On the Integration of Intensity, Interferometric Coherence and Polarization Diversity in Flood Detection from Long Stacks of Multi-Frequency SAR Data Through a Bayesian Framework

Bayesian estimation of posterior probabilities for the presence of floodwaters, coupled with accurate time series regression methods, show good performance in the monitoring of inundations at high temporal and spatial resolution from long stacks of synthetic aperture radar (SAR) data. We report results on the integration of SAR intensity and cascaded InSAR coherence time series in different polarization channels within a Bayesian framework. The method is being tested over sites in both northern and southern Italy, with X- and Cband SAR data. The results indicate some advantage in using more than one independent channel in the Bayesian inference for some types of land cover, in terms of area under the curve (AUC) when compared to independent flood maps acquired over known events. Stacks of surface water confidence levels computed over both test sites show promising characteristics, both on agricultural and coastal areas.