Near-real-time non-obstructed flood inundation mapping using synthetic aperture radar

• Published in: Remote sensing of environment Vol. 221; pp. 302 - 315
• Main Authors: Shen, Xinyi, Anagnostou, Emmanouil N., Allen, George H., Robert Brakenridge, G., Kettner, Albert J.
• Format: Journal Article
• Language: English
• Published: New York Elsevier Inc 01.02.2019; Elsevier BV
• Subjects: algorithms; artificial intelligence; Automation; Classification; Cloud cover; Coarsening; Data processing; Earth orbits; Flood inundation mapping; Flood mapping; Floods; Harvey; High resolution; Land cover; Learning algorithms; Machine learning; Mapping; Missions; Near real-time system; Nepartak; Radar; Radarsat; Real time; Remote sensing; Rivers; Satellite observation; Satellites; Sentinel; statistics; Synthetic aperture radar; Flood inundation mapping; Nepartak; Harvey; Sentinel; Synthetic aperture radar; Near real-time system
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2018.11.008

In the event of a flood disaster, first response agencies need inundation maps produced in near real time (NRT). Such maps can be generated using satellite-based information. In this study, we developed mapping techniques that rely on synthetic aperture radar (SAR) on-board earth-orbiting platforms. SAR provides valid ground surface measurements through cloud cover with high resolution and sampling frequency that has recently increased through multiple missions. Despite numerous efforts, automatic processing of SAR data to derive accurate inundation maps still poses challenges. To address them, we have developed an NRT system named RAdar-Produced Inundation Diary (RAPID). RAPID integrates four processing steps: classification based on statistics, morphological processing, multi-threshold-based compensation, and machine-learning correction. Besides SAR data, the system integrates multisource remote-sensing data products, including land cover classification, water occurrence, hydrographical, water type, and river width products. In comparison to expert handmade flood maps, the fully-automated RAPID system exhibited “overall,” “producer,” and “user” accuracies of 93%, 77%, and 75%, respectively. RAPID accommodates commonly encountered over- and under-detections caused by noise-like speckle, water-like radar response areas, strong scatterers, and isolated inundation areas—errors that are in common practice to ignore, mask out, or be filtered out by coarsening the effective resolution. RAPID can serve as the kernel algorithm to derive flood inundation products from satellites—both existing and to be launched—equipped with high-resolution SAR sensors, including Envisat, Radarsat, NISAR, Advanced Land Observation Satellite (ALOS)-1/2, Sentinel-1, and TerraSAR-X. •We built a fully automated system to retrieve inundation maps from high resolution SAR images.•Statistical, morphological and machine learning approaches are integrated.•The system extensively uses multi-source remote sensing ancillary data.•The system can detect both connected and isolated open flood areas.•The system shown reliable performance in state scale during multiple flood events.