Predicting condensed tannin concentrations in Lotus uliginosus Schkuhr using near‐infrared reflectance spectroscopy

Abstract

Near-infrared reflectance spectroscopy (NIRS) was used to develop equations to predict condensed tannins (CT) in greater lotus (Lotus uliginosus Schkuhr) with a precision satisfactory for the screening of genotypes in a plant breeding programme. NIRS equations were developed using both partial least squares and stepwise multiple linear regression techniques, and a wide range of mathematical treatments of the log 1/reflectance NIRS data. In general, equations developed using partial least squares regression techniques had 10–15% lower standard errors than those developed using stepwise multiple linear regression, during both calibration and prediction. Standard errors of calibration ranged from 10 to 17 g kg−1, standard errors of prediction from 12 to 17 g kg−1. These errors equate to coefficients of variation in the order of 20% The use of NIRS to predict CT in greater lotus will allow more rapid evaluation and selection of genotypes than could be achieved using butan-1-ol/HCl hydrolysis. © 1997 SCI.