Advisors: Markus Hollaus, Anna Iglseder

The monitoring of green spaces is becoming increasingly important in urban areas in order to take meaningful measures to combat climate change and mitigate its effects on urban areas. For this reason, this project was initiated to refine the monitoring of green spaces in Vienna. The greening of public spaces is already recorded in high spatial resolution by the tree register, whereas other greening measures are currently mapped to a much lesser extent. This also includes the greening of roofs, a greening measure that is currently becoming increasingly common. In the past, this mapping was only carried out as a binary yes/no classification, but a more detailed classification of the stock into categories such as tree, shrub, meadow, overhanging tree or sealed roof would be useful. To implement this project, this master thesis aims to develop an automated script for the detailed classification of green roof types in Vienna, which is to be carried out based on current airborne laser scanning (ALS) data from June 2023. The data was collected during the leaf on season and therefore offers an ideal basis, as the vegetation is fully developed at this time. In addition to geometric information, the ALS data also contains spectral data (RGBI information), which enables an extended analysis. In addition to the ALS data, data from the City of Vienna's multi-purpose area map will also be incorporated into the classification process described below. Relevant test areas in Vienna are to be defined to validate the results and to assess the quality of the classification. Representative areas for each greening class should be found and mapped, and the mix of sealed roof surfaces should be as heterogeneous as possible. To reliably identify the different types of vegetation, an explorative statistical analysis of the ALS data is carried out. Characteristic attributes of the individual vegetation classes are extracted and analyzed. To handle and process the point cloud efficiently, a Python script based on OPALS modules is developed. To include only the roof areas in the analysis, the point cloud is clipped with a mask before classification using the building polygons from the multi-purpose area map. Expert-based decision trees and thresholds will be combined in this script to further automate the classification. For further optimization, a vegetation map derived in another ongoing project will also be used as an additional decision feature to distinguish between vegetated and non-vegetated areas.The final output of this study is a pixel-based classification map of roof surfaces in Vienna. To assess the quality of the classification, the generated map is compared with manually mapped reference data. The validation results indicate that, depending on the test area, an overall accuracy between 0.84 and 0.96 is achieved. These accuracy values confirm that the proposed approach reliably identifies the different vegetation and roof types in most cases. The resulting classification map therefore provides a detailed and robust representation of the current state of green roofs in Vienna. It serves as a valuable decision-support tool for urban planning processes and supports future measures in the fields of climate protection and green infrastructure.