Deciduous tree segmentation using airborne lidar point clouds and geometric networks: Examples from the Park Sanssouci, Potsdam, Germany
- 1Institute of Environmental Science and Geography, Universität Potsdam, 14476 Potsdam, Germany
- 2Institute of Geosciences, Universität Potsdam, 14476 Potsdam, Germany
- 3Helmholtz Centre Potsdam - German Research Centre for Geosience (GFZ), 14473 Potsdam, Germany
Automated tree segmentation in anthropogenically shaped environments is challenging. Large differences in deciduous tree species composition narrows down suitable solutions. Tree crowns and vegetation structures exhibit wide varieties and trees often form complex tree aggregations. Clear delineation of individual deciduous tree crowns in dense tree stands is difficult, and methods developed for conifer trees may not be applicable. The dense overlapping structures require alternative approaches.
To overcome some of these limitations, we have developed a new method using individual point locations and their neighborhood information. We build a network of connected lidar points with the following steps: (1) We use a k-d tree to derive immediate neighborhoods of points. (2) For each point a vector is generated pointing to the centroid of higher located points within a neighborhood. (3) Each point is then connected to the nearest point above its corresponding centroid. Points which do not connect to higher points based on this rule are identified as endpoints. Since each point is connected to only one higher point if such a point exists in the neighborhood, these connections form a forest of trees network. (4) With adaptive parameters each tree crown is represented by a single component of the network, i.e., each tree crown is represented by a mathematical tree in the network.
Our new method is tested with a dense airborne point cloud (88 pts/m2) collected in 2018 for the Park Sanssouci in Potsdam, Germany. This area of high biodiversity contains more than 25.000 mapped trees from over 75 different species that we use as reference and validation datasets. We demonstrate the successful application of our algorithm and present segmentation uncertainties.
How to cite: Hess, M., Rheinwalt, A., Brell, M., and Bookhagen, B.: Deciduous tree segmentation using airborne lidar point clouds and geometric networks: Examples from the Park Sanssouci, Potsdam, Germany, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-4155, https://doi.org/10.5194/egusphere-egu21-4155, 2021.