Abstract: Models based on FT-NIR spectroscopy and PLS-regressions were developed over three harvest years to determine the condensed tannins contents (CT, %(w/w)) in sainfoin powders. The three data sets corresponding to the three harvest years were used as calibration and then validation sets, successively. Finally, a global model gathering all the three years data sets has been developed. The developed models predict CT in a range of 2.06 to 11.28 %(w/w). The accuracy of the models depended on the range of CT-values of the calibration and validation sets. Finally, it was possible to predict the CT with a SEP-value lower than 0.5% (w/w) and R2-value higher than 0.9. In the present study, some of the PLS-parameters such as bias, slope and SEP have been statistically evaluated using the international standard ISO 12099:2017. The final global model was very promising since bias was not significantly different from 0, the slope was not significantly different from 1, and obtained SEP was 0.49% (w/w) while the calculated SEP-limit was 0.47% (w/w). The presently developed model was robust over the three years and the global model presented very interesting values. Such approach would be very useful to develop a new quantitative, rapid and low cost method to assess the CT of sainfoin powders. This method will allow us to free ourselves from the traditional chemical method that consumes time, money and chemicals.
