Flood Extent Mapping From Time-Series SAR Images Based On Texture Analysis And Data Fusion

The expansion and contraction of floods can be mapped using multi-temporal satellite observations. This method examines the observations of the moving boundary of this flood edge. In this paper, flood extent means the distance between the river centre and either the right/left side of the flood edge, along a given cross section. All cross sections are leaning at a 90 degree angle to the river’s axis. The flood extents for both sides (left and right) of river centre are measured separately. The idea of this method is continuous observations of the flood extents can indicate the location of high flood defenses. This method talks about establishing stage-extent relationships for a given cross section (either left/right).

If after measurement, the lateral flood extent expands but the level of water does not, it means that floodwaters are increasing within the floodplain. On the other hand, if the flood extent stops increasing but the level of water still continues to increase, it means that there is an impediment to the expansion of the flood. The moment that the flood extent starts increasing perpendicularly after a level plateau it is said that a structure has been overtopped at that point. The method that is being proposed in this paper plans to use this stage-extent relationship, and deeply searches for extended horizontal lines at each cross-section, to deduce the position of the embankments of flood.

The method explained here contains two parts; Part 1 talks about using the images taken by SAR to derive fluvial flood maps. Part 2 talks about implementing the stage-extent plot and the derived embankment location for designed cross-sections.

Part 1

SAR images have a large distinction between smooth water surfaces and rougher land areas, this textural difference makes the data retrieved from it a very useful means for mapping flood extent. Sentinel-1 SAR data was used because it has a large spatial coverage, it is widely available and it has shirt revisit times. For this study, pre-processed and geo-referenced Level 1 Sentinel-1 data were downloaded, after that, the SAR products were calibrated, terrain-corrected and clipped. A Lee Sigma filter option was applied to the images that had clipped to tackle speckle effects. To limit the loss of data from the images, a 3 by 3 target window was set as the filter.

Part 1a uses a method called grey-level thresholding which is used while developing flood mapping algorithms. For part 1b, a spread operation for creating a fluvial flood map was used. It was initiated at the river centre.

Part 2

In the stage-extent plots, plateaus are used to mark the location of the embankment. For this, other methods namely plotting discharge data against extent, stage-extent is used here for the sake of consistency. Using SAR images, the gauge of flood extent at discreet time-steps is plotted against stage level for said time-steps. The same process is used over again for all SAR images that are available with flood extent plotted against the stage. Ordering the stage by increasing magnitude instead of using date was necessary due to the fact that during a flood, stage does not rise and fall in the course of time efficiently.

To get the positional error of the predicted embankment location, reference data was used on the location of the embankment for computation and also for justification purposes. The positional error was calculated as the absolute error between the predicted embankment location and the reference location with the unit of meters along each cross section and it is also calculated as the median absolute error over all cross sections at each site. It was calculated as the median absolute error because errors were varying and mean results became confusing because of some external factors.

The values of the selected water threshold gotten form the SAR imagery determine the binary water map and the associated fluvial flood map. Because of this, the method with a range of threshold values was tested and the positional error was evaluated as a function of the selected threshold. The positional error that was obtained in this sensitivity test are given as the signed error, rather than the absolute error, to highlight any under/over estimation of the embankment location.