Ditch mapping for Switzerland by using semantic segmentation

In Switzerland, with its steep topography and frequent precipitation, forests face major challenges in terms of soil erosion, flooding and drainage. Ditches within forests are therefore essential for controlling excess surface water, lowering the groundwater level in wet areas of forests and helping to prevent landslides, flooding and the degradation of forest roads and infrastructure. With increasing awareness of environmental sustainability, ditch systems are being used to balance water flow without overly disrupting ecosystems. Proper ditch detection and maintenance within forests is critical to protect the environment and infrastructure, especially in the context of climate change, which brings more intense rainfall  vents and poses risks to forest ecosystems. 

The project focuses on the ditch detection for the entire country using semantic segmentation based on UNet. The input data is the SwissALTI3D DEM data with a resolution of 0.5 meters and a corresponding drainage label image. In addition, the high pass median filter and impoundment size index mean feature are also implemented as input features. On going experimental results have illustrated the potential effectiveness.