The potential of UAV-based remote sensing for the prediction of aboveground biomass and N fixation in legume-grass mixtures

The world population is growing steadily and with it the demand for meat and animal protein. A major source of feed for livestock is grassland, which serves as high quality and protein rich forage. In addition to permanent grassland, temporary grassland is a valuable crop rotation element, especially in organic farming, since legumes in legume-grass mixtures can fix atmospheric nitrogen (NFix). To further increase the intensification in agriculture, reliable non-destructive measurement techniques are needed as an alternative to traditional labor- and time-consuming methods in grasslands. Precision agriculture offers an efficient information-based and site-specific approach to support farm management decisions. In this context, remote sensing is the key driver for precision agriculture to gain these goals. Especially unmanned aerial vehicle (UAVs) are a low-cost and flexible platform for different sensor applications. Temporary grassland management in precision agriculture plays only a secondary role compared to other agricultural crops. To further close this research gab, this thesis aimed to examine whether remote sensing measurements can be used to estimate biomass and N fixation under field conditions in two legume-grass mixtures including varying proportions of legumes (0-100%). In this thesis three different approaches were tested with three multi-temporal studies: (i) point cloud-based canopy surface height measurement, (ii) multispectral information and (iii) the combination of both in form of sensor fusion. The first approach was based on the strong correlation between canopy surface height (CSH) and aboveground biomass (Chapter 3). Photogrammetric structure from motion (SfM) processing of UAV-based red, green, blue (RGB) images into point clouds can be used to generate 3D spatial information about CSH. The study showed that biomass estimation by SfM-based UAV RGB imaging provided similar accuracies across all treatments as the ruler height measurements, even under extreme weather conditions (drought). Nevertheless, the high variability of the canopy surface required supplementary spectral and structural information. The second approach used UAV-based multispectral information and vegetation indices for aboveground biomass and NFix estimation (Chapter 4). To include also horizontal spatial relationships of pixels to each other in the images, texture features where calculated. The study compared the estimation models with and without the inclusion of texture features. The results showed, that biomass prediction accuracy for the whole dataset as well as for crop-specific models were substantially improved by the inclusion of texture features. The third approach combined point cloud-based CSH measurement with multispectral data and texture features (Chapter 5). This study included two-year data and to increase the spatial resolution, a terrestrial laser scanner was used for a high-density point cloud. Sensor fusion was found to provide better estimates of aboveground biomass and NFix than separate data analysis of point clouds and multispectral information. Furthermore, the study showed the important role of texture based on multispectral data, but also based on CSH data, which contributed greatly to the estimation model generation. The results of this thesis showed the challenges and diverse possibilities of the UAV-based biomass and NFix estimation of two legume-grass mixtures (Chapter 6). The rapid technical development of sensors, platforms and information technology offers constant improvements in precision agriculture and grassland monitoring. The approaches used in this work offer an interesting method for new possibilities at field and farm level.