Enhanced Individual Tree Detection and Crown Delineation Using Generic VHR Data for the Example of Augsburg for Operational Use in Urban Areas

Kurzfassung

Accurate knowledge of urban trees’ number and location is crucial for climate change adaptation and urban climate modelling. However, complete information on the urban tree population is rarely available. Remote sensing offers a suitable alternative to manually recording trees, which is costly and time-consuming. This thesis focuses on optimizing and enhancing a method for detecting trees from very high-resolution (VHR) remote sensing data for a generic data setting using the example of Augsburg. Identifying and segmenting urban trees is carried out using a Local Maximum (LM) filter and Marker-Controlled Watershed Segmentation (MCWS). Preprocessing and LM filter parameters are optimized through grid search in two representative areas of the urban forest: a park and a street. The results are quantified and analysed by comparing the detected tree positions with those in the Augsburg tree cadastre, serving as a reference. An F-score of 0.77 is achieved for both sample areas. A local sensitivity analysis is conducted on the results of individual tree detection (ITD), and the characteristics of both detected and overlooked tree markers are examined. To properly assess tree crowns extending above buildings, local maxima searches are refined by incorporating Light Detection and Ranging (LiDAR) point clouds, distinguishing between buildings and tree canopy through backscatter signal intensity and height distribution. The improved approach is then applied across Augsburg’s urban area. From the grid search results, a suitable parameter setting is selected for individual tree detection and crown delineation (ITD/CD) across the city. A total of 655,864 trees are identified in Augsburg, a significant improvement compared to the 66,000 trees in the existing tree cadastre. The newly created dataset is utilized to characterize the urban forest of Augsburg, analysing the spatial distribution of detected trees and crown cover in the area. Additionally, urban forest characteristics are compared with satellite-based land surface temperatures (LST) from June to August in the years 2020-2024, focusing on differences among Augsburg’ built-up Local Climate Zones (LCZs): 'Compact Midrise', 'Open Midrise', 'Open Low-Rise', and 'Large Low-Rise'. The strong negative correlation between LST and both tree height and crown cover supports the assumption that trees contribute to cooling temperatures. Similarities can be observed between the two 'Open' categories, as well as for 'Compact Midrise' and 'Large Low-Rise' regarding temperature and tree characteristics. Detailed analysis of LCZs in 10% increments of canopy cover related to temperatures reveals that, especially in areas with low canopy cover (up to approximately 30%), tree cooling capacity is not the only factor affecting temperatures. However, as expected, the cooling impact of the trees generally increases with canopy cover.