Flood delineation based on time series change detection and region growing of SAR images

Authors: Jiayong Liang*, The Ohio State University, Desheng Liu, The Ohio State University
Topics: Remote Sensing, Hazards, Risks, and Disasters
Keywords: SAR, flooding, change detection
Session Type: Paper
Day: 4/12/2018
Start / End Time: 1:20 PM / 3:00 PM
Room: Napoleon B1, Sheraton 3rd Floor
Presentation File: No File Uploaded

Flooding is an influential disaster and SAR imagery is effective to map flooding during adverse weather and illumination. Improved spatial and temporal resolution of SAR images are made possible by the recent Sentinel-1 mission. Due to SAR imaging mechanism, robust interpretation usually requires specific information for the region of interest, so flood mapping method need to be tailored in different places and scenarios. Aiming at global operational observation, this paper uses a series of SAR backscatter intensity images acquired at different time to detect flooding area. An image pair is first converted into modified saturation for both polarizations. The product of the two polarizations’ saturation is used to identify flooding event by co-occurrence matrix. The correlation of the co-occurrence matrix measures how correlated a pixel is to its neighbors over the whole image and this correlation would increase dramatically when change happens. With such activated indicator, a region growing procedure is conducted to detect flooded region in the corresponding SAR image. Gaussian mixture model is used to cluster the SAR image with three classes, including intensity decreased, increased, and no obvious change. The seed for region growing algorithm is selected when the pixels are with obvious decreased backscatter. The pixels with smaller backscatter than the decreased cluster mean are used for seeds. After region growing from the selected seeds with backscatter threshold of half variance of the cluster, resulting inundation extent is delineated, with water producer accuracy of 85% (either polarization) and water user accuracy of 91% (both polarizations).

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login