Prediction of potato processing quality by near infrared reflectance spectroscopy of ground raw tubers

The technique of near infrared reflectance spectroscopy was tested to predict the processing quality of potatoes by scanning 18502500 nm) ground raw potatoes (n=2517) from nine growing seasons and four to five locations. Calibration equations were developed for dry matter, starch, reducing sugars, sucrose and total sugar as well as lightness of dehydrated potatoes and potato crisps, and quality score of finished French fries. Two strategies in modelling were compared, an annually repeated (incremental) and a retrospective calculation at the end of the project with inclusion or removal of spectral outliers. The best coefficients of determination (r(2)) within independent validation sets were about 0.99 for dry matter, 0.96 for starch, 0.43 for reducing sugars, 0.71 for sucrose and 0.66 for total sugar. The RPD statistic (ratio of standard error of prediction to sample standard deviation) was up to 8.5 and 5.4 (dry matter and starch content, respectively), but was unacceptably low in the case of reducing sugars (1.7) and low for sucrose (2.4) and total sugar (2.3). Direct prediction of processing quality resulted in r(2) values of 0.52 for dehydrated potatoes, 0.69 for potato crisps and 0.56 for French fries, but the RPD of these models were too low even for screening purposes (RPD < 2.0).