
21
21
20
19
19
18
18
18
17
17
16
16
16
15
15
14
14
13
13
12
12
12
12
12
12
13
14
14
16
18
20
23
28
32
142
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Appendix A (Continued )
Wavelength
692.894
694.329
695.764
697.198
698.633
700.068
701.503
702.938
704.372
705.807
707.242
708.677
710.112
711.546
712.981
714.416
715.851
717.286
718.720
720.155
721.590
723.025
724.460
725.894
727.329
728.764
730.199
731.634
733.068
734.503
735.938
737.373
738.808
740.242
741.677
743.112
744.547
745.982
747.416
748.851
750.286
751.721
753.156
Anthesis models
Milk-grain models
6LR
6MR
6MI1
6MI2
6HI
7LR
7MR
187
190
160
162
248
234
134
137
205
12
46
101
245
175
297
368
478
490
627
618
635
808
700
822
871
833
846
789
843
782
775
759
764
645
655
642
620
692
571
544
568
485
452
23
11
2
13
29
46
64
82
92
112
132
147
163
178
198
215
226
245
262
282
292
307
314
328
335
336
341
336
334
325
314
302
285
267
251
231
209
187
169
147
131
107
96
461
408
365
298
237
199
162
108
73
53
43
21
30
22
48
43
25
1050
888
1179
1118
1308
1476
1445
1521
1700
1956
1531
878
495
564
814
954
1031
923
755
698
739
937
1250
1593
1651
1697
73
80
83
89
102
99
125
113
116
102
107
100
91
88
87
70
74
73
65
55
8
5
6
47
71
58
87
117
159
163
151
173
167
194
186
182
165
165
165
147
146
120
95
16
21
26
28
35
41
48
52
57
64
69
74
77
84
90
94
99
104
110
116
119
122
127
131
132
134
132
132
130
125
123
117
111
102
95
87
76
69
58
49
41
33
24
52
54
55
57
58
59
61
63
64
64
65
66
66
67
67
67
67
68
67
66
66
66
65
63
62
60
60
58
55
53
51
50
48
45
43
42
41
38
36
35
34
32
30
75
88
112
111
129
136
126
127
106
88
51
31
19
54
65
105
149
199
258
251
287
312
367
383
399
424
430
450
435
425
413
414
403
355
379
339
319
301
272
257
252
228
193
7MI1
105
110
115
120
125
129
134
138
141
145
148
150
152
154
156
157
158
157
156
154
151
148
143
137
130
122
113
102
91
80
68
54
41
28
14
2
11
23
35
45
55
64
72
7MI2
7HI
58
11
29
11
12
10
31
31
61
57
107
91
137
148
181
224
220
282
324
348
379
393
376
437
420
401
412
394
426
393
333
332
312
285
232
252
206
184
193
149
165
131
111
37
42
48
53
59
65
70
76
80
85
89
94
98
102
106
110
114
118
122
125
127
129
130
130
129
128
125
120
116
110
103
95
88
79
70
61
53
44
36
29
21
15
8
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
143
Appendix A (Continued )
Wavelength
Anthesis models
6LR
754.590
756.025
757.460
758.895
760.330
761.764
763.199
764.634
766.069
767.504
768.938
770.373
771.808
773.243
774.678
776.112
777.547
778.982
780.417
781.852
783.286
784.721
786.156
787.591
789.026
790.460
791.895
793.330
794.765
796.200
797.634
799.069
800.504
801.939
803.374
804.808
806.243
807.678
809.113
810.548
811.982
813.417
814.852
490
540
408
51
369
13
95
65
222
199
251
304
172
221
165
153
127
98
222
48
81
62
8
27
138
30
106
178
53
257
214
219
227
244
232
461
351
405
384
337
423
431
464
6MR
80
65
60
37
7
8
11
12
9
5
9
9
15
13
14
18
16
23
24
19
25
20
26
25
23
26
25
27
23
27
26
27
26
26
28
23
29
24
23
27
25
27
17
Milk-grain models
6MI1
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
1668
814
155
5
40
288
258
47
31
48
64
105
109
106
119
111
99
99
79
82
74
40
24
28
0
0
53
10
17
18
6
7
25
25
14
55
38
10
1
60
131
249
342
111
88
109
100
33
