Cross-regional and temporal generalizability of portable visible spectroscopy for olive maturity classification

Olive maturity governs lot destination and the yield-quality trade-off in olive oil production, yet it is still monitored by subjective visual scoring of the Jaén index. Spectroscopic alternatives have relied mainly on near-infrared instruments, with scarce evidence from Brazilian orchards and almost no evaluation of spatial or temporal transfer. The preserny study investigated whether a portable visible spectrophotometer (400-700 nm, 31 wavelengths) combined with CIE L*a*b* colorimetry and machine learning can recover per-fruit ripening stage (six classes) with encouraging robustness across the evaluated Brazilian orchards.

A dataset of 7058 Arbequina olives from three orchards in Rio Grande do Sul and São Paulo across two harvest seasons (2023-2024) was classified by 10 algorithms under 15 preprocessing hypotheses and five validation protocols of increasing rigour. Within-dataset accuracy reached 99.06% (Extra Trees, $\kappa$  = 0.986), whereas leave-one-orchard-out validation reached 95.74% (SVM-RBF, $\kappa$  = 0.935), geographic transfer between states 96.60% and the conservative inter-annual direction 93.98%. Ordinal metrics (quadratic weighted kappa ≥ 0.982; accuracy within ± 1 class $\ge$ 99.7%) showed that residual errors were almost exclusively single-step. The 530-680 nm region carried most of the discriminative information, consistent with chlorophyll-to-anthocyanin pigment dynamics.

Visible-only portable spectroscopy classified per-fruit olive maturity with high accuracy under the evaluated conditions, and quantified a deployment gap of 3-5 percentage points between within-dataset cross-validation and cross-region or inter-annual hold-outs, comprising a gap that random cross-validation alone conceals. Because errors are confined to adjacent maturity stages, the approach suggests potential for future automated lot-level maturity index estimation in post-harvest quality control of extra virgin olive oil. © 2026 The Author(s). Journal of the Science of Food and Agriculture published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.