Estimation of grain yield by near-infrared reflectance spectroscopy in durum wheat

Near-infrared Reflectance Spectroscopy (NIRS) is widely used for routine, non-destructive assessment of grain quality in cereals. In this work, we assess the use of NIRS as an indirect indicator of grain yield (GY) in durum wheat grown in Mediterranean conditions. Plants were grown under rain-fed conditions in two sites in NW Syria, Breda and Tel Hadya, and a third trial was planted at Tel Hadya under support irrigation. We measured the spectral reflectance signature between 1100 and 2500 nm in flour from the same grains harvested for yield determination. By Partial Least Squares (PLSR), we obtained a model of the association between agronomic determinations of GY and these spectra. We found significant (p < 0.001) relationships between GY and NIRS-predicted values, with r(2) = 0.36 (N = 96), r(2) = 0.72 (N = 72) and r(2) = 0.52 (N = 138) in Breda, and Tel Hadya rainfed and irrigated, respectively. Full cross validation of these models gave r(2) between 0.31 and 0.33 (p < 0.001). When we applied the models to the samples from the other two trials, we found a general decrease in their predictive ability. Compared with other grain traits ( carbon isotope discrimination, grain weight, ash and nitrogen content), NIRS showed a similar or even stronger relationship with GY, at least when validated in the same trial. We concluded that NIRS could be used to rank GY during the early generations of a breeding program, when performing yield trials is less feasible. However, further research is required to identify the environments in which this technique is more reliable, as well as to evaluate the heritability of NIRS-predicted GY.