Role of chemometrics for at-field application of NIR spectroscopy to predict sugarcane clonal performance

This paper demonstrates the crucial role of chemometrics in facilitating the knowledge transfer of a laboratory derived rating for sugarcane clonal performance (based on GC measurements) towards a rapid near infrared (NIR) spectroscopic methodology for at-field screening applications. NIR sugarcane stalk spectra were collected from 65 sugarcane clones, which were obtained from two widely different geographical regions. An exploratory principal component analysis (PCA) of an NIR spectral data matrix of the sugarcane clones showed a clear trend along PC1 similar to that found from the GC measurements, with the resistant clone (Q90) being well discriminated from the susceptible one (NCo310). Biplots related the NIR spectral frequencies with the spectral objects. Cellulose and lignin bands (4248 and 4406 cm− 1 respectively) were associated with the well performing clones and epicuticular wax bands (4346 and 5781 cm− 1) with the poorly performing ones. Several partial least squares (PLS) regression models were constructed from the spectral objects, and satisfactorily predicted the traditional sugarcane performance ratings (SEP range +/− 0.7–0.9 traditional units for the verification data, and the Measured versus Predicted R2 range 0.880–0.910). In addition, the stalk spectra were investigated with the use of multicriteria decision-making methods, PROMETHEE and GAIA. This demonstrated that the spectral objects could be ranked on the basis of the NIR information alone, and that this rank order was in agreement with traditionally determined ratings. Thus, the performance of sugarcane clones could be assessed from molecular information as from the earlier laboratory GC measurements, and therefore, can be obtained completely independently of any conventional nominal ratings scales.