Absolute accuracy problem resolution


I just fired my subscription to WebODM back up to process a flight I recently conducted with ground control points. I had previously started a thread about having problems with the absolute accuracy in the processing. I am very happy to say that I may have gotten to the bottom of the problem.

My Emlid unit gives me Latitude, Longitude, and Ellipsoidal Height measurements as well as measurements within the CRS of my choosing. In my case, I use NAD83 (2011) for the horizontal components and NAVD88 for the vertical component. This is how my state’s NTRIP is configured. I typically get errors of about .3 to .4 meters in the XY and anywhere from .5 to 1 meter in the Z. The reason for this may be that ODM converts input coordinates to the nearest UTM CRS. In my case that’s 17N. It should also be noted that the conversion is to UTM 17N WGS84. I found the following article on Quora about converting NAVD88 to WGS84. In short, accurate conversions between the two are not possible.

NAD83 (2011) seems to have identical XY values to LAT/LONG in my survey file. My accuracy is a bit better in this direction but I’m still wondering if there’s a bit of a translation error in changing from WGS84 to UTM 17N. Extremely high absolute accuracy may not be a possibility for me given my state’s NTRIP configuration.

I would like to mention that my Horizontal CE90 relative accuracy was .009 m and my Vertical CE90 relative accuracy was .021 m. This leads me to believe that the software does an amazing job of reconstructing reality.


Glad to hear your reconstructions are sane (and really, really precise, it seems)!

Yeah, unfortunately, some CRS conversions are pretty lossy, and beyond that, some reference systems are inherently poor for storing accurate coordinates. Ensemble datums like WGS84 are an example of this, having an inherent inaccuracy of 1-3m, plus a shifting of coordinates of something like 80mm a year due to plate tectonics (and it being not plate-fixed).

The older the reference point recorded in WGS84, the more the location could have shifted… Meeps.

1 Like

I would like to mention that I processed the same image set without the GCP list and got exactly the results one might expect. Absolute Horizontal CE90 was 1.083 m and Vertical CE90 was 2.689 m. Compared to H CE90 at .356 m and V CE90 at .978m, you can see that the accuracy is better than not shooting points at all. I think this reinforces that the real error comes from converting between coordinate reference systems.

Relative accuracy was spot on in both cases.

1 Like

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