Flood Inundation Mapping From Sentinel-1A GRD Images

Most of the authors use the change detection technique (i.e., difference between pre- and post- flood event) to map the inundated flood area, however in the manual provided by UN-SPIDER and modified by the Alaska Satellite Facility, they use only the one scene for the day of the flood event. Could you please explain which method is more beneficial, particularly if the GRD images are available for the day of the flood. I would appreciate mentioning the citation references if possible.

Please take a look at this post it is helpful

It is possible to use the technique is mentioned in the post shared above , and process the pre and post flood event , then it is possible to use for example the band math to subtract the inundated area,

“Post”_"Pre "

Thank you Dr. Falah for the helpful comment. However, I am still concerning about using the GRD image for only the day of the flood as mentioned by UN-SPIDER, is it enough without applying change detection.

In general, it is possible, but basically in case you’d apply let’s say change detection technique only for the day of flooding, then you’ll get only the final result of inundated area. It depends on your goal, I think the manual is an instruction and example, it is not firm application for all cases, but would you please to share the manual,

Please find the below link, they use only SAR image for the flood day, no change detection method was applied.

I think a flood map can be generated with a single image if it was taken during the flood.
However, if there are areas in the image with very low backscatter (smooth surfaces, permanent water bodies, ect.) you won’t be able to distinguish them from the actual flood. Consequently, over-estimating the flooded areas is very likely to happen.
If that is okay to you, change detection is not needed. But whenever the actual extent of the flood matters, an image without flood is required for comparison.

Thank you ABraun for your helpful reply. I am wondering if you can provide me with scientific reference to detect flood from one scene during the day of the flood. Most of the papers I have read talking about detecting water bodies not exactly flood from a single scene.

Try to classify flood using thresholding technique and then mask out the existing waterbodies using water mask(ex. SRTM warer mask) availabe. Finally, you will be left out with only flooded area.
But you should take care while choosing the threshold and follow all the preprocessing steps required for a GRD product before choosing the threshold and then applying it to the same image.

1 Like

Hello Dr. Falah;

Both Radar images for pre and post flood events should be in binary format (1,0). Is it Ok to use band math to to subtract the inundated area, “Post”_“Pre”. Please advice if it will work or please suggest another method.

It’s possible, but I think you could also create the simple RGB, from the two post and pre, inundated area, to identify the affected flooded area,

Thank you, I am going to try it