Generation of digital terrain models from digital surface models using a watershed transformation approach

The basis of Urban Digital Twins are 3D city models generated from airborne or spaceborne remote sensing imagery. These remote sensing data allows the extraction of digital surface models (DSMs) representing the absolute geographical height of each pixel together with orthorectified image mosaics. But for the derivation of a city model also the height of the urban objects and the height of the ground – the digital terrain model (DTM) – is needed. So in a first step there has to be a method to derive the digital terrain model (DTM) from the generated digital surface model (DSM). The extraction of DTMs from DSMs is already a basic task for many decades and there are many approaches but still there is no general fully satisfying solution for this demand. In this paper we present a novel method for deriving a DTM from a provided urban DSM based on the inverted watershed-transformation. The watershed-transformation normally fills up sinks in a DSM up to a ridge-line where two sinks meet. So the result is a segmentation of a DSM to distinct sinks. Using the inverted watershed-transformation delivers a segmentation to single elevated objects. The boundary points where each of these elevated segments meet are taken as possible candidates for ground areas. But such boundaries may also lie for example between adjacent roofs. Using a dynamically determined threshold in the height distribution of each watershed segment gives a possible ground candidate area of the segment. Merging all ground candidate areas and taking only larger areas with special properties into account gives all ground candidates. Keeping only the heights of these areas as real ground and filtering and filling them gives the final DTM. In the presented paper we describe the method in detail and apply the method to different DSMs of rather flat and hilly urban and non-urban environments, discuss the results and give finally an evaluation of the quality of the results.