6
6
8
11
9
16
29
31
18
23
31
42
21
52
45
36
27
63
73
67
102
74
96
86
77
100
105
116
102
121
98
104
112
107
144
119
168
174
17
12
8
4
3
5
11
12
17
19
21
23
24
26
26
29
30
29
31
31
32
32
31
31
31
32
32
34
32
35
33
32
33
34
35
34
32
33
34
35
31
32
33
28
28
26
24
25
22
23
21
20
21
19
19
18
17
15
15
14
14
12
12
11
10
7
8
7
7
6
5
3
4
2
0
1
0
0
2
2
4
4
4
5
8
7
188
152
54
80
121
40
35
51
66
43
47
54
47
18
31
37
37
44
18
35
9
5
1
25
6
13
22
29
35
57
76
64
71
64
89
79
81
84
91
100
106
142
163
78
85
90
95
104
108
109
110
110
111
111
111
112
113
112
113
113
113
112
111
111
112
111
110
110
109
109
108
107
107
106
105
105
104
104
103
102
102
101
100
99
99
99
117
84
114
115
125
145
142
120
106
43
12
5
4
53
15
3
4
3
2
9
21
54
24
58
22
41
14
22
9
4
32
16
11
41
22
7
34
33
15
15
10
21
69
3
2
7
11
16
21
21
22
23
24
25
26
27
27
28
28
28
29
28
28
27
27
26
26
26
25
26
26
26
26
25
26
25
26
25
25
25
24
24
23
22
20
18
144
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Appendix A (Continued )
Wavelength
Anthesis models
6LR
816.287
817.722
819.156
820.591
822.026
823.461
824.896
826.330
827.765
829.200
830.635
832.070
833.504
834.939
836.374
837.809
839.244
840.678
842.113
843.548
844.983
846.418
847.852
849.287
850.722
852.157
853.592
855.026
856.461
857.896
859.331
860.766
862.200
863.635
865.070
866.505
867.940
869.374
870.809
872.244
873.679
875.114
876.548
394
513
532
437
486
404
490
534
397
374
475
370
403
407
253
366
375
410
296
228
227
281
148
5
298
436
160
338
350
38
192
67
197
84
20
29
38
60
156
12
29
96
100
6MR
18
26
30
31
15
25
19
22
16
14
15
7
13
6
7
5
3
0
0
2
3
8
7
10
5
10
4
3
5
10
9
9
9
15
16
5
2
12
15
24
7
12
24
Milk-grain models
6MI1
312
284
217
170
232
136
82
139
152
117
91
69
35
10
44
76
64
77
88
100
125
152
145
111
139
167
34
170
297
150
117
147
180
192
111
66
208
210
125
142
168
205
129
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
190
213
165
220
151
187
204
154
183
159
161
154
181
167
140
155
125
149
117
134
139
154
152
112
166
150
106
187
156
139
117
130
135
129
129
101
153
163
147
121
94
134
149
33
30
29
32
33
31
29
29
29
32
28
26
30
28
28
27
26
29
28
26
25
29
28
28
26
30
28
27
30
26
30
28
28
27
25
29
24
28
27
27
28
24
25
8
9
10
11
11
13
13
15
13
14
15
15
17
18
18
19
19
20
20
21
23
24
23
24
26
26
27
27
27
28
28
31
31
33
32
35
32
33
35
33
34
35
39
133
142
105
119
175
102
94
135
169
122
142
102
104
109
82
65
32
27
45
70
20
25
72
25
29
16
50
83
6
24
37
46
86
12
89
68
1
25
36
22
8
20
45
98
97
95
94
94
92
92
90
89
88
85
85
84
83
82
80
80
79
77
77
76
75
74
74
72
72
72
69
69
67
68
67
66
65
66
64
63
61
60
60
59
59
58
73
118
40
44
84
119
59
12
37
27
59
46
19
12
59
12
7
37
36
25
22
76
30
60
72
73
61
101
50
2
50
22
54
58
74
2
4
39
14
3
25
13
66
16
17
18
18
19
16
16
17
16
16
15
16
15
16
15
15
16
14
15
15
13
14
13
14
13
13
14
12
12
13
14
14
13
13
15
13
12
12
11
14
9
10
10
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
145
Appendix A (Continued )
Wavelength
Anthesis models
6LR
877.983
879.418
880.853
882.288
883.722
885.157
886.592
888.027
889.462
890.896
892.331
893.766
895.201
896.636
898.070
899.505
900.940
902.375
903.810
905.244
906.679
908.114
909.549
910.984
912.418
913.853
915.288
