The Order of DEM Creating Steps

I am trying to create a DEM over my AOI. After reading plenty of topics,

I have obtained a pair of S1A and S1B data. The baseline is 143.48 which is the longest I could find. As I read, it should be between 150 and 300 and around 70% of the values in the coherence histogram should be above 0.4. This is the histogram of my recent and best attempt.

After applying the steps below;

  1. S1 TOPS Coregistration
  2. Creating Interferogram
  3. Debursting
  4. Goldenstein Phase Filtering
  5. Multilooking
  6. Exporting to Snaphu, unwrapping and importing back
  7. Phase to Elevation
  8. Range-Doppler Terrain Correction

This is the elevation model I got and it’s the best of all my attempts:

I wonder if I applied the required steps in the right order. And I’m not sure whether the incoherencies and distortions distributed around image is caused by that or not having long enough baseline and not very good coherence. I would like to know what else I can do to improve the result.

Another issue is the mismatch of the elevation values and the real values from Google Earth. Is it related to the incoherency as well?

And last, I wonder if there is any way to predict whether the baseline is long enough before downloading the datas.

I know that’s a lot of question, but I really need to get this done correctly in short amount of time. I appreciate all the answers in advance.

Note from @ABraun (23.01.2020): A tutorial on DEM generation with Sentinel-1 data based on this area is now available: http://step.esa.int/docs/tutorials/S1TBX%20DEM%20generation%20with%20Sentinel-1%20IW%20Tutorial.pdf

15 Likes

the order is fine, good job looking for this information in the forum! :+1:

I also think the quality of the DEM is astoningly good compared to the attempts I made so far with S1 data. May I ask where your data is located?
Hillshading your image will reveal the true quality of the data in terms of surface forms.

I’d say faulty areas in your case are mostly are caused by lack of coherence and atmospheric disturbances. You can also once terrain correct your coherence layer and overlay it with your result to see where are the areas with least coherence.
Abolute heights are always hard to retrieve by InSAR techniques. Errors in altitude have many sources, one of them is coherence but also unwrapping (the single fringes are added up wrongly). Lastly, even the widely-used DEMs of SRTM and ASTER have differences of several hundres in extreme regions. So it’s hard to tell which one out of all is the ‘true elevation’.

You could try if you find other image pairs with similar good results and then later average all DEMs to one final product in order to minimize errors.

8 Likes

Thank you for the reply.

The data is located at north-east Black Sea region of Turkey.

Is Hillshade a function in toolbox? I looked for it but I couldn’t find.

I will try to produce another DEM of the same area.

Regards

1 Like

these are probably not the worst conditions for InSAR. Coherence will probably be low at the north end where the green coastal area begins.

Hillshading is not available in SNAP but can be done in QGIS, for example. Make sure that you select a projected coordinate system instead of WGS84 in the terrain correction module in order to retrieve correct hillshading.

nice example:

5 Likes

I’m wondering why you included Step 5 - Multilooking in your process. I am very new to InSAR so I’ve been following along the tutorial here: http://sentinel1.s3.amazonaws.com/docs/S1TBX%20Stripmap%20Interferometry%20with%20Sentinel-1%20Tutorial.pdf

They suggest here doing Topographic Phase Removal then the Goldstein Phase Filtering. Now that tutorial I believe is for deformation analysis - are these steps different for DEM generation?

Thanks!

1 Like

multi-looking is not obligatory in SAR processing, but reduces file size and, to a certain extent, speckle effects. If your interferogram doesn’t contain much speckle and you have the computing capacity, you can as well skip multi-looking.

2 Likes

Great, thanks for the quick response.

Here’s an example from an area where I am working. The image below is the phase with only the Goldstein Phase Filtering performed:

Here’s the same area with Multilooking performed:

Running the multilooking seems to have put the image back into Slant range if I am correct. Is that typical behavior?

the opposite is the case. All interferometry is performed in slant geometry until terrain correction. Applying multi-looking corrects for pixels of different sizes due to the incidence angle and makes them all squared.

Have a look at this comparison:

1 Like

Excellent, thank you for the detailed info.

I hope I am not hijacking the thread here. Feel free to split or move this if necessary.

Is it incorrect to perform Topographic Phase Removal when processing to Generate DEMs? I see @musicnerd did not include that step.

yes[quote=“dhill269, post:9, topic:5771”]
Is it incorrect to perform Topographic Phase Removal when processing to Generate DEMs?
[/quote]
yes. This is only if you want to detect subsidence or uplift (differential interferometry) and absolute heights don’t matter. But if you are interested in the topography itself, topographic phase removal must not be applied.

As long as the questions are addressing the same subject I see no need to split.

Thank you very much for this. However, after applying colour ramp, when I try to export in hillshade window, it gives an error that saying the file have one band only. It requires three bands. How did you get over this issue? I know this is not related to here, but it annoyed me a bit. :slight_smile:

Thanks again. Performing the Topographic Phase Removal is likely what has been causing me issues.

I will also provide my processing steps below as they may help others:

1. S1 TOPS Coregistration

(Only IW2 swath, Bursts 1-3, VV band only)

Output RGB image (RGB: Intensity Apr 10, Intensity Apr 22, N/a):

2. Interferogram Formation

(Default options used)

Phase image:

Coherence image:

3. S1 TOPS Deburst

(Default options used)

Phase image:

Coherence image:

4. Goldstein Phase Filtering

(Default options used)

Phase image:

Coherence image:

5. Multilooking

(Default options used)

Phase image:

Coherence image:

6. Snaphu Export, unwap and import

(8 processors, 20x20 row and colums, MCF, TOPO)

Phase image:

Unwrapped phase image:

7. Phase to Elevation

Elevation image:

The resulting elevation looks really wacky but I suspect this is simply due to very low coherence between the two images.

Here’s the coherence image for comparison:

The low coherence areas have very wacky elevation as we would expect.

Here is a histogram of the coherence showing that generally the values are quite low.

My study site is located in Canada and the images were acquired in April. My hunch is that the snow cover at that time is what is causing the low coherence. This is probably further exacerbated by the forest cover in much of the area. I will look into using some images from a difference season to compare results.

6 Likes

please friends how can i solve this problem in dem

this is natural data
i have big differences

Can you post a screenshot of the coherence image?

that is it >> i have mini project about this one please help me

i did subset to the image after Interferogram … that is right or no

you need better coherence for interferometry. Only areas with high coherence can be used later.

Please have a look at these answers:


2 Likes

S1A_IW_GRDH_1SDV_20170407T115610_20170407T115635_016038_01A770_709D


getting interferometric DEMs in tropical areas is nearly impossible with C-band data because it is scattered ad the canopy and coherence is lost after a few seconds. Even with bistatic data (e.g. TanDEM-X) this is a challenge. So getting a DSM of the canopy is also not possible.

Unless you have P-band data (wavelenghs of ~74 cm) these canopies are not penetratable.

When are the exact dates of your two images?

1 Like