Robust PLS models for soluble solids content and firmness determination in low chilling peach using near-infrared spectroscopy (NIR)

• Main Author: Nascimento, Paloma Andrade Martins [UNESP]
• Publication Date: 2016
• Other Authors: Carvalho, Lívia Cirino de [UNESP], Júnior, Luis Carlos Cunha, Pereira, Fabíola Manhas Verbi [UNESP], Teixeira, Gustavo Henrique de Almeida [UNESP]
• Format: Article
• Language: eng
• Source: Repositório Institucional da UNESP
• Download full: http://dx.doi.org/10.1016/j.postharvbio.2015.08.006; http://hdl.handle.net/11449/168061
• topic: Aurora 1; Chemometrics; Maturity stages; PLS; Prunus persica L

The objectives of this study was to develop partial least square (PLS) models using NIR spectroscopy for the determination of SSC and firmness in intact low chilling 'Aurora-1' peach fruit, and verify the influence of maturity stage and harvest season on the models to be developed (robustness). FT-NIR spectra were obtained as log 1/R with fruit harvested in 2013 at 3 maturity stages and in 2014. The spectra were collected on the background and blush colour skin areas of the each fruit. Model performance was evaluated based on the values of root mean square error for prediction (RMSEP) and coefficient of determination (RP 2) obtained from validation fruit set (Kennard-Stone), and prediction fruit set (2014). PCA could not group the fruit based on blush and background skin colour, maturity stages, and harvest season. The model constructed using the external validation method obtained a RMSEVE of 1.08 % with 11 latent variables (LVS) and a RVE 2 of 0.59. The prediction set, independent data, resulting in a less accurate model (RMSEP 1.04 %, Rp 2 0.45 and 11 LVS). The same trend happened for determining firmness with the external validation resulting in better model with RMSEVE 9.51N and RVE 2 of 0.40 and the prediction set with RMSEP of 13.2N, RP 2 0.40 with 7 LVS. The NIR spectroscopy showed to be a potential analytical method to determine SSC and firmness of intact low chilling 'Aurora 1' cultivar. However, it is necessary to optimize the models in other to reduce the prediction errors.