Search

• EPO was recommended for sunlight correction. • Results support use of hand-held NIRS with varying sunlight. • Results support use of hand-held NIRS estimation of optimal harvest date. The models for hand-held near infrared spectroscopy (NIRS) instrument calibrated in the laboratory are prone to failure during orchard application because of measurement conditions change especially on sunlight interference. The models transfer from lab to orchard for assessment of soluble solids content (SSC) of navel orange on tree was investigated with consideration of sunlight influence to the fruit spectra. Chemometrics methods of global model, external parameter orthogonalization (EPO) and generalized least square weighting (GLSW) were attempted to correct the influence of the sunlight. Comparison of global model, both EPO and GLSW improved the predictive ability of partial least square regression (PLSR) models. With few corrected samples requirement, EPO was recommended and validated in predicting the samples in the different harvest date and blocks. The best results of coefficient of determination (R2) of 0.73 and 0.90, root mean square error of prediction (RMSEP) 0.85 and 0.63 % and ratio of standard deviation (SD) to RMSEP of 1.94 and 3.27 in predicting the samples in days after full bloom (DAFB) 237 (block A) and 164 (block B) were obtained. The best model corrected by EPO method was also adopted to estimate the optimal harvest date with a good consistency with destructive analysis. The results suggested that EPO correction aided the model transfer from lab to orchard was feasible for assessment of SSC of navel orange on tree. [ABSTRACT FROM AUTHOR]