Dealing with HTTP data in modern crop breeding programs

Abstract

Modern crop breeding programs are data-driven. A breeder’s decisions are based on the prediction of the genotype performance from a large number of field trials. These trials should account for environmental variability of the target region, and more importantly, they should possess a high degree of accuracy. In recent years, different robotic and sensor technologies for collecting high-throughput field-based plant phenotyping (HTTP) data have been developed. Thereby, the possibility for gaining higher overall precision, as well as data and decision accuracy from crop breeding field trials was gained. Prediction of end-of-season yield and quality will become faster with the use of cameras for hyperspectral imaging, which is important for large scale producers. Comparing big sets of images generated in the field with results of classical chemical analyses serves as an advanced crop quality prediction tool for breeders. Important steps in such data analysis are calibration, noise reduction and the search for the most significant relations. Nevertheless, assessing phenotypic traits within genetic collections is made more accurate with the aid of phenotyping platforms that record plant growth from germ to seed. Like many types of phenotypic data, HTTP data collected from the images may also have some amount of unknown variability.