Flood mapping of Danube River at Romania using single and multi-date ERS2-SAR images

详细信息	查看全文 | 推荐本文 |

• 作者：T.Y. Gan^a ; ^{tgan@ualberta.ca} ; F. Zunic^b ; C.-C. Kuo^a ; T. Strobl^b
• 关键词：Flooding of Danube River ; ERS2-SAR images ; Permanent water ; Flooded area ; Dry land ; Single-image un-supervised ; contextual ; PCA and Isodata classifications
• 刊名：International Journal of Applied Earth Observation and Geoinformation
• 出版年：2012
• 期刊代码：91_03032434
• 类别：ear
• 出版时间：August, 2012
• 卷：18
• 期：Complete
• 页码：69-81
• 文件大小：4067 K

摘要

Several flood mapping classification techniques, applied to single-date and multi-date SAR images of ERS2, based on the Danube River flooding of 2006 in Romania are compared, as part of an effort to explore the feasibility of mapping flooded areas by SAR images acquired through radar sensors. Among 7 SAR images analyzed for the same study site located around Bistret of Romania, several represent 鈥渄ry鈥?and several 鈥渨et鈥?conditions, where the latter represent the major Danube flooding event of 2006. The images were classified into (1) permanent water (Danube River and lakes), (2) flooded area, and (3) dry land, using single image, pixel-based classification, frequency-based contextual classification, and principal component analysis (PCA) combined with Isodata classification. The flooded areas delineated from the above procedures for the study site at Bistret are visually compared with that of Landsat-TM images and MODIS mosaic and digitally compared with referenced flooded area produced by the DEM data of SRTM. Apparently there is no one technique that is clearly better partly because of the nature of SAR data (radar echoes) and partly because of data noise even though the images were first subjected to speckle filtering and geometric corrections, and partly because SAR images could appear dark not only because of flooding but also because of smooth surfaces, target sizes, etc. However, if multi-date SAR images of both DRY and WET (flooding) conditions are available, it seems that PCA combined with the Isodata classifier would give better defined flooded areas of the Danube River than the simple single image, pixel-based classification or the contextual classification.