Spectral reflectance indices were developed using formulations based on simple mathematical operations, such as ratios or differences, between the reflectance at given wavelengths. Most spectral indices use specific wavebands in the range 400 to 900 nm and their most widespread application is in the assessment of plant traits related to the photosynthetic size of the canopy, such as LAI and biomass.

The most widespread vegetation indices (VI), for measurements not only at ground level but also at aircraft and satellite level (Wiegand & Richardson, 1990) are the normalized difference vegetation index (NDVI = RniR-RREd /Rnir +RREd) and the simple ratio (SR= RNIR/RRED) (see Table 1 for their definition). The ratio between the reflectances in the near­infrared (NIR) and red (RED) wavelengths is high for dense green vegetation, but low for the soil, thus giving a contrast between the two surfaces. For wheat and barley a wavelength (A) of around 680 nm is the most commonly used for Rred, and one of 900 nm for Rnir (Penuelas et al., 1997a). These indices have been positively correlated with the absorbed photosynthetically active radiation (PAR), the photosynthetic capacity of the canopy and net primary productivity (Sellers, 1987). According to Wiegand & Richardson (1984, as cited in Wiegand et al., 1991), the fraction of the incident radiation used by the crops for photosynthesis (FPAR) may be derived from vegetation indices through their direct relationship with LAI, according to Equation (1):

FPAR(VI) = FPAR(LAI) x LAI(VI) (1)

For this reason, vegetation indices have proven to be useful for estimating the early vigor of wheat genotypes (Bellairs et al., 1996; Elliot & Regan, 1993), monitoring wheat tiller density (J. H. Wu et al., 2011), and assessing green biomass, LAI and the fraction of radiation intercepted in cereal crops (Ahlrichs & Bauer, 1983; Aparicio et al., 2000, 2002; Baret & Guyot, 1991; Elliott & Regan, 1993; Gamon et al., 1995; Penuelas et al., 1993, 1997a; Price & Bausch, 1995; Tucker 1979; Vaesen et al., 2001). They tend to minimize spectral noise caused by the soil background and atmospheric effects (Baret et al., 1992; Collins, 1978; Demetriades-Shah et al., 1990; Filella & Penuelas, 1994; Mauser & Bach, 1995).

Positive and significant correlations of SR and NDVI with LAI (Fig. 6), GAI and biomass (either on a linear or a logarithmic basis) have been reported in bread wheat and barley (Bellairs et al., 1996; Darvishzadeh et al., 2009; Fernandez et al., 1994; Field et al., 1994; Penuelas et al., 1997a). In a study conducted with 25 bread wheat genotypes, NDVI explained around 40% of the variability found in biomass (Reynolds et al., 1999). Studies involving 20-25 durum wheat genotypes have demonstrated a strong association between SR and NDVI and biomass under both rainfed and irrigated field conditions (Aparicio et al., 2000, 2002; Royo et al., 2003). Spectral reflectance measurements are also being used increasingly as a tool to detect the canopy nitrogen status and allow locally adjusted nitrogen fertilizer applications during the growing season (Mistele & Schmidhalter, 2010). Since grain yield is closely associated with crop growth and the vegetation indices are sensitive to canopy variables such as LAI and biomass that largely determine this growth, spectral data have also been proposed as suitable estimators in yield-predicting models (Aparicio et al., 2000; Das et al., 1993; Ma et al., 2001; Royo et al., 2003).

Fig. 6. Patterns of the relationships of leaf area index (LAI) with the normalized difference vegetation index (NDVI) and the simple ratio (SR). Data correspond to 7 field experiments involving 20-25 durum wheat genotypes and conducted under contrasting Mediterranean conditions for 2 years, with spectral reflectance measurements done at anthesis and milk — grain stage. Each point corresponds to the mean value of a genotype, experiment and growth stage. Adapted from Aparicio et al. (2002)

Another way to formulate the relationship between biomass and VI is to use the light use efficiency (є) model (Kumar & Monteith, 1981) based on the fact that the growth rate of a crop canopy is almost proportional to the rate of interception of radiant energy. Thus, the crop dry weight of a crop canopy at a given moment (t) may be expressed as a function of the incident radiation (Io), the fraction of the radiation intercepted by the crop canopy (FPAR), and the radiation use efficiency (є), as follows:

CDW = jlo x FPAR(LAI) x є dt (2)

0

Small increases in biomass in a small period (expressed as days or thermal units) may then be calculated as a function of LAI from the derivative of Equation (2)

= Io x FPAR(LAI) x є (3)

The incident radiation (Io) may be obtained from meteorological stations or, alternatively, it can be estimated from air temperatures (Allen et al., 1998). FPAR(LAI) may be calculated from vegetation indices on the basis of the linear relationship existing between vegetation indices and the FPAR of green canopies (Daughtry et al., 1992), and particularly between NDVI and FPAR (Bastiaansen & Ali, 2003). Radiation use efficiency (є) is assumed to be constant during the crop growing season (Casanova et al., 1998). Values of radiation use efficiency have been summarized by Russell et al. (1989) for different crops and environmental conditions; moreover, є-values can also be derived for a particular species
and environment from the slope of the relationship between total aboveground biomass and absorbed PAR energy (Liu et al., 2004; Serrano et al., 2000).

