Improving the Precision of NDVI Estimates in Upland Cotton Field Trials

Abstract

Controlling for experimental error attributable to field heterogeneity is important in high‐throughput phenotyping studies that enable large numbers of genotypes to be evaluated across time and space. In the current study, we compared the efficacy of different experimental designs and spatial models in the analysis of canopy spectral reflectance data collected on upland cotton (Gossypium hirsutum L.). Canopy spectral reflectance, as measured by normalized difference vegetation index (NDVI), was measured at first bloom on three upland cotton performance trials conducted in Florence, SC, during 2014 and 2015. The relative efficiency and estimates of genotype effects were compared among randomized complete block, an α‐lattice incomplete block, row–column incomplete block, nearest neighbor adjusted, and spatially correlated error models. The row–column model provided the greatest improvement in the precision of genotype effect estimates compared with the randomized complete block model. Genotype rankings based on NDVI varied substantially between the randomized complete block and alternative models, particularly at 5 and 10% selection intensities. These results suggest that the use of more complex experimental designs and spatial analyses should be routinely considered to minimize experimental error due to field heterogeneity and improve the precision and reliability of traits measured using high‐throughput phenotyping systems. These findings also indicate that further research into the effects of field heterogeneity on the relationship between NDVI and lint yield in upland cotton is warranted.