Implementation of an Unoccupied Aerial System in a Sorghum Breeding Program

Abstract

Sorghum bicolor (L.) is a diploid, commonly grown as a hybrid grain crop in developed countries for animal feed, food, and bioenergy. Breeding methodology of this crop is being transformed due to unoccupied aerial systems (UAS also called drones) which allow researchers to collect large amounts of phenotypic data on existing, and new traits. However, determining how to consistently collect this data and apply it to breeding programs is still being determined. The objective of the first chapter of this thesis was to compare vegetative indices collected in the morning and at solar noon at various altitudes to further define the limitations of data collected with a UAS. The objectives of the second chapter were to determine if camera settings significantly impact vegetive indices extracted from UAS, and to compare methods of phenotyping herbicide tolerance in sorghum. To collect imagery, a UAS with an RGB camera was flown over multiple days, over two fields containing grain sorghum. The flights were used to determine how indices can be consistently collected across various times and altitudes, along with how to use vegetative indices when breeding for herbicide tolerant populations. When 22 indices were extracted using images from varying altitudes and times on different flight dates, there was no significant differences in the rankings of the genotypes. This indicated that flights earlier or later in the day produced the same results as collecting indices at solar noon, which is the industry recommended flight time. The collection of indices was further improved through determining that the settings of the camera does not significantly impact the rankings of indices. Within grain sorghum sprayed by tembotrione, which is a post-emergent used to control broadleaves and grasses in corn, these indices could also determine which populations were the most susceptible early in the season. Machine learning models, such as random forest and neural network showed promise for estimating ratings of plots at 14, 21 and 53 days after tembotrione application. Overall, this thesis suggests that UAS can consistently collect reliable data in the mornings and is a beneficial tool when breeding for herbicide tolerance within sorghum.