A Simple Adaptive DEM Void Filling Method Based on Raster Distance Transform

Digital elevation models (DEMs) are indispensable in the fields of photogrammetry and remote sensing. During generating DEMs from multiview satellite images, nonvalues often occur in areas with cloud cover or water bodies, resulting in void areas. The existing interpolation or fusion methods for filling DEMs often struggle with the challenge of ensuring smooth elevation transitions at the seams, while deep-learning-based methods are encumbered by substantial memory and computational demands. To address these issues, this study proposes a simple adaptive DEM void filling method based on raster distance transform to reconstruct a complete topographic surface, using publicly available global DEMs as auxiliary data. First, an initial weight grid is created by performing distance transform on the mask of void areas. Then, the buffer zone and weight map for DEM fusion are adaptively determined by setting appropriate threshold in the initial weight grid. Finally, the target DEM is seamlessly fused with the auxiliary DEM according to the weight map. Comparative experiments conducted across different topographic terrains demonstrate the effectiveness of the proposed method in seamless DEM void filling. It achieves a substantial reduction in root mean square error (RMSE) of elevation differences at seams from 4.27 to 1.47 m, significantly mitigating the staircase effect in visual, and the simulation experiments also prove that this method has a higher terrain fidelity.