916.723
918.158
919.592
921.027
922.462
923.897
925.332
926.766
928.201
929.636
931.071
932.506
933.940
935.375
936.810
136
62
2
134
108
71
124
316
407
224
120
358
222
293
52
149
211
96
195
185
140
644
466
426
139
269
364
320
130
306
101
529
357
177
99
557
343
767
405
208
278
446
6MR
6
19
22
13
25
21
18
7
32
28
30
20
28
35
29
25
17
20
26
12
32
30
32
19
26
22
34
47
24
24
15
42
78
25
51
34
63
59
34
18
30
31
Milk-grain models
6MI1
133
122
148
170
184
165
119
190
224
84
84
22
52
158
260
373
333
207
151
32
144
291
252
263
255
339
272
281
225
178
1
30
5
101
26
86
337
528
966
1115
1157
1175
6MI2
6HI
7LR
7MR
7MI1
7MI2
7HI
191
111
141
158
98
122
85
163
118
58
109
66
199
99
104
167
182
247
153
175
179
149
141
146
231
125
203
120
108
182
114
176
12
63
136
173
151
18
100
112
206
241
27
27
25
27
23
20
26
24
25
20
23
17
24
19
21
16
13
22
14
13
17
20
16
17
8
16
24
8
2
14
8
15
6
8
14
4
8
2
13
10
25
7
39
37
37
39
40
40
40
40
38
39
41
42
41
41
46
40
42
45
44
44
45
41
46
43
43
40
39
41
34
42
48
46
44
38
45
42
43
34
43
45
25
22
4
52
17
47
27
8
25
37
5
20
31
28
81
54
46
54
111
23
102
141
38
102
106
41
104
33
105
84
34
31
10
31
102
159
113
14
22
23
262
636
765
964
56
57
54
54
53
52
51
50
49
46
47
46
45
44
42
43
41
40
37
37
36
36
36
31
31
31
29
29
25
25
23
22
20
19
19
14
14
10
10
9
2
10
24
130
139
57
70
167
50
5
4
20
95
38
35
81
86
23
51
3
55
217
192
259
209
392
249
288
247
138
227
214
16
32
65
76
58
109
125
94
132
66
227
384
9
9
8
8
8
8
6
7
7
8
8
6
5
4
2
3
1
1
1
0
0
1
3
4
1
6
3
4
4
6
6
3
3
1
5
5
6
7
0
9
5
3
146
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Appendix A (Continued )
Wavelength
938.245
939.680
941.114
942.549
943.984
945.419
946.854
948.288
949.723
951.158
952.593
954.028
955.462
956.897
958.332
959.767
961.202
962.636
964.071
965.506
966.941
968.376
969.810
971.245
972.680
974.115
975.550
976.984
978.419
979.854
981.289
982.724
984.158
985.593
987.028
988.463
989.898
991.332
992.767
994.202
995.637
997.072
Anthesis models
Milk-grain models
6LR
6MR
6MI1
6MI2
6HI
7LR
7MR
1530
1027
247
752
195
1309
1270
516
526
969
697
321
133
399
1023
375
696
490
339
727
640
523
453
417
163
932
1098
578
125
965
508
708
461
798
582
699
1001
445
1297
1215
367
326
95
36
10
5
49
23
25
31
58
13
12
41
9
106
75
78
56
99
114
120
135
122
89
101
166
158
137
215
192
136
133
108
157
174
164
115
127
171
109
172
83
147
1277
850
571
417
771
789
556
500
500
734
390
325
183
209
67
40
455
126
122
384
703
757
648
905
785
918
980
914
704
1054
1023
883
937
992
1103
1207
1182
1309
1018
1267
1035
924
210
470
348
462
88
195
220
268
309
132
338
3
132
106
272
128
106
238
200
116
435
257
143
411
150
380
155
146
577
221
316
19
130
298
350
423
31
411
383
562
344
97
32
6
3
3
30
4
0
1
20
26
23
41
51
46
28
36
57
67
31
77
61
65
65
61
50
77
57
63
60
62
47
45
76
56
70
52
52
70
55
51
34
80
33
25
30
28
41
32
33
32
5
28
20
12
10
5
7
9
10
1
2
3
8
12
11
13
6
3
16
9
2
30
0
22
29
11
27
0
9
5
14
2
36
18
593
293
486
227
34
278
794
667
372
56
538
558
490
280
118
38
274
223
416
439
411
447
650
758
1018
643
785
641
558
608
636
368
542
874
456
823
710
712
1219
563
640
722
7MI1
14
17
15
23
24
28
23
34
36
44
50
49
55
66
76
74
78
80
87
95
86
93
100
101
94
93
94
104
