Assessment of durum wheat yield using visible and near-infrared reflectance spectra of canopies

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 Produccio´ Vegetal i Ciencia Forestal, Universitat de Lleida, Rovira Roure 191, 25198 Lleida, Spain b IRTA, A`rea de Conreus Extensius, Centre UdL-IRTA, Rovira Roure 191, 25198 Lleida, Spain c Unitat de Fisiologı´a 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 Pen˜uelas 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 (Pen˜uelas 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 (Pen˜uelas 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 Vitro´n) 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 ha
1) Productivity classification Trial code

El Cano´s

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 m
2. 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 ha
1, 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 ha
1) 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 N
1 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 ha
1, 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 ha
1); 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, Pen˜uelas et al., 1997a), chlorophyll content (550–680 nm, Haboudane et al., 2002), water content (970 nm, Pen˜uelas et al., 1996) and carotenoids (430–445 nm, Pen˜uelas 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 Pen˜uelas (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 (Pen˜uelas 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 (Pen˜uelas 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 (Pen˜uelas 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 (Pen˜uelas 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 ` rea de Conreus Extensius is gratefully the staff of the A acknowledged. We thank Dr. J. Puy, from the

J.P. Ferrio et al. / Field Crops Research 94 (2005) 126–148

Departament de Quı´mica of Universitat de Lleida, for his useful advises on multivariate calibration. We also thank J. Casadesu´s, 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., Casadesu´s, J., Royo, C., 2002. Relationship between growth traits and spectral vegetation indices in durum wheat. Crop Sci. 42, 1547–1555. Aparicio, N., Villegas, D., Casadesu´s, 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., Casadesu´s, 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., Casadesu´s, 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., Casadesu´s, 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., Catala`, 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., Pen˜uelas, 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., Pen˜uelas, 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., Pen˜uelas, 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. Pen˜uelas, J., Baret, F., Filella, I., 1995a. Semi-empirical indices to assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31, 221–230. Pen˜uelas, J., Filella, I., Lloret, P., Mun˜oz, F., Vilajeliu, M., 1995b. Reflectance assessment of plant mite attack on apple trees. Int. J. Remote Sens. 14, 1887–1905. Pen˜uelas, J., Filella, I., 1998. Visible and near-infrared reflectance techniques for diagnosing plant physiological status. Trends Plant Sci. 3, 151–156. Pen˜uelas, 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. Pen˜uelas, J., Isla, R., Filella, I., Araus, J.L., 1997a. Visible and nearinfrared reflectance assessment of salinity effects on barley. Crop Sci. 37, 198–202. Pen˜uelas, J., Pin˜ol, 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., Casadesu´s, 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., Garcı´a 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., Pen˜uelas, 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.