Flood Mapping with SAR and Multi-Spectral Remote Sensing Images Based on Weighted Evidential Fusion

Synthetic Aperture Radar (SAR) and Multi-spectral (MS) remote sensing images are commonly used for flood mapping. SAR images can provide valid backscattering measurements of inundated areas through cloud cover, while MS data is able to monitor the spectral changes of ground surface, but usually affected by clouds. The complementary characteristics of the two data indicate the potential of their combining application for flood monitoring in emergency. This paper proposes a novel weighted evidential fusion method to take full advantages of the SAR and MS data for change detection during the flood. First, pre-processing and classification are performed with the SAR and MS data, independently. Second, a modified PCR6 rule for evidential fusion is proposed, which introduces the confusion matrixes to calculate the weight of evidences so that the conflicting degree in the fusion process can be reduced. Then, the flood inundating, standing and receding patterns are identified, which can be used to describe the flooding process in details. Practically, the proposed method is applied to flood mapping of the Typhoon Rumbia in 2018, in Shouguang City, China. The experiments show that the proposed fusion scheme efficiently use both of the SAR and MS data, and improve the flood mapping accuracy.