106
98
101
109
104
108
104
104
105
113
105
101
103
103
7MI2
7HI
68
307
183
359
832
4
404
11
153
211
180
66
497
141
118
127
16
267
88
349
417
508
465
327
1007
909
759
210
632
1283
347
191
23
193
165
17
141
184
740
225
28
504
10
10
13
17
21
11
12
1
2
3
10
19
15
7
22
23
29
33
38
37
50
46
40
46
45
48
51
53
50
44
38
38
38
31
42
35
43
44
41
42
40
35
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
147
Appendix A (Continued )
Wavelength
998.506
999.941
1001.380
Anthesis models
Milk-grain models
6LR
6MR
6MI1
6MI2
6HI
7LR
7MR
72
1485
872
154
107
55
1065
1280
1199
683
237
459
45
54
35
6
23
4
801
882
575
References
Aparicio, N., Villegas, D., Araus, J.L., Casadesús, J., Royo, C.,
2002. Relationship between growth traits and spectral vegetation
indices in durum wheat. Crop Sci. 42, 1547–1555.
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.
Aparicio, N., Villegas, D., Royo, C., Casadesús, J., Araus, J.L.,
2004. Effect of sensor view angle on the assessment of agronomic traits by ground level hyper-spectral reflectance measurements in durum wheat under contrasting Mediterranean
conditions. Int. J. Remote Sens. 25, 1131–1152.
Araus, J.L., Amaro, T., Casadesús, J., Asbati, A., Nachit, M.M.,
1998. Relationships between ash content, carbon isotope discrimination and yield in durum wheat. Aust. J. Plant Physiol. 25,
835–842.
Barnes, R.J., Dhanoa, M.S., Lister, S.J., 1989. Standard normal
variate transformation and de-trending of near-infrared diffuse
reflectance spectra. Appl. Spectrosc. 43, 772–777.
Bauer, M.E., 1975. The role of remote sensing in determining
the distribution and yield of crops. Adv. Agron. 27, 271–
304.
Beebe, K.R., Kowalski, B.R., 1987. An introduction to multivariate
calibration and analysis. Anal. Chem. 59, 1007A–1017A.
Bort, J., Casadesús, J., Araus, J.L., Grando, S., Ceccarelli, S.,
2002. Spectral vegetation indices as nondestructive indicators
of barley yield in Mediterranean rain-fed conditions. In: Slafer,
G.A., Molina-Cano, J.L., Savin, R., Araus, J.L., Romagosa,
I. (Eds.), Barley Science: Recent Advances from Molecular
Biology to Agronomy of Yield and Quality. Haworth Press,
Binghamton, NH, pp. 387–412.
Ceccarelli, S., 1989. Wide adaptation: how wide? Euphytica 40,
197–205.
Ferrio, J.P., Bertran, E., Nachit, M.M., Català, J., Araus, J.L., 2004.
Estimation of grain yield by near-infrared reflectance spectroscopy in durum wheat. Euphytica 137, 373–380.
Field, C.B., Gamon, J.A., Peñuelas, J., 1994. Remote sensing of
terrestrial photosynthesis. In: Schulze, D.C., Caldwell, M.M.
(Eds.), Ecophysiology of Photosynthesis. Springer-Verlag, Berlin, pp. 511–528.
Filella, I., Peñuelas, J., 1994. The red edge position and shape as
indicators of plant chlorophyll content, biomass and hydric
status. Int. J. Remote Sens. 15, 1459–1470.
7MI1
114
106
108
7MI2
7HI
705
752
115
31
45
50
Filella, I., Serrano, L., Serra, J., Peñuelas, J., 1995. Evaluating wheat
nitrogen status with canopy reflectance indices and discriminant
analysis. Crop Sci. 35, 1400–1405.
Gamon, J.A., Field, C.B., Goulden, M.L., Griffin, K.L., 1995.
Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecol. Appl. 5, 28–
41.
