Flood Extent Estimation Using Remote Sensing

Abstract

Using the example of flood in eastern Croatia in May 2014, the paper compares three flood mapping methods: reflection in near infrared (IR) interval of the electromagnetic spectrum (EMS) (one band method), and two multiband methods, namely Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI). Input data for all three methods are satellite images recorded with the sensor carried by the Landsat 8 satellite. Four data sets were analysed: one recorded before the event, one during the flood and two images recorded after the event. Estimation based on the one band method resulted in the area smaller than the area estimated using the multiband methods. Such results may be explained with the complexity of the spectral signature of the shallow flood waters containing suspended load, which alters the near IR reflection from water surfaces and classifies them as non-water surfaces. Combining the various spectral bands in the multiband methods that impact is compensated and classification of water and non-water surfaces is more precise.