Developing OBIA methods for measuring plant traits and the legacy of crops using VHR UAV-based optical sensors
Summary
Unmanned aerial vehicles (UAV) have emerged as a flexible, affordable, and easy-to-use technology for remote sensing within fields such as geoscience, precision agriculture and forestry management. The capacity of taking very high spatial resolution (VHR) optical images makes UAVs highly suitable for individual crop detection and analysis. Following an object-based image analysis (OBIA) approach, the thesis investigates to what extent image-objects can be built for cichorium endivia (endive) crops at the end of the growing season. The main capabilities of OBIA are delivering a highly accurate detection of crops (99.8% accuracy) and precise measuring of crop-covered areas without including any shadows or bare soil.
Hyperspectral data and a digital surface model (DSM) are combined with the OBIA-based image-objects, so-called data fusion. These data-layers can be used for plant-volume calculations and application of vegetation indices (VI) on the object-level. A randomized agro-ecological field experiment by Barel et al. (2018) is used to validate results from spectral and geometric object analysis. They assess legacy effects of preceding crops via the soil, known as plant-soil feedback (PSF), by destructive sampling and qualifying cichorium plant traits such as biomass. Four different cover crop species and their mixtures are used as a plant-soil treatment: Lolium perenne (perennial ryegrass), raphanus sativus (radish), trifolium repens (white clover), and vicia sativa (common vetch). Optical and hyperspectral UAV-data was acquired prior to destructive sampling. OBIA shows comparable results. The raphanus monoculture and mixture raphanus + vicia cause highest plant-volumes and high scores for most VIs used. The lolium treatment has overall the lowest average crop volume and VI scores.