Hyperspectral measurements of Cannabis sativa fan leaves during early floral development predict final cannabinoid yield

Abstract

Accurate, non-destructive prediction of cannabinoid concentrations in Cannabis sativa is critical for optimising crop value and ensuring regulatory compliance in the industrial and medicinal cannabis sectors. In this study, we demonstrate for the first time that fan leaf hyperspectral reflectance (FLHR) measurements taken across the canopy, performed both early and late in the flowering period, can reliably predict final cannabinoid content in mature inflorescences in two Cannabis cultivars under seven distinct lighting regimes. Machine learning models trained on FLHR spectra achieved high predictive accuracy, with R² values up to 0.89 for CBD, 0.77 for THC, and 0.8 for total cannabinoids, outperforming previous approaches. Importantly, our method utilises a hand-held, non-destructive hyperspectral device, enabling rapid, in situ assessment of intact fan leaves without the need for sacrificial sampling or laboratory analysis (e.g., for HPLC or GC-MS). FLHR measurements were able to differentiate cultivars and lighting treatments, offering a tool for germplasm classification. The capacity to predict cannabinoid profiles weeks before harvest has significant implications for cannabis production, enabling growers and breeders to enhance product quality, reduce costs, and ensure regulatory compliance, particularly for industrial hemp crops subject to strict THC limits, or to track and predict yields for medicinal cannabis operations.