Since I have searched in this forum and found no topics that demonstrate a clear workflow for ALOS-2 SLC stripmap data interferometric processing, I think we should make it clearer here about the processing workflow with ALOS-2 PALSAR-2 L1.1 SLC stripmap images, using SNAP. I really hope this topic could end with a clear and common agreement that would help others handling ALOS-2 stripmap interferometry by SNAP.
Should it be like this?
(1) Open product (import ALOS-2 CEOS file)
(2) Coregistration
(3) Interferogram Formation
(4) Coherence Estimation
(5) Topographic Phase Removal (SRTM 1sec)
(6) Multilook
(7) Goldstein Phase Filtering
(8) Snaphu export
Your workflow looks good to me. It is better to do the topographic phase removal before multilook and filtering. You might want to remove the topographic phase before the coherence estimation, but I donāt think it will make much difference because ALOS-2 satellite has small baselines, unlike ALOS satellite.
I would recommend doing the topographic phase removal after the interferogram formation. The ALOS satellite had large variations of baseline, so it is important to remove the topographic phase as early as possible in the processing.
Hello, I am new to using the step forum and I am not experienced in InSAR processing.
I am using SNAP 8.0 with all updates.
My question is also about InSAR processing a single pair of ALOS-2 L1.1 SLC images.
The images are ALOS-2 stripmode (SM) collected on ascending paths on 16 Jan 2018 an 27 March 2018 with a 6.5 m perpendicular baseline. The images are of a mountainous, forested Pacific island. I read a few earlier posts concerning ALOS InSAR processing and from those changed some processing inputs. Processing steps followed
1- Open product (import ALOS-2 CEOS file),
2- Coregistration using defaults except for ābicubic_interpolationā in the stack creation and 2000, 10000, 20000 gcpās in the cross-correlation. Using 2000 gcps, the two images were not aligned having a visible shift. Following an earlier post the gcpās to were increased to 10000 then 20000. the 10000 and 20000 looked similar.
3- Interferogram Formation āflat earth degreeā was changed to 8 and āorbit interpolation degreeā to 5.
4- Topographic Phase Removal used defaults and SRTM 1sec.The change was very slight.
5-I tried to run the Coherence Estimation but there was a very quick message flash that couldnāt be read and then nothing happened.
Below is shown the coregistration alignment, master in red, slv in green, within the full scene, and zoomed into the island. Next is the phase and the coherence in the āinterferogram formationā without subtracting topographic phase, and the topographic phase removal.
My questions are (1) does the alignment look reasonable? (2) why is the coherence so low (mostly <0.3) and patchy? and (3) in āCross-correlationā there is an option for āCoherence based registration.ā Should that be used? (4) I am wondering if alignment is the cause. Can the image be subset pre Coregistration? Should that help?
thank you for any help!
if the yellow part is the island, yes. Althouth all areas outside the island drastically changed between both acquisitions.
The temporal difference is quite long and these areas probably undergo much changes. Where exactly is your study area located? What is the surface cover of the island (and the areas outside)? You can only retrieve high coherence and usable phase information from coherent scattering. If the area is affected by growing vegetation or inundation, InSAR becomes rather ineffective.
The type of coregistration will only affect the coherence if it is really bad. Images can be well coregistered and still have low coherence because of temporal decorrelation. So you can try another method, but it is unlikely to increase your quality. The black line however looks suspicious. Please try nearest neighbor resampling of the slave images instead of bicubic.
The accuracy of coregistration has limited impact on the coherence. As I said, you can try a different resampling to keep the values as original as possible, but changes in surface scattering mechanisms are the real problem, Iām afraid.
thank you ABraun for your reply, information and suggestions
The temporal difference is quite long and these areas probably undergo much changes.
āmy perception was that a 3-months time difference was within the range of L-band providing a workable InSAR product.
Where exactly is your study area located? What is the surface cover of the island (and the areas outside)?
āits a mountainous island in the Pacific (Pohnpei in the Federated States of Micronesia). it is heavily forested, however, from my scanning of the literature, I was thinking L-band and a low baseline could still provide reasonable coherence. In the large part, the forest should be mature and unchanged. The outside area is ocean.
So you can try another method, but it is unlikely to increase your quality. The black line however looks suspicious. Please try nearest neighbor resampling of the slave images instead of bicubic
ā thanks, I did change to nearest neighbor without changing any of the other parameters. The relatively higher coherence extents decreased, overall the product quality decreased.
it is true that L-band is less sensitive to temporal decorrelation compared to X-band or C-band, but in cases of dense vegetation you will still not retrieve coherent scattering when images are several days (or even weeks) apart. It was shown in studies that decorrelation occurs already when canopies move by wind. So repeat pass differential interferometry over forest is nearly impossible.
From documentation by the satellite operator. Or you could just process the data as PALSAR-1 data should be aband d see what happens. I would guess deskewing might since the coherence looked so weird.
mengdahl, sorry for taking so long to get back. I did import the two ALOS-2 scenes as ALOS PALSAR as you suggested. Did the coregistration, could not Deskew the coregistered images (the procedure would not accept ALOS-2), and created the interferometric product. although there were differences compared to the earlier results, the coherence was low and phase still completely without pattern. I did not see a d band. thanks for the suggestion.
hi mengdahl, I do have access to another InSAR pair to test and I will. sorry, I am pressed with a deadline and will try another set and get back as soon as possible. Thank you for the article. I did look it through and it seems that the band mode overlap problem with ALOS-2 Stripmap was solved for post 1 Jun 2015 images. The AlOS-2 images I am trying to process are post 2018. thanks mengdahl.
According to āALOS-2/PALSAR-2 Level 1.1/1.5/2.1/3.1 CEOS SAR Product Format Descriptionā by JAXA (2012), āL1.1 image is focused onto zero Doppler direction.ā Therefore, no deskewing is needed.
hi mengdahl
I did process another pair of ALOS-2 SM3 slc images, 15 March 2020 and 21 June 2020. the images, however, are from an orbit slightly more west but still ascending and same right look direction. The temporal separation is about the same as the images I previously posted as well. the only non default processing options I changed were the use of bilinear in the CreateStack, 10,000 gcpās in Cross-Correlation, Cubic convolution in Warp, 1001 estimation points in the Interferogram Formation, and the 1Sec HGT in the Topographic Phase Removal. The Goldstein filter process came after the topographic phase removal.
Iāve attached four screen captures, first the red-green overlay of the mst and slv HH intensity and second the HH coherence. For the third and fourth images, I used output jpgās for the HH phase and HH phase with Goldstein filter to highlight the filter output. From the paper you posted earlier, I believe problems with ALOS were in the 2014 to 2015 period, although banding shows up on the posted filtered phase image. Visually, the spatial alignment looks good to me, however, as ABraun suggested, the steep terrain and vegetation cover are probably the reason for the low coherence throughout most of the image even with just a 3 month time separation. The banding in the filtered phase image though seems to be more questionable, although, me being a novice at InSAR, that had to be pointed out to me.
thank you for all you help and suggestions.
