Guided Super-Resolution of Digital Elevation Models with Diffusion-Based Image Generators

High-resolution digital surface models remain scarce and expensive to acquire, even as detailed aerial and satellite optical imagery has become widely accessible. This work bridges that resolution gap by using denoising diffusion models guided by spectral images to upscale elevation maps tenfold, from 5 meters to 0.5 meters. By transferring structural boundaries and rooflines directly into elevation data, the approach recovers geometric fidelity that standard interpolation fails to reconstruct.