Deriving Digital Surface Models from Geocoded SAR Images and Back-Projection Tomography

Digital surface models (DSMs) are sets of elevation data of the Earth's surface, useful for applications such as urban studies and height estimation of buildings. They can be derived from a set of synthetic aperture radar (SAR) images acquired in an interferometric or tomographic configuration. Each image acquisition is usually focused in radar geometry. In this work, we present steps required to derive DSMs from SAR single-look complex (SLC) products focused in map geometry (geocoded). We modified existing tomographic reconstruction techniques to be able to operate with geocoded SLCs and extended methods to operate with 3-D geocoded SLCs. The performance analysis showed that methods using 3-D geocoded SLC products yielded DSMs with fewer outliers and retained more information of the illuminated area, with a cost of higher computational complexity. Compressive sensing methods using 2-D geocoded SLCs can be a good alternative due to their comparatively moderate computational complexity.