Gamon, J.A., Serrano, L., Surfus, J.S., 1997. The photochemical
reflectance index: an optical indicator of photosynthetic radiation use efficiency across species, functional types, and nutrient
levels. Oecologia 112, 492–501.
Haboudane, D., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J.,
Dextraze, L., 2002. Integrated narrow-band vegetation
indices for prediction of crop chlorophyll content for application to precision agriculture. Remote Sens. Environ. 81, 416–
426.
Martens, H., Naes, T., 1991. Multivariate Calibration. Wiley, New
York.
Peñuelas, J., Baret, F., Filella, I., 1995a. Semi-empirical indices to
assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31, 221–230.
Peñuelas, J., Filella, I., Lloret, P., Muñoz, F., Vilajeliu, M., 1995b.
Reflectance assessment of plant mite attack on apple trees. Int. J.
Remote Sens. 14, 1887–1905.
Peñuelas, J., Filella, I., 1998. Visible and near-infrared reflectance
techniques for diagnosing plant physiological status. Trends
Plant Sci. 3, 151–156.
Peñuelas, J., Filella, I., Serrano, L., 1996. Cell wall elasticity and
water index (R970 nm/R990 nm) in wheat under different nitrogen availabilities. Int. J. Remote Sens. 17, 373–382.
Peñuelas, J., Isla, R., Filella, I., Araus, J.L., 1997a. Visible and nearinfrared reflectance assessment of salinity effects on barley.
Crop Sci. 37, 198–202.
Peñuelas, J., Piñol, J., Ogaya, R., Filella, I., 1997b. Estimation of
plant water concentration by the reflectance water index WI
(R900/R970). Int. J. Remote Sens. 18, 2869–2875.
Raun, W.R., Solie, J.B., Johnson, G.V., Stone, M.L., Lukina, E.V.,
Thomason, W.E., Schepers, J.S., 2001. In-season prediction of
potential grain yield in winter wheat using canopy reflectance.
Agron. J. 93, 131–138.
Richardson, A.D., Berlyn, G.P., Gregoire, T.G., 2001. Spectral
reflectance of Picea rubens (Pinaceae) and Abies balsamea
(Pinaceae) needles along an elevational gradient, Mt. Moosilauke, New Hampshire, USA. Am. J. Bot. 88, 667–676.
148
J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148
Royo, C., Aparicio, N., Blanco, R., Villegas, D., 2004. Leaf and
green area development of durum wheat genotypes grown under
Mediterranean conditions. Eur. J. Agron. 20, 419–430.
Royo, C., Aparicio, N., Villegas, D., Casadesús, J., Monneveux, P.,
Araus, J.L., 2003. Usefulness of spectral reflectance indices as
durum wheat yield predictors under contrasting Mediterranean
environments. Int. J. Remote Sens. 24, 4403–4419.
Royo, C., Villegas, D., Garcı́a del Moral, L.F., El Hani, S., Aparicio,
N., Rharrabti, Y., Araus, J.L., 2002. Comparative performance of
carbon isotope discrimination and canopy temperature depression as predictors of genotype differences in durum wheat yield
in Spain. Aust. J. Agric. Res. 53, 561–569.
SAS Institute Inc., 1987. SAS/STATTM Guide for Personal
Computer, 6th ed. SAS, Cary, NC.
Sellers, P.J., 1987. Canopy reflectance, photosynthesis, and transpiration. II. The role of biophysics in the linearity of their
interdependence. Remote Sens. Environ. 21, 143–183.
Serrano, L., Filella, I., Peñuelas, J., 2000. Remote sensing of
biomass and yield of winter wheat under different nitrogen
supplies. Crop Sci. 40, 723–731.
Vaesen, K., Gilliams, S., Nackaerts, K., Coppin, P., 2001. Groundmeasured spectral signatures as indicators of ground cover and
leaf area index: the case of paddy rice. Field Crops Res. 69, 13–
25.
Walburg, G., Bauer, M.E., Daughtry, C.S.T., Housley, T.L., 1982.
Effects of nitrogen on the growth, yield and reflectance characteristics of corn. Agron. J. 74, 677–683.
Wold, S., 1978. Cross validatory estimation of the number of
components in factor and principal components models. Technometrics 20, 397–406.
Zadoks, J.C., Chang, T.T., Konzak, C.F., 1974. A decimal
code for the growth stage of cereals. Weed Res. 14, 415–
421.