Normalized difference spectral indices and partial least squares regression to assess the yield and yield components of peanut

High-throughput hyperspectral passive reflectance sensing can acquire timely information to make more informed management decisions in real time compared with the more laborious destructive measurements. Early prediction of yield and yield components of peanut by spectral reflectance measurements prior to harvest could reduce the phenotyping time and expenses compared to destructive measurements. In this study, the performance of hyperspectral passive reflectance sensing was tested at three growth stages, the beginning of pod development and at 50% and 80% pod development, to assess their relationship to the pod yield, seed protein content, seed oil content, and straw yield of peanut cultivars Two peanut cultivars, Giza 5 and Giza 6, were grown under field conditions and subjected to three levels of nitrogen application. Simple linear regression and partial least squares regression (PLSR) models were compared to analyse the spectral data. The closest relationships were obtained for the spectral index (R₆₁₀ - R₄₂₄)/(R₆₁₀ + R₄₂₄) with the pod yield (R² = 0.70, significant at p ≤ 0.001), as well as the straw yield (R² = 0.53, significant at p ≤ 0.001) and the protein content (R² = 0.69, significant at p ≤ 0.001. For the relationships between PLSR with the pod yield, protein content and oil content and the straw yield of peanut cultivars, the coefficients of determination reached values up to R² = 0.82 (significant at p ≤ 0.001) through the individual measurements. Both, the PLSR and normalized difference spectral index analysis of spectral data performed better for assessing the pod yield and protein content than the oil content and straw yield of peanut cultivars. In conclusion, phenotyping yield and quality related parameters of peanut by PLSR analysis of non-invasive reflectance measurements represent a promising strategy for management action as well as for screening peanut cultivars.