Flood Detection in Built-Up Areas Using Interferometric Phase Statistics of PALSAR-2 Data

This study investigated a method of flood detection in built-up areas by L-band synthetic aperture radar (SAR) images, using interferometric phase statistics instead of the widely used coherence. Simulation and actual SAR data analysis showed that phase statistics are more accurate than coherence is, especially when there are many point-like scatterers. Accuracy was further improved by using an interferometric phase filter. The flood detection accuracies obtained by the phase statistics were κ = 0.535 (0.304 by coherence) in the 2015 flood in Joso and κ = 0.679 (0.609 by coherence) in the 2018 flood in Mabi. These results showed that coherence should be used with great care with point targets and that the phase-only statistics can be more robust in such cases. The proposed method is expected to apply to other types of disasters where coherence has been used previously.