Evaluating spectral vegetation indices for a practical estimation of nitrogen concentration in dual-purpose (forage and grain) triticale

There is an ample literature on spectral indices as estimators of the crop’s chlorophyll concentration, and, by extension, of the nitrogen concentration. In this line, the suitability of 21 of these indices was evaluated as nitrogen concentration indicators for the dual-purpose (fodder and grain) triticale (X Triticosecale Wittmack). The interval of interest was the one in that it would be possible to intervene to correct the deficiency of nitrogen (defined according to practical criteria); one peculiarity of this study is that it only develops a model for that period; more developments complicate the profitability, because the annual stability is not guaranteed and calibration studies are expensive. The results showed that, although there are significant correlations between the greenness indices and the crop’s nitrogen concentration, for none of the spectral indices the relationship can reach acceptable values that encourage their use in the new techniques of precision agriculture of low cost. One solution for improving the effectiveness and reduce costs could be to use the information contained in the spectral signature beyond what is easily explicable by biochemistry and biophysics, in other words, using data mining in the search for new spectral indices directly related to the concentration of nitrogen in plant and stable throughout crop development. At present, the squared correlation coefficient ( R 2 ) of the best fits reach 0.5 for the later phenological stages, this mark is reduced to 0.3 with an approach of low cost. Resumen


Introduction
Reflectance measurements can be used to obtain the values of the most widely used spectral indices (reviewed in Ustin et al., 2004) as indicators of chlorophyll concentration. These indices, together with the canopy radiative transfer models, allow one to estimate the state of the vegetation from satellite and aerial images. Our ultimate goal is to make this approach a reality in agriculture and the action plan begins with identifying the most appropriate spectral index and estimating the magnitude of its relationship with nitrogen concentration. This study assesses the potential of the methodology in easily reproducible conditions on farms. The immediate objective is to determine if the spectral indices keep the effectiveness reported by other authors when implementation costs are reduced (only a model for relating reflectance and nutritional status by campaigning on a stage with high variability), which is necessary to facilitate the transfer.
Some of the spectral indices commonly used as indicators of chlorophyll concentrations include corrections for the effect of the soil on the measurements of the canopy reflectance (Huete, 1988;Rondeaux et al., 1996;Zarco-Tejada et al., 2004), unnecessary precautions in this study because it works with reflectance measurements of the leaf.
This work is in line with that described by Heege et al. (2008). They related the measurements of greenness, obtained by sensors mounted on farm equipment, with the dose of nitrogen fertilizer. They used fluorescence and reflectance measurements, in the case of reflectance; they tested the determination of the red-edge inflexion point both numerically and empirically.
The work of Li et al. (2010), although similar to this one, differs in that in this study is limited to one the number of models that relate the spectral index and the nitrogen concentration in the plant throughout the development, instead of looking for relationships for specific growth stages, that is what has been done so far because the plant changes during its development (Marschner, 1995;Azcón-Bieto and Talón, 2003) complicate another approach.
With the dual-purpose triticale, one has the possibility of allowing livestock to graze more than once without ruining the harvest. After each grazing by livestock the plant has to regenerate the above-ground part and with it the ability to synthesize chlorophyll, but the below-ground part is unaffected so that the plant's nitrogen absorbing capacity remains intact. The result is an imbalance in the first weeks after each cutting which is manifest in a yellowing of the plant. This is not a symptom of nitrogen deficiency, since there has been neither a decrease in the concentration of this element nor an interruption in plant growth. This characteristic is a major additional obstacle to the transference of radiometric technology to the dual-purpose triticale case.

