Automatic Open Water Flood Detection from Sentinel-1 Multi-Temporal Imagery

Many technical infrastructure operators manage facilities distributed over large areas. They face the problem of finding out if a flood hit specific facility located in open countryside. Physical inspection after every heavy rain is time and personnel consuming, equipping all with detection expensive. Therefore, methods are being sought to ensure that these monitored at minimum cost. One possibilities using remote sensing, especially radar data regularly scanned by satellites. A significant challenge this area was launch Sentinel-1 providing free-of-charge adequate spatial resolution relatively high revisit time. This paper presents developed automatic processing chain for landscape from data. Flood can be started on-demand; however, it mainly focuses on autonomous near real-time monitoring. It based combination algorithms multi-temporal change histogram thresholding open-water detection. The solution validated five events four European countries comparing its results delineation derived reference datasets. Long-term tests were also performed evaluate potential false positive occurrence. In statistical classification assessments, mean value user accuracy (producer accuracy) class reached 83% (65%). typically provided flooded polygons same areas as dataset, but smaller size. fact attributed use universal sensitivity parameters, independent location, which almost complete successful suppression alarms.