Elevation offset when using Pleiades RPC

[Kathan Mehta]

Hello all, 

Firstly thanks a lot for developing and maintaining this really cool tool. Apologies if this is a repeat of questions.

I am a new AMES Stereo user and I went through the book/document to run a few tests on a Pleiades L1B Stereo Pair in order to generate a dsm output. Unfortunately, all the outputs I get have a negative offset of about ~30m in elevation when compared to a SRTM dem.

Here are some of the tests I have run : 

1. Directly ran parallel_stereo on the raw input stereo (using the default parameters) followed by point2dem on the point cloud output. 
2. Map-projected the inputs to an external low resolution SRTM-dem and used the map-projected outputs as inputs to the parallel_stereo command (same default parameters as in point 1) followed by point2dem.
   
3. Same as point 2 but also modified the parameters in the parallel_stereo command. Mainly using "asp_mgm" stereo-algorithm and sub-pixel mode 2 (affine adaptive window, Bayes EM weighting) 
4. Lastly, instead of using the low resolution SRTM Dem as reference, used output of step 2 as the low resolution reference. Apart from that, followed same steps and methods as step 3.

All 4 output dam's have a vertical shift as compared to the SRTM dem. I have also attached a screenshot of the elevation profile here for reference. Red -> SRTM dem, Navy Blue -> Test 1 (-30m), Green -> Test 2 (-30m), Purple -> Test3 (-30m), Light Blue -> Test 4 (-70m). 

Is there a step or a parameter I am missing?  Any inputs would be greatly appreciated.

Thanks, 

Kathan

[Oleg Alexandrov]

To have an elevation change from SRTM is quite expected. First SRTM is sometimes relative to the geoid, while ASP-produced DEMs are relative to the WGS84 ellipsoid. Second, there can be systematic issues with the cameras. That is why we have the pc_align tool. https://stereopipeline.readthedocs.io/en/latest/tools/pc_align.html This one aligns the ASP DEM to your preferred DEM. 

I will also suggest reading https://stereopipeline.readthedocs.io/en/latest/next_steps.html#choice-of-initial-guess-terrain-model when it comes to getting an SRTM DEM relative to WGS84. 

Aside from this, I will suggest running point2dem with the option --errorimage and inspect the resulting intersection error. If it is no-uniform and not small (so not under 1-2 m, you may need bundle adjustment as well). https://stereopipeline.readthedocs.io/en/latest/tools/bundle_adjust.html

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/ab8e12e4-27f1-4cca-9cb0-4c92eb173c96n%40googlegroups.com.

[David Shean]

I agree - sounds like a geoid offset issue.  The Pleiades RPCs should be better than 30 m.

Kathan, where is the site?  You can quickly see what the expected offset between the EGM96 geoid (default for many standard SRTM products) and the WGS84 ellipsoid (default for ASP and the L1B camera models) should be: https://en.wikipedia.org/wiki/Earth_Gravitational_Model#/media/File:Earth_Gravitational_Model_1996.png 

Here’s another version of that figure that I think is a little easier to interpret: https://uwgda-jupyterbook.readthedocs.io/en/latest/modules/07_Raster2_DEMs_Warp_Clip_Sample/07_Raster2_DEMs_Warp_Clip_Sample_exercises.html?highlight=geoid%20offset#prepare-egm96-offset-grid-figure

What is the source/version of your SRTM data?

I recommend either removing the geoid offset from your SRTM product or applying the offset to your output ASP DEM if you need orthometric heights (relative to geoid).  Either can be done with ASP dem_geoid tool.  Alternatively, you can just use a version of SRTM that has heights above the WGS84 ellipsoid (see the OpenTopography Global DEM API).

Hope that helps.

-David

--

David Shean
Civil and Environmental Engineering
University of Washington
https://www.ce.washington.edu/facultyfinder/david-shean 

201 More Hall, Box 352700
3760 E. Stevens Way NE
Seattle, WA 98195-2700
Office: (206) 543-3105, Wilcox Hall 265
Pronouns: he, him, his

To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/CADDkNmMm-Sfq%2BjXOnzwGfx2usEA7vrTQXzc87BidbUnSHqqw1w%40mail.gmail.com.

[Kathan Mehta]

Thanks a lot for the prompt response. Both were really helpful.

The vertical offset that I was having was in line with what David mentioned in the two links. It was indeed because of the offset between the EGM96 geoid and the WGS84 ellipsoid. I was able to get rid of the vertical offset by using the dem_geoid tool as suggested. 

This might be a naive question but I had errors when I try to use the image_align tool to aligns the ASP DEM to srtm DEM and the error says : "RANSAC was unable to find a fit that matched the supplied data. Alignment transform computation failed". This might be because of the difference in resolution between the two dem's (1m vs 10m)

I know the suggested tool was pc_align but I do not have a point cloud for the srtm dem which is why I am unable to use that tool. Am I using the image_align tool correctly or is there some parameter that I should specify? Also, what is the correct method to use the pc_align tool for my case?

Thanks again for the help.

Regards, 

Kathan

[Oleg Alexandrov]

Glad to know you got to the bottom of the shift issue.

This might be a naive question but I had errors when I try to use the image_align tool to aligns the ASP DEM to srtm DEM and the error says : "RANSAC was unable to find a fit that matched the supplied data. Alignment transform computation failed". This might be because of the difference in resolution between the two dem's (1m vs 10m)

I know the suggested tool was pc_align but I do not have a point cloud for the srtm dem which is why I am unable to use that tool. Am I using the image_align tool correctly or is there some parameter that I should specify? Also, what is the correct method to use the pc_align tool for my case?

The pc_align tool should run on DEMs just as well (per its manual). The image_align tool likely is not the right tool for aligning DEMs of very different resolution. And it will not be accurate enough for alignment purposes, I think.

So, try pc_align, with the denser DEM being passed in first, so the ASP one, then the other DEM passed in second. Both of these should be relative to WGS84 and not geoid. I will also suggest reading carefully the pc_align tool manual page, as it makes certain assumptions about the data. It also prints out some stats which should tell you how well it did.