High-throughput non-destructive biomass determination during early plant development in maize under field conditions

Abstract

Breeding, genomic, and physiological research on early growth in plants is hampered by the lack of suitable tools for non-destructive phenotyping of the above-ground biomass of a large number of genotypes in field trials. We designed a high-throughput phenotyping platform employing light curtains (LC) and spectral reflectance (SR) sensors mounted on a tractor and evaluated its performance under field conditions. The objectives of our study were to (i) compare biomass determination by LC, SR sensors, and their combination (LC⊕SR) using various biometric methods, (ii) evaluate the effect of the composition of the calibration data set on the mean relative error of prediction (MRE) and coefficient of determination of validation ( R v 2 ), and (iii) assess the repeatability ( w 2) of biomass determination. Twenty maize genotypes were grown in field trials in five environments. Sensor measurements were taken at three stages (full development of the fourth, sixth and eighth leaf). After recording sensor measurements, plots were harvested to determine fresh biomass. Biomass prediction was based on linear, non-linear and locally weighted polynomial regression for LC measurements. Partial least squares regression (PLSR) and support vector machine regression (SVMR) were used for SR and LC⊕SR measurements. The LC⊕SR data using SVMR and global sampling resulted in the lowest MRE (0.11) and the highest R v 2 (=0.97). Repeatability based on duplicate measurements of each plot was very high. In conclusion, this study provided a proof-of-concept that the described high-throughput, non-destructive phenotyping platform based on LC and SR sensors has a great potential for early biomass determination in field trials of maize and other row-crops.