GCPs with high vertical accuracy but low horizontal accuracy (to force leveled water body shores)

I’ve surveyed several lakes, and I need to measure lake volume changes and model how much water is stored for a given increase in lake level. Therefore I don’t need accurate absolute coordinates. I surveyed without GCPs and only with the drone GPS (navigation GPS, not fancy PPK or RTK).
Now I’m having trouble getting the lake shore all at the same vertical level, as it shows variations of several meters. I don’t know how to solve this, but one option I’ve thought of is to create GCPs along the lake shore using the coordinates obtained from an orthomosaic created in an initial run of the reconstruction. Then use those GCPs for a second run, but setting their elevation all to the same value. Ideally, I would like to find a way to tell ODM to give more importance to the vertical coordinate of those GCP than to the horizontal ones, which might be wrong due to the warping of the model in the first run. Is that possible? Any other ideas to get the lake shore level?




We don’t have any weighting capabilities for GCP components, nor GCPs themselves.

Can you tell us more about your survey and data?

What is your mean GPS error for a given dataset? Have you adjusted the --gps-accuracy flag to 2x that to start? Have you enabled --rolling-shutter to test?

If you need high repeatability, then yes, you’ll likely need to use GCPs to keep things constrained.

Hello, this is a task you can solve in the „Cloudcompare“ Software which is Open Source. In Cloudcompare you can align (register) different pointclouds to a reference pointcloud. It is necessary to find equal points, at least 4 but more are better, that haven’t changed coordinates and elevation. With the aligned pointclouds you can also do volume calculations in Cloudcompare.


You can also use align files within WebODM:

The align option looks very interesting. Is there more documentation on how it works? It only makes translations, or also rotations, de-warping of some sort?

1 Like

Thanks a lot for the suggestion. I’ll give it a try. Thanks!

1 Like

None yet! I need to go and collect some good data and learn the function before I write it up.

I’ll dig for what I can find later this evening or this week.

That would be great! Thanks a lot, I look forward to learn more of that function.

1 Like

Hi @Saijin_Naib,
Did you find any documentation on the align option?
I haven’t. Hopefully, you had better luck.

1 Like

This is what we have currently.

It is based upon this explanation:

I’ve not had a chance to re-synthesize this into something a bit more friendly, nor collect my own data. Sorry -_-

Looks like linear transformations only:

Each registration module solves a 6- or 7-parameter similarity transformation (three translations, three rotations, one optional scale).

So it assumes the elevation models are broadly correct, but need translated, rotated, and optionally scaled.

1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.