Wilke, Norman: UAV remote sensing for the spatial differentiated assessment of plant traits based on multispectral images. - Bonn, 2023. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-69368
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-69368,
author = {{Norman Wilke}},
title = {UAV remote sensing for the spatial differentiated assessment of plant traits based on multispectral images},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2023,
month = jan,

note = {Rising global temperatures, carbon dioxide (CO2) concentration, and the occurrence of extreme weather events are constantly changing the environmental conditions for crop productivity. Identifying climate-adapted and high-yielding cultivars has become a challenging task for the plant breeding community. Time-efficient and objective crop phenotyping is advantageous in reliably identifying desired varieties. Moreover, the identification of variabilities within the agricultural system is an important means of enabling site-specific management and increasing agricultural sustainability. In recent years, unmanned aerial vehicles (UAVs) have opened new frontiers for high-throughput phenotyping. Small-sized and lightweight sensors can be equipped on UAVs, facilitating the mapping of field heterogeneity, with a high spatial and temporal resolution. The enormous potential to assess different crop parameters nondestructively using the three-dimensional (3D) or spectral information has been investigated in this work. In addition, the potential limitations and the effort of applying the developed methods to unknown datasets are examined in this thesis, to meet the needs of precision agriculture and plant breeding.
The first study focuses on the lodging assessment of barley, based on the 3D information. Canopy height models were utilized to quantify the presence or absence of lodged plants (lodging percentage), as well as the intensity of the permanent displacement of plants from their upright position (lodging severity). Height thresholds that make use of a mathematical approach enabled to distinguish between the naturally occurring height variations and the lodged plants. The results showed a high correlation of lodging percentage to reference data (RMSE = 7.66 %) when applied to breeding trials. A similar accuracy could be observed in the case of a farmer’s field. The use of the 3D information is nearly independent of the abiotic or biotic factors and is characterized by high repeatability and applicability.
The second study of this thesis makes use of multispectral imagery for the plant density assessment of barley and wheat. The methodology was based on an empirical regression model and the relationship between plant density, counted in the field, and calculated fractional cover (number of plant pixels per square meter). The results illustrated that, at an early seedling stage, fractional cover is closely related to the number of plants (RMSE < 34 plants m-2). In order to expose the robustness of this statistical relationship, a developed model was applied to an unknown dataset and showed a high degree of accuracy (RMSE = 24 plants m-2). In addition, 11 reference measurements have proved to be sufficient, to enable the adaptation of the empirical regression model to the scene.
The third study explores the application of high-resolution RGB imagery for the spike density assessment of wheat. Similar to the previous study, the statistical relationship between the manually counted spike density and the UAV-derived spike cover (number of spike pixels per square meter) was examined. The spike density could be accurately modelled (RMSE < 18 spikes m-2) for three different genotypes, with phenotypic variations of canopy and spike characteristics. Investigations in this study have led to the assumption that 11 reference measurements can enable the adaptation of the empirical regression model.
In summary, the work demonstrated the great diversity of mapping crop heterogeneity with UAVs on a field scale. This potentially enables the faster selection of superior lines, the spatially differentiated prediction of crop yield or site-specific management for different nutrition and soil conditions. The independency or known effort of adapting the methodologies to unknown datasets increases the application potential in practice. However, the greatest challenge in remote sensing is to demonstrate universal applicability or the satisfactory adaptation of the methods.},

url = {https://hdl.handle.net/20.500.11811/10571}

The following license files are associated with this item: