HSI and ML for non-destructive pistachio quality assessment: influence of location and irrigation on nutrient and fat composition

Abstract

Pistachio quality is a major determinant of market value, particularly in Mediterranean environments strongly influenced by irrigation and climate. This study evaluated effects of orchard location and irrigation on pistachio nutritional content and assessed VIS–NIR hyperspectral imaging coupled with machine-learning techniques as a non-destructive predictive approach. Field experiments were conducted in 2022 in two commercial orchards in Castilla y León, Spain, under control and high-irrigation treatments. A total of 2818 pistachio kernels were analysed using a portable VIS–NIR hyperspectral camera (400–1000 nm). Mean spectra were extracted for each nut, pre-processed using standard normal variate correction, and linked to reference measurements of minerals, proximate composition and fatty acids. Partial Least Squares regression, Support Vector Regression and Extra Trees Regressor models were calibrated and validated to predict quality parameters from spectral data. Location, irrigation and their interaction significantly affected most nutritional and lipid traits. Pistachios from Moraleja de las Panaderas showed higher nitrogen, phosphorus, protein, ash and oleic acid contents, whereas samples from La Seca exhibited relatively higher sodium and linoleic acid levels. Increased irrigation enhanced the accumulation of several minerals and saturated fatty acids. Among the evaluated algorithms, Partial Least Squares regression provided the most consistent performance, accurately predicting nitrogen (R2 = 0.75), zinc (R2 = 0.81), oleic acid (R2 = 0.91), linoleic acid (R2 = 0.87), ash (R2 = 0.81), carbohydrates (R2 = 0.87) and humidity (R2 = 0.84). Overall, VIS–NIR HSI with machine learning enables non-destructive, data-driven optimisation of pistachio irrigation management.