I’ve processed inSAR coherence images using the SNAP software on Sentinel SLC images and was quite pleased with the results, until I compared them to the ASF coherence maps using the same Sentinel images (from the ASF Earthdata search site). The ASF images are much lower resolution but the coherence is generally much higher than mine. Does anyone know what they’ve done to do this?
I’ve attached a picture of the graph of the processes I used in the SNAP software. I tried making the coherence window as big as possible (90) and multi-looked the output but it didn’t improve the coherence, it just reduced the resolution.
and here is the ASF processed image with its histogram of values. (They haven’t masked out the sea, which explains the little hump around 0.3 on the histogram.)
I was hoping to use the ASF coherence images alongside my own processed SNAP images, but I don’t know which is right, or why they are different - Can anybody help???
Yours Hopefully,
Unless you want to include coherence in mapping, filtering can artificially increase the coherence value and give a false estimation of the phase quality. Of course you can achieve high coherence values (as shown above) by using large windows and strong filters but the interferometric quality of the phase won’t improve. You simply get ‘nicer’ coherence images.
Which one you use (range/azimuth or Goldstein) depends on what you want to do with the data and also the incidence angle of your image which determines the ratio between range and azimuth resolution. I personally rarely use range/azimuth filtering because it seems more complicated to me to find suitable parameters.
Thanks for your thoughts - I’ve decided to not use the ASF imagery, as I’m not sure how they’ve processed it, and I’ll stick to my self processed ones using the SNAP software.
In this case, multi-looking the coherence will reduce the spatial resolution, but probably give you more homogenous areas to work with. Makes sense to me. Depending on the scale of your agricultural fields, reducing the spatial resolution to 20 or 30 meters with multi-looking could not only reduce the amount of data, but also enhance the contrast between areas of high and low coherence.
yes, the number of looks determines the output resolution. You have to find a balance between effective multi-looking and the lowest tolerable resolution.
Hello! Forrest here from the ASF SAR development team. An increase in coherence when comparing high-resolution SNAP outputs and ASF HyP3 product is actually expected. When you multi-look SAR imagery prior to interferogram generation, you introduce a positive coherence bias (see Touzi et., al. 1999: “Coherence estimation for SAR imagery”). If you are using SNAP to create interferograms with less multi-looking, then it’s expected that the HyP3 interferograms will have a higher coherence.
For those who want to dive deeper into the ASF InSAR workflow (which uses GAMMA software under the hood), here our processing code can be found in GitHub in the hyp3_gamma repository.
If you have further questions, feel free to reach to ASF at uso@asf.alaska.edu
