Floodsar: Automatic mapping of river flooding extent from multitemporal SAR imagery

Abstract

Floodsar is an open-source tool for automatic mapping of the flood extent from a time series of synthetic aperture radar (SAR) imagery. Floodsar is unsupervised, however, it requires defining the parameters search space, geographical area of interest, and some river gauge observations (e.g. water levels or discharges) time series that overlap temporarily with the SAR imagery. Applications of Floodsar are mainly in real-time monitoring and elaborating long-time series of historical data. Floodsar features two algorithms for flood extent mapping. The 1D algorithm identifies a flood/no-flood threshold in one SAR polarization. The 2D algorithm performs clustering on two SAR polarizations at the same time. Floodsar chooses the optimal threshold or the composition of clusters by maximizing the correlation between the resulting flood area and the river gauge observations. Floodsar was tested on three case studies with different land uses to illustrate its performance. The 2D algorithm performed on average the best with the average kappa=0.78, yet the less complex 1D/VV algorithm obtained similar results.

Keywords

• Flood extent
• SAR
• Thresholding
• Classification
• Sentinel-1