A Novel Fully Automated Mapping of the Flood Extent on SAR Images Using a Supervised Classifier

Abstract

When a populated area is inundated, the availability of a flood extent map becomes vital to assist the local authorities to plan rescue operations and evacuate the premises promptly. This paper proposes a novel automatic way to rapidly map the flood extent using a supervised classifier. The methodology described in this paper is fully automated since the training of the supervised classifier is made starting from water and land masks derived from the Normalized Difference Water Index (NDWI), and without any intervention from the human operator. Both a pre-event Synthetic Aperture Radar (SAR) image and an optical Sentinel-2 image are needed to train the supervised classifier to identify the inundation on the flooded SAR image. The entire flood mapping process, which consists of preprocessing the images, the extraction of the training dataset, and finally the classification, was assessed on flood events which occurred in Tewkesbury (England) in 2007 and in Myanmar in 2015, and were captured by TerraSAR-X and Sentinel-1, respectively. This algorithm was found to offer overall a good compromise between computation time and precision of the classification, making it suitable for emergency situations. In fact, the inundation maps produced for the previous two flood events were in agreement with the ground truths for over 90% of the pixels in the SAR images. Besides, the latter process took less than 5 min to finish the flood mapping from a SAR image of more than 41 million pixels for the dataset capturing the flood in Tewkesbury, and around 2 min and 40 s for an image of 19 million pixels of the flood in Myanmar.

Patch Similarity Convolutional Neural Network for Urban Flood Extent Mapping Using Bi-Temporal Satellite Multispectral Imagery

“…Unlike pixel-based classification for flood mapping [6][7][8]12,39,40], this study investigated image patch based flood mapping similar to the study in [3]. Major motivations include: (1) reducing the impact of heterogeneous image background over urban area, which is challenging for pixel-based classification; and (2) accelerating human annotation of training samples since pixel-wise labeling would be much more time-consuming and labor-intensive.…”

Decision Support on Monitoring and Disaster Management in Agriculture with Copernicus Sentinel Applications

“…Generally, in the NDWI images, water areas correspond to higher pixel values compared to the remaining areas. In case of Sentinel-2 images, the positive values threshold cannot be adopted, since it results in the misclassification of building rooftops, shadows, dark objects, or asphalt, confused with water [28,51]. In order to eliminate these commission errors, the threshold for Sentinel-2 NDWI images is 0.2, which in practice means that pixels with value greater than 0.2 are classified as 'water'.…”

“…Examples in the literature include the use of satellite data for mapping the extent of a burned area using optical and thermal sensors of different spatial resolution [21][22][23][24] and many methods are particularly developed for Sentinel data (e.g., [25,26]). Various approaches were developed for example for the automatic flood detection [27,28], and the free and open access policy of Copernicus Sentinel-1 triggered the development of more [29][30][31]. An important asset for monitoring disasters of the Sentinel-1 constellation is the frequency of data acquisition (six days) and the fact that they can be downloaded within an hour and 24 hours of reception depending on the severity of the event.…”

Decision Making on Disaster Management in Agriculture With Sentinel Applications

“…Concerning flood events, the synthetic aperture radar (SAR) data are the key information input for the mapping the extent of the flood, as cloud coverage hinders the usage of optical data in these cases. There are various approaches developed for the automatic flood detection (Hess et al 1995, Martinis et al 2015, Benoudjit et al, 2019) and many emerged after the free and open access to Copernicus Sentinel-1A and -1B data. A major advantage of the constellation is the systematic data acquisition every 6 days.…”