Using Unmanned Aerial Vehicle and Ground-Based RGB Indices to Assess Agronomic Performance of Wheat Landraces and Cultivars in a Mediterranean-Type Environment

• Published in: Remote sensing (Basel, Switzerland) Switzerland), 2021-03, Vol.13 (6), p.1187
• Main Authors: Rufo, Rubén, Soriano, Jose Miguel, Villegas, Dolors, Royo, Conxita, Bellvert, Joaquim
• Format: Article
• Language: English
• Subjects: drought stress; ground-based RGB image; high-throughput phenotyping; phenology; UAV imagery; vegetation indices
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs13061187

The adaptability and stability of new bread wheat cultivars that can be successfully grown in rainfed conditions are of paramount importance. Plant improvement can be boosted using effective high-throughput phenotyping tools in dry areas of the Mediterranean basin, where drought and heat stress are expected to increase yield instability. Remote sensing has been of growing interest in breeding programs since it is a cost-effective technology useful for assessing the canopy structure as well as the physiological traits of large genotype collections. The purpose of this study was to evaluate the use of a 4-band multispectral camera on-board an unmanned aerial vehicle (UAV) and ground-based RGB imagery to predict agronomic traits as well as quantify the best estimation of leaf area index (LAI) in rainfed conditions. A collection of 365 bread wheat genotypes, including 181 Mediterranean landraces and 184 modern cultivars, was evaluated during two consecutive growing seasons. Several vegetation indices (VI) derived from multispectral UAV and ground-based RGB images were calculated at different image acquisition dates of the crop cycle. The modified triangular vegetation index (MTVI2) proved to have a good accuracy to estimate LAI (R2 = 0.61). Although the stepwise multiple regression analysis showed that grain yield and number of grains per square meter (NGm2) were the agronomic traits most suitable to be predicted, the R2 were low due to field trials were conducted under rainfed conditions. Moreover, the prediction of agronomic traits was slightly better with ground-based RGB VI rather than with UAV multispectral VIs. NDVI and GNDVI, from multispectral images, were present in most of the prediction equations. Repeated measurements confirmed that the ability of VIs to predict yield depends on the range of phenotypic data. The current study highlights the potential use of VI and RGB images as an efficient tool for high-throughput phenotyping under rainfed Mediterranean conditions.