Green roof detection

First, an NDVI raster was generated from digital orthophotos (DOP) (target resolution 0.5 m). Analogously, a Digital Surface Model (DOM) was first generated from the point clouds and then an nDSM was generated using the NRW-DEM (target resolution 0.5 m). The study area was limited to house perimeters (the source of the house perimeters is variable in principle, here the OpenNRW perimeters were used as a starting point, the buildings newly detected in the previous project can also be used).

Based on this, a region growing segmentation was calculated based on the NDVI and nDSM grids. Statistics (mean, range, standard deviation, etc.) of these grids were assigned to the created objects.
An NDVI threshold was then determined, which can be used to extract candidate vegetation objects on roof surfaces. This threshold can be set manually or selected automatically from vegetation covered areas (from land use mapping – FNK). The vegetation objects were then compared to the NDVI / nDSM statistics of the surrounding roof area to minimize false detections. The remaining vegetation objects were intersected with the house perimeters and the latter were assigned the binary attribute greened (yes / no).

Optionally, a user-defined threshold can be used to set a minimum percentage of green roof area per building. The analysis was conducted for both spring and summer aerial surveys, as well as the combination of both time points, to derive recommendations for appropriate input data.