Enhanced Total Recoverable Sugar (TRS) Estimation in Sugar Cane Juice by HPLC and Rapid TRS Prediction via NIR Spectroscopy Coupled with Partial Least-Squares Modeling

Abstract

This study proposes a rapid and straightforward method for determining total recoverable sugars (TRS) in sugar cane juice using near-infrared spectroscopy (NIR) and partial least-squares (PLS) regression. PLS models were built using NIR spectra obtained directly from sugar cane juice as independent variables and TRS values as dependent variables. TRS was obtained via two estimations: (i) technological analysis (TRS^TA), indirectly determined using CONSECANA equations, and (ii) high-performance liquid chromatography, used to quantify sucrose, glucose, and fructose to calculate TRS, defined as TRS^LC. The estimated values for TRS^TA and TRS^LC ranged from 82.08 to 180.03 kg ton^–1 and from 116.14 to 216.83 kg ton^–1, respectively. A high correlation coefficient (R = 0.90) and a consistent bias of 24.75 kg ton^–1 were observed between TRS^TA and TRS^LC. Thus, a median bias correction was proposed to improve the alignment. To enhance model performance, variable selection was applied by using the ordered predictor selection (OPS) algorithm. PLS models built with the OPS showed improved accuracy compared to models using the full spectral range. The best PLS-OPS model for TRS^TA and TRS^LC presented root-mean-square errors of prediction of 7.8 and 9.8 kg ton^–1 and R of 0.91 and 0.86, respectively. Finally, the models were successfully applied in a practical genotype screening scenario, showing a high selection efficiency and strong agreement with HPLC-based rankings. These findings demonstrate the potential of NIR-PLS models as reliable, low-cost, and scalable alternatives for TRS prediction in sugar cane breeding and industrial decision-making.