Urban floods, especially those in dense built-up areas, present high impacts to resident populations and infrastructure. Real-time geographic extents of flooded areas delineated using remote sensing data and technologies is one of the key information inputs for effective disaster management and rapid rescue response. Images from visible band remote sensors are the most common and cost-effective for the real-time applications. Based on an understanding of the differing characteristics of floodwater and those of urban land surface classes, a robust method has been developed and automatized to extract floodwater using RGB band DNs. The methodology has been applied to delineate flood extent visible in imagery from very high-resolution aerial image data. The methodology development involved rule development, segment- and pixel-based feature analysis, automated feature extraction and result validation processing. The accuracies for the visible floodwater class are above 0.8394% and the overall accuracies are above 0.9668% at both pixel and segment levels for three test sites with diverse urban landscapes.
|