Towards high-precision flood mapping: Multi-temporal SAR/InSAR data, Bayesian inference, and hydrologic modeling

This article addresses the application of Bayesian Networks (BNs), to perform data fusion of SAR intensity, InSAR coherence imagery and ancillary data to detect flooded areas. Results show the advantage of integrating heterogeneous sources of information (satellite, topographic, land cover, hydraulic modeling) in order to reduce uncertainties in the mapping of the presence of water on different land cover types, e.g. on agricultural areas, where the presence of vegetation may produce backscatter/coherence flood signatures which tend to confuse automatic classifiers based on simple thresholding approaches.