Classification algorithms for Synthetic Aperture Radar (SAR) imagery that are based on statistical modelling typically require the parameterization of distribution functions of backscatter values of different land cover classes to classify a scene. To parameterize accurately the distribution functions of the individual classes a sufficient number of pixels is needed and this criterion is not always satisfied, especially for classes occupying only a small fraction of the scene Here we propose an automatic algorithm that aims to map buildings based on the SAR intensity backscattering feature. It makes use of a hierarchical split-based approach that does not fix the size of the tiles a priori but, rather, searches for tiles of variable size where the distribution functions attributed to classes of interest can be parameterized in a robust way. The algorithm has been developed in the framework of the Urban Round-Robin exercise, supported by the European Space Agency (ESA) through the ESA Land Cover Climate Change Initiative (CCI), and tested on Sentinel-1 data from five different test sites located in semiarid and arid regions in the Mediterranean region and Northern Africa.