Sentinel-1 for urban area monitoring — Analysing local-area statistics and interferometric coherence methods for buildings' detection

This study provides a comparison of built-up area detection methods from Sentinel-1A SAR images. The selected methods are based on local area statistics (speckle divergence and difference between the local mean and local median) and repeat-pass interferometric coherence. The dataset comprises VV and VH polarization SAR images acquired during ascending and descending orbits over two study areas in Estonia, and ancillary map information. Our results show that combining measurements from both orbits and both polarisations yielded the highest accuracy for built-up area detection. In the case of methods based on local statistics, classification accuracy increased twofold, up to 72% (user accuracy, mean-median method). For all methods, the results differed between the study areas, indicating the existence of site specific effects. The methods yielded lower classification accuracy when applied to a more rural study area.