Material and methods
As part of a study at the «La Orden-Valdesequera» Research Centre aimed at determining the optimal combination of seeding density, number of grazing and doses of nitrogenous fertilizers for growing triticale (X Triticosecale Wittmack), the reflectance of the leaves throughout the growth of the crop was measured.
The experimental design was a split-split-plot with four replicates. The first factor was the seeding density (400, 500 and 600 plants m -2 ), the second the number of times the crop was cut to simulate grazing (0, 1 and 2 grazing by livestock) and the third the dose of nitrogenous fertilizer (0, 75 and 125 kg ha -1 ). Each factor had three levels, so that there were 108 experimental plots in total, each of 30 m 2 . The leaf reflectance measurements were made at 80, 117, 132, and 164 days after seeding (campaign 2009). Together with these measurements, crop samples were taken and sent to the laboratory for the determination of the concentration of total nitrogen by the Kjeldahl method.
The weather is a key factor in plant growth and climate variability is sufficient to cause changes in the evolution of the crop year after year; for this reason, more important than the number of days from sowing is to indicate the phenological stage in which was the plant. Table 1 shows the correspondence between the number of days after seeding, the crop's phenological stage and the description of growth stage.
The growth stages were determined using the Zadoks (Zadoks et al., 1974) and Feekes scales (Feekes, 1941;Large, 1954). On this matter illustrative charts can be found in the book of Rawson and Gómez Macpherson (2000).
The influence of each factor on the nitrogen concentration measured for each of the 108 plots on each sampling day was analyzed (it is unknown whether all the factors at all levels have an effect on the concentration of nitrogen in plant). The analysis of variance for factorial designs (statistical analysis that corresponds 682 F. Rodriguez-Moreno and F. Llera-Cid / Span J Agric Res (2011) 9(3), [681][682][683][684][685][686] to the split-split-plot design) determines a grouping of the elementary experimental plots according to the nitrogen content on each of the four sampling dates. All calculations were done using R 2.9 (R Development Core Team, 2004). The crop's spectral signature is sensitive to phenological changes; this is a good reason to test the relation between the spectral index and the nitrogen concentration for each sampling date. This approach suffers from a serious handicap, a commercial application is difficult since in that case calibration studies would be needed for each growth stage, which compromises the economic viability of this technique. This is the reason because in this work, the restriction that only one model should cover the entire period of interest was established.
On each sampling date, 20 leaves at random were collected in each of the 108 elementary plots. Ten estimates of the reflectance (each averaging 50 readings) were made of these samples, using the ASD FieldSpec 3 spectroradiometer for this. This device has a spectral range of 350-2,500 nm, a sampling interval of 1.4 nm for the range 350-1,000 nm and 2 nm for the range 1,000-2,500 nm, and a spectral resolution of 3 nm at 700 nm and 10 nm from 1400 nm to 2100 nm. Readings were performed using a plant probe plus leaf clip. The light source of the plant probe is a halogen bulb with a colour temperature of 2,901 ± 10 K.
For each of the different groups of elementary plots defined according to their concentration of nitrogen on each sampling date, the mean reflectance was cal-Practical estimation of nitrogen concentration in triticale by spectral vegetation indices 683 culated by averaging the readings taken in their respective elementary plots. The average spectral signature for each of the groups identified during the growth of the crop was then used to determine the value of each of the selected spectral indices given in Table 2. The relationship between the different spectral indices and the nitrogen concentrations was evaluated by calculating the correlation coefficient squared (R 2 ), the adjusted squared correlation coefficient (aR 2 ), the square root of the mean square error (RMSE) and the statistical significance (p-value) of the model (analysis of the variance explained as against the residual). Figure 1 is a flowchart that summarizes the whole process. Table 3 shows that all the models, except the one developed for the modified chlorophyll absorption in reflec-tance index, explained a significant portion of the variance of the plant's nitrogen concentration, but in none of the cases the magnitude of the relationship was enough for developing a profitable methodology. The highest R 2 was 0.31, and corresponded to the green reflectance index.

Results and discussion
This study determines the reduction in the effectiveness if one tries to estimate the concentration of nitrogen in the plant by a low-cost approach. The reduction is 38% (the value of R 2 is reduced from 0.5 to 0.31), a significant reduction received with a positive evaluation because of the complexity involved in developing a single model for different phenological stages and the enormous handicap which entails the double duality of the crop. It is understood that this reduction in effectiveness is the highest possible and that's not enough to discard the approach of a precision agriculture of low cost, at least remains to evaluate the potential of data mining in the exploitation of the information contained in spectral signature. Looking for a valid model for the 4 days of sampling, so gather the data in a single set Those parcels that differ from each other by a factor or level without effect are repetitions of the same configuration A certain factor-level could have a significant effect at a given time and not have it in another, the analysis is performed by sampling day The spectral signature of a leaf is extremely sensitive to the conditions affecting the leaf itself and to the conditions under which the measurements are made. These conditions vary considerably throughout the period during which it would be possible to act to correct a possible nitrogen def iciency in the crop, for this reason the models were specific to a growth stage, something that complicates the profitability and thus the transfer of technology, this study explores this field and its conclusion suggests that the cost of using a single model is not as high as could be expected.
In light of the results, exactly of the results of F test and associated p-value for significance of the model, one can say that the independent variables, spectral indices, are important in explaining the observed variation in dependent variable, concentration of nitrogen in plant. These indices have a biochemical and biophysical basis, the existence of that relationship was expected, different is the magnitude of it, so small that makes one think about the weakness of using non-specific index for the concentration of nitrogen in plant.
It is confirmed that, in cultivating dual-purpose triticale, calculations based solely on a greenness measurement, such as the green reflectance index (the index which gave the best correlation in this study), would lead to obtaining erroneous estimates of the nitrogen concentration.
The single model developed for the entire period of interest -from stem elongation (when tillering has ended) through booting and heading to floweringgave lower correlations between the vegetation spectral indices and the crop's nitrogen concentration than those reported in similar studies in which different phenological intervals were modeled separately (Heege et al., 2008;Li et al., 2010). In particular, the value of R 2 was reduced by about 40%, from 0.5 for the f inal growth stages (Li et al., 2010) to the value of 0.3 for the entire period of crop growth.
The techniques of precision agriculture have to reach to the cultivation of the dual-purpose triticale. Since nitrogen's role in a plant is not restricted to be a component of chlorophyll and the spectral signature of the leaf is function of its composition and configuration, it has to be possible to derive new spectral indices which estimate nitrogen concentrations based on other variables besides of the greenness of the plant.
The results do not provide reliable indications of spectral features of those indices which best correlate with the crop's nitrogen concentration. Some authors, among others, the already cited Heege et al. (2008) Practical estimation of nitrogen concentration in triticale by spectral vegetation indices 685 and Li et al. (2010), have attempted to find relationships between the various spectral indices to explain their rank in terms of suitability (greater R 2 ). The lack of correspondence between the different rankings might be considered a further reason for not addressing this issue.