UAV-Based Estimation of Grain Yield for Plant Breeding: Applied Strategies for Optimizing the Use of Sensors, Vegetation Indices, Growth Stages, and Machine Learning Algorithms

Non-destructive in-season grain yield (GY) prediction would strongly facilitate the selection process in plant breeding but remains challenging for phenologically and morphologically diverse germplasm, notably under high-yielding conditions. In recent years, application of drones (UAV) spectral sensing has been established, data acquisition processing have to be further improved with respect efficiency reliability. Therefore, this study evaluates measurement dates, sensors, parameters, as well machine learning algorithms. Multispectral RGB were collected during all major growth stages winter wheat trials tested GY using six machine-learning Trials conducted 2020 2021 two locations southeast eastern areas Germany. most cases, milk ripeness stage was reliable from individual maximum accuracies differed substantially between drought-affected (R2 = 0.81 R2 0.68 both locations, respectively), wetter, pathogen-affected conditions 0.30 0.29). The combination multiple dates (maximum 0.85, 0.81, 0.61, 0.44 four-year*location combinations, respectively). Among parameters investigation, best RGB-based indices achieved similar predictions multispectral indices, while differences algorithms comparably small. However, support vector machine, together random forest gradient boosting performed better than partial least squares, ridge, linear regression. results indicate useful sparser canopies, whereas improvements are required dense canopies counteracting effects pathogens. Efforts measurements more rewarding enhanced information (multispectral versus RGB).