SAR and Social-Media-Based Change Detection With Dual-Threshold Fusion for Flood Inundation Mapping

As one of the most destructive natural disasters, floods are increasingly frequent and severe due to urban development and population growth. The threshold-based method is widely acknowledged as an effective approach for detecting flood extent in synthetic aperture radar (SAR) imagery. However, determining the accurate threshold value poses a significant challenge. During periods of flooding, social media (SM) data posted by users provide a wealth of real-time information for flood inundation mapping (FIM) purposes. This study presents a new semi-automatic threshold determination method called SAR and social media-based dual threshold (SSM-DT) for FIM. SSM-DT aims to improve accuracy by avoiding traditional method inaccuracies through a semi-automatic approach. Integration of SM data enhances real-time flood situation monitoring, enriching FIM comprehensiveness. Firstly, the SAR images are processed and analyzed using change detection techniques to identify potential flood inundation areas. Simultaneously, a deep learning model is utilized to classify and filter SM data, enabling the retrieval of real-time flood-related information. Finally, the flood information obtained from both SAR images and SM data is fused together to generate a more accurate and comprehensive flood inundation map, leveraging the complementary nature of these two data sources. A case study focused on the extensive flooding caused by Hurricane Harvey in Houston in 2017 is discussed. The results demonstrate that the proposed method can provide near real-time depiction of flood extent, which is crucial for mitigating economic losses and minimizing casualties.