For the accurate and automatic mapping of forest stands based on very-high-resolution satellite imagery and digital orthophotos, precise object detection at the individual tree level is necessary. Currently, supervised deep learning models are primarily applied for this task. To train a reliable model, it is crucial to have an accurate tree crown annotation dataset. The current method of generating these training datasets still relies on manual annotation and labeling. Because of the intricate contours of tree crowns, vegetation density in natural forests and the insufficient ground sampling distance of the imagery, manually generated annotations are error-prone. It is unlikely that the manually delineated tree crowns represent the true conditions on the ground. If these error-prone annotations are used as training data for deep learning models, this may lead to inaccurate mapping results for the models. This study critically validates manual tree crown annotations on two study sites: a forest-like plantation on a cemetery and a natural city forest. The validation is based on tree reference data in the form of an official tree register and tree segments extracted from UAV laser scanning (ULS) data for the quality assessment of a training dataset. The validation results reveal that the manual annotations detect only 37% of the tree crowns in the forest-like plantation area and 10% of the tree crowns in the natural forest correctly. Furthermore, it is frequent for multiple trees to be interpreted in the annotation as a single tree at both study sites.

Deep learning models in the field of individual tree detection and crown delineation (ITDCD) rely on large and high-quality annotation datasets to produce accurate predictions. Training data or annotations for most ITDCD studies are collected through manual labeling. Manual labeling, especially for complex structures like tree crowns, is a time-consuming process that often results in error-prone annotations. Error-prone annotations, in turn, can lead to significant errors in the predictions of deep learning models. Semi- or fully-automated training data creation shows the potential to make the creation process more efficient and ensure high quality of the training dataset. In this work, we present a methodology for generating semi-automated synthetic training data for deep learning-based ITDCD applications. Furthermore, a systematic criteria-based - validity, efficiency, variety and scalability - comparison is conducted between the manual and synthetic training data creation methods to structurally and practically illustrate the advantages and disadvantages of the two approaches. Overall, the semi-automated synthetic data approach outperforms manual labeling in terms of validity, efficiency, and scalability; once the algorithm is implemented, it rapidly generates arbitrarily large, high-quality, reproducible tree crown annotation datasets. In contrast, a manual creation approach shows its advantages as an efficient way to create small, low-quality datasets (e.g., for fine-tuning a pre-trained model) compared to developing a semi-automated method from scratch.