Assessment of durum wheat yield using visible and near-infrared reflectance spectra of canopies

The estimation of grain yield before harvesting could be a very useful tool for breeding programs and productivity forecasting. Canopy reflectance indices have been used for yield estimation, but with limited success. This work was carried out to study the suitability of the visible and near-infrared reflectance spectrum of the canopy for the assessment of grain yield in a set of durum wheat genotypes. Five field experiments, each one including 25 genotypes, were conducted in low, medium and high productivity environments, with average yields of 2.5, 4.5 and 7 t/ha. Spectral reflectance measurements between 400 and 1000 nm were made at anthesis and milk-grain stages. Partial least squares regression (PLSR) was used in the construction of models that were tested by simple regression between genotype means of predicted and observed grain yields. The empirical models for the estimation of grain yield showed generally stronger and more robust assessment of grain yield than previously assayed spectral indices. For the best model, correlation coefficients between genotype means of predicted and measured yield within each of the five environments ranged from 0.53 to 0.76. We concluded that, although the models did not provide an accurate quantification of grain yield, they could still be used to rank genotypes for breeding purposes. The most reliable ranking of genotypes was attained using measurements made at milk-grain stage on medium to high productivity environments.