Of Plants and Pixels: Leveraging Biological Priors for Ecological Data Analysis When Generic Methods Fall Short

As ecological research continues to expand across temporal and spatial scales, it generates immense amounts of data. The constantly growing vol ume and complexity of this data poses a major bottleneck in the scientific process. The primary objective of this interdisciplinary PhD project was to ac celerate several data analysis tasks in ecological research that are too time consuming to perform manually. In a first step, the feasibility of achieving this goal through the use of existing, generic methods was evaluated. For some specific tasks, such as instance segmentation of tree rings or root turnover estimation, the performance of these generic methods failed to meet the desired level of accuracy because of unique requirements that are not adequately addressed by generic computer vision research. This led to the development of new algorithms addressing these new challenges which constitute the main part of this thesis. Incorporating domain-specific prior knowledge of the underlying biological systems and processes has shown to be essential in guiding the development of these
algorithms. The presented algorithms are packaged as user-friendly analysis tools that do not require a high level of technical expertise. They have already proven their usability within several ecological studies and led to new follow-up projects.