• Tidak ada hasil yang ditemukan

CHAPTER 6: Feasibility studies on the use of near-infrared reflectance

6.4 Discussion

Observed differences in sugarcane varieties in terms of resistance may be due to biochemical differences in the leaves. This study aimed to predict differences between sugarcane varieties with respect to pest resistance based on NIR spectra obtained from their intact leaf surfaces and reference values obtained from various screening techniques. Previous studies have shown that NIR can penetrate up to 2.5 mm into plant material which infers that NIR spectra should represent the

biochemical and structural composition of the leaf that could be related to sugarcane resistance to pests such as C. partellus and F. serrata which feed on them (Purcell et al., 2010a). Although the results from this investigation are preliminary in nature, some of the calibrations do show a degree of potential for the development of predictive models for F. serrata and Chilo spp. resistance in sugarcane.

Although the PLS regression calibrations were fairly good for certain reference parameters, with R2 values higher than 0.8 achieved, results obtained from the

184

majority of the cross validation models (leave one out method) were discouraging.

This is probably due to the calibration data sett being too small. When a sample is left out for cross validation, the effect on the predictive model is too great resulting in a prediction that differs greatly from that of the completely inclusive calibration model (Table 6.8).

Calibrations and cross validations for all reference values obtained during oviposition experiments were particularly poor, indicating that no correlation occurred between the spectral data and the measured values for ovipositing. This could be due to the lack of any significant differences between varieties for egg number and batch

number found during oviposition experiments, suggesting poor quality reference data and consequently a poor predictive model. Perhaps the differences in sugarcane varieties with respect to ovipositing behaviour were a result of olfactory stimuli

(Thompson and Pellmyr, 1991). Herbivore induced plant volatiles include terpenoids, benzenoids, and green leaf volatiles which are released from the plant to either attract or repel insects. It has been shown that ovipostion by herbivorous insects on plants can result in a change in volatile emissions which may deter further deposition of eggs (De Moraes et al., 2001; Fatouros et al., 2012).

Results from the test validations were more promising than those from the cross- validations for the selected reference parameters. The mean number of shotholes from Pot Trial Two used as a reference parameter was shown to have the best calibration and test validation performance, with 75% of the variation in the reference data being accounted for by the spectral data obtained from leaf surfaces. Perhaps it is because the number of shotholes is a direct measurement of damage caused to the leaves by larva and is therefore more closely related to spectral data taken from the leaves than other parameters measured. The mean number of shotholes is a good indication of susceptibility of sugarcane varieties to C. sacchariphagus and C.

partellus and has been shown to be a non-destructive measure by which to rate sugarcane varieties (Conlong et al., 2004). Using this parameter to develop NIR models for predicting for resistance may be very useful in the future because it indicates that compounds in the leaf effect larval feeding and should be explored further. The mean number of shotholes was predicted to be fairly close to the actual number of shotholes for sugarcane varieties M1135/64 and R570, indicating that they

185

could possibly have a strong constitutive resistance against C. partellus larvae. Test validation models built using larval survival ratings of C. partellus from diet bioassays gave an R2 value of 0.63 and a SEP of 2.9. Field based ratings are known to have an associated error of +/- 1 units, indicating that the SEP for survival rating is too high and is therefore unsatisfactory to be used for screening purposes (Purcell et al., 2009). Similarly, a high SEP of 2.2 was observed for the test validation model using F. serrata ratings as reference values. High SEP values could be attributed to a poor or skewed range of values within the sample set (Edney et al., 1994).

In general terms, for NIR calibrations to be of use in the prediction of unknown samples (e.g. for total wheat nitrogen (N) content), R2 values for calibration and validation should be greater than 0.80. If R2 values are between 0.7 and 0.8 then the calibration can be used for rough prediction or classification, while R2 values less than 0.7 require further calibration development (Williams, 2001). The SEC and SECV should be as small as possible for good calibrations, while a large gap between SEC and SECV or SEP indicates that the sample sett is too small

(Dardenne, 2010). Acceptable predictions are characterized by low SEP values and high R2 and RPD values (Chen et al., 2002).

However, unlike analyses such as total N content, the determination of reference resistance ratings using live plant inoculation assays quantifies total resistance, whereas NIR scans of undamaged plant material can only be linked to constitutive preformed resistance. Since the inducible component of resistance is not accounted for, high R2 values in calibration or validation should not be expected when calibrating against total resistance reference values determined in live plant bioassays. If an equal contribution of constitutive and induced resistance were assumed, then an R2 in calibration of 0.5 would be reasonable.

Relatively low R2 values are still useful in plant breeding programmes where the aim might be to discard all susceptible clones at an early stage in the selection process.

Due to large numbers, conventional screening at early selection stages is not

possible. According to calculations made by Schenk and Westerhaus (1993), clones predicted as susceptible (of the three groupings; resistant, intermediate, and

susceptible) would include only 4% of actual resistant clones, based on a R2 during