Single Tree Detection

Single Tree Detection is a data package from mundialis that records individual trees across a whole area automatically. Municipalities, landowners, urban planners, foresters and authorities responsible for road safety and climate protection need accurate, current data on tree populations for tree registers, their duty to keep roads safe, green-infrastructure planning and climate adaptation. Municipal tree registers, where they exist, are often incomplete or outdated.

The method applies machine learning to aerial orthophotos, LiDAR point clouds and derived surface models. Each detected tree becomes a digital object with trunk coordinates, tree height, crown diameter, circumference and volume, plus an NDVI value per tree that indicates its health status. Optional attributes include the distance to the nearest tree or building, the surface type beneath the tree, and a deciduous/coniferous distinction derived from Copernicus Sentinel-2 satellite time series. At Sentinel-2's 10 m resolution, that distinction is made per group of trees rather than per tree. Repeat runs support automated change detection over time. Results come in common geodata formats or as an OGC web service, and mundialis states that each product can be customised.

mundialis developed the approach for the Regionalverband Ruhr (RVR) with an open-source toolchain built on GRASS GIS, and computed an area-wide dataset for the city of Herne for 2020. Quality was checked against a reference tree register, and the software, results and documentation were handed over to the client. mundialis is certified to DIN EN ISO 9001:2015 and DIN EN ISO 14001:2015.