An example of use of Kumar & Monteith’s model to assess the pattern of changes in biomass from the LAI estimated from spectral reflectance measurements is shown in Fig. 7. In the example, LAI and CDW values were calculated from destructive samplings, and a comparison is made between the pattern of changes in CDW derived from the mathematical model and that assessed by destructive samplings (Fig. 7b). The model requires frequent reflectance measurements to accurately assess the pattern of changes in LAI over time (Christensen & Goudriaan, 1993), and proper estimations of the incident radiation.

Studies conducted in bread wheat (Asrar et al., 1984; Serrano et al., 2000; Wiegand et al., 1992) and durum wheat (Aparicio et al., 2002) have demonstrated that SR increases linearly with increases in LAI, while NDVI shows a curvilinear response (Fig. 6). When the LAI of wheat canopies exceeds a certain level, the addition of more leaf layers to the canopy does not entail great changes in NDVI (Aparicio et al., 2000; Sellers, 1987), because the reflectance of solar radiation from the underlying soil surface or lower leaf layers is largely attenuated when the ground surface is completely obscured by the leaves (Carlson & Ripley, 1997). The consequence is that for LAI values higher than 3, NDVI becomes relatively insensitive to changes in canopy structure (Aparicio et al., 2002; Curran, 1983; Gamon et al., 1995; Serrano et al., 2000; Wiegand et al., 1992), which constitutes an important limitation for the use of NDVI to estimate LAI. In this context the linearity of the relationship between SR and LAI is not advantageous, because SR may be directly derived from NDVI as SR=(1+NDVI)/(1- NDVI), thus leading to similar statistical significances of both indices when LAI values are predicted (J. M. Chen & Cihlar, 1996). Because of the sensitivity of NDVI and SR to external factors —particularly the soil background at low LAI values—and the developments in the field of imaging spectrometry, a set of new vegetation indices have been developed in order to minimize the effect of disturbing elements in the capturing of the spectra (Baret & Guyot, 1991; Broge & Mortensen, 2002; Gilabert et al., 2002; Meza Diaz & Blackburn, 2003; Rondeaux et al., 1996).

In order to compare the suitability of the classical vegetation indices and the new ones mentioned in the literature as being appropriate for estimating growth traits in wheat and other cereals (P. Chen et al., 2009; Haboudane et al., 2004; Li et al., 2010a; Prasad et al., 2007), 83 hyperspectral vegetation indices were tested using durum wheat data from our own research. The indices were calculated from spectral reflectance measurements taken at different growth stages in 7 field experiments each involving 20-25 durum wheat genotypes, conducted under contrasting Mediterranean conditions for 2 years. Principal component analysis performed with the complete set of vegetation indices and LAI, GAI and CDW revealed that the vegetation indices most closely correlated with durum wheat growth indices were the 29 shown in Table 1. The correlation coefficients between growth traits and the selected indices are shown in Fig. 8. The results show that the majority of indices explained more than 50% of variation in LAI, GAI and CDW when determined at anthesis and milk grain stages, most correlation coefficients being statistically significant at P<0.001. However, the correlation coefficients were significant only for a small number of indices when measurements were taken at physiological maturity. From these results we can conclude that despite the large number of vegetation indices described to improve the appraisal of growth indices given by NDVI and SR, this objective was attained in only a few cases.

Fig. 8 shows that some indices changed from positive values determined at milk-grain to negative ones determined at physiological maturity, confirming that the utility of vegetation indices to assess growth traits decreases drastically when the crop starts to senesce (Aparicio et al., 2000). Young wheat plants normally absorb more photosynthetically active radiation and therefore reflect more NIR. As the plants progress in growth stage, new tissues are formed but older green tissues lose chlorophyll concentration, turning chlorotic and then necrotic. These senescent tissues increase reflectance at the visible wavelengths and decrease reflectance at the NIR wavelengths, causing a decrease in the values of the vegetation indices compared with that obtained at earlier growth stages. Aparicio et al. (2002) concluded that genotypic differences were maximized in durum wheat when growth traits were determined by spectral reflectance measurements taken at anthesis and milk-grain stage.

Mistele and Schmidhalter (2010) Xue et al. (2004)

R780/R740 R780/R740 R780/R740 RI Ratio index R810/R560

Though a large number of studies demonstrate the utility of vegetation indices for assessing growth traits in small-grain cereals when there is a wide range of variability involved in the experimental data, the results indicate that the value of the indices decreases drastically when the range of variation caused by the environment or the crop canopies is low (Aparicio et al., 2002; Royo et al., 2003). In such cases the success of the indices at tracking changes in growth traits becomes much more experiment-dependent (Babar et al., 2006; Christensen & Goudriaan, 1993). Nevertheless, as stressed above, one of the practical applications of spectral reflectance may be its use as a routine tool for screening germplasm in breeding programs, when measurements are taken on a genotype basis, usually in one or a reduced number of experiments. Moreover, vegetation indices are more appropriate for assessing LAI than for estimating biomass (Aparicio et al., 2000, 2002; Serrano et al., 2000), particularly when measurements are taken with low variability backgrounds.