Flood inundation mapping in SAR images based on nonlocal polarization combination features

Accurately and promptly monitoring of flood inundation is crucial for disaster relief and loss assessment. Although some advancements have been achieved in flood mapping based on synthetic aperture radar (SAR) data, the current research usually extracts flood information at a level such as a single temporal image, single polarization or local neighborhood features. In addition, these methods often rely on learning samples, geographic features, or priori hypotheses (e.g. the histogram of the image has bimodal characteristics), which limits their practicality in large-scale rapid automated mapping. Therefore, we propose an automated flood inundation mapping method for multi-temporal SAR images by using a nonlocal polarization combination feature (NPCF). In this study, we first construct multiple polarization combination features of Sentinel-1 data, and then select the optimal combination feature suitable for flood mapping. On this basis, the different information, before and after the flood disaster, is extracted based on NPCF. Finally, the flood inundation area is obtained through an optimal automated threshold method. Experimental results show that: 1) (VV + VH)2 performs better than the other dual-polarization combination features and the single polarization (VV or VH) feature in flood inundation mapping; 2) NPCF performs better than 10 other advanced flood inundation mapping methods. Additionally, NPCF was further used for rapid mapping of two flood events, and the results showed that Kappa was greater than 0.85. Overall, NPCF achieves automated mapping of flood inundation areas, and can provide timely and reliable disaster information for disaster relief in flood events with high timeliness requirements.