Key points are not available for this paper at this time.
Urban trees are essential for cities as they reduce the risk of flooding and provide shade and coolness in the summer months. However, these trees are exposed to environmental stresses, e.g. due to limited soil resources and unfavorable hydrological conditions caused by impervious surface and drought. As part of the CliMax project, our research aims to develop a method that uses LiDAR (light detection and ranging) and Sentinel-2 to monitor and estimate the vitality of urban trees in Braunschweig and Brandenburg a.d.H. Estimating urban tree vitality by conducting ground measurement requires a huge number of work force and resources, which is expensive and time consuming. Remote sensing enables continuous monitoring of trees within the urban area. In this research, we implemented a four-step methodology to detect individual urban trees based on airborne LiDAR data. Firstly, a pre-classified subset of the upcoming digital twin LiDAR data (harmonized Germany-wide data) was used to train a machine-learning model. This model is designed to distinguish between trees and buildings by relying on geometric features describing the three-dimensional LiDAR point distribution, such as planarity, sphericity or verticality. Secondly, the LiDAR data was rasterized into a Canopy Height Model (CHM) to delineate single trees by applying the slope break technique. We modified the conventional slope break computation to counteract the underestimation of the crown diameters. Third, the slope break values defined the different window sizes for the Local Maximum Filter (LMF) used to determine the spatial position of the treetops. Fourth, the extracted treetops were used as seeds in a watershed segmentation to partition a CHM into individual tree polygons based on the topology of its intensity surface. We tested our method on a subset with heterogeneous landscape elements (park, building, and street) in Braunschweig and used tree cadastral data provided by city authorities - for validation. The tree cadastre documents the location, height and crown diameter of each tree based on on-site surveys. With that, we evaluated the performance of our individual tree detection procedure and achieved a commission error of 36.72% and an omission error of 5.41%. A comparison of the cadastral data with the remotely sensed derived parameters results in an R2 of 0.246 and 0.7452 for the crown diameter and tree height respectively. Sentinel-2 data from June 2023 served as the basis for calculating the Normalized Difference Vegetation Index (NDVI), which we initially used as proxy for tree vitality. Additionally, we calculated the percentage of fraction tree cover per Sentinel-2 pixel. We found that pixels tree cover correlates with the average NDVI values, but individual observations are often influenced by the tree's understory, resulting in higher NDVI values. In the next step, we will evaluate NDVI time series for the vitality analysis of urban trees and investigate pixels spectral components in more detail.
Shrestha et al. (Mon,) studied this question.