How to correct the elevation of the raster?

Question

While comparing the DEMs of relatively small area raster (1) obtained of UAV data and (2) EU-DEM elevation model, I spotted around 30 meter difference in elevation between them.

Since no GCP was used while harvesting data and I know that commercial grade UAV GPS chips are not able to deliver accurate Z values, I checked my findings via elevationfinder.com which gave same results as EU-DEM.

To correct the data, I've established Z values difference for points which are easiest to identify on both rasters and recalculated it using Raster Calculator. I am not sure however whether this is the proper way to do it. Can somebody confirm or suggest other, more scientific and reliable approach?

Answer 1

This appears to me to be a geoidal separation issue.

Not knowing the type of GPS equipment used on the UAV is forcing me to make several assumptions.

Assuming the GPS chip is single frequency with EGNOS corrections (Assuming you are in Europe) You should get approximately three meter relative horizontal accuracy, and around five meter vertical accuracy. This can vary wildly, but I am comfortable stating these values. (I have seen much better, and I have certainly seen much worse.)

Without ground control points, or real time corrections, you can have some fairly significant positional errors in all three coordinate dimensions.

Starting with the raw UAV data, convert the center of the pixels to points, or do a raster to vector conversion creating polygons which would contain Easting, Northing, and Elevation data.

Then calculate Geoidal Separation values for the specific vertical datum and geoid model you wish to employ. Apply these values to each point. (Depending on the size of the area, and the size of the pixels, you will probably get a lot of them.) This should provide a pixel for pixel vertical correction rather than a bulk correction based upon interpolation of several known points.

Hopefully, this will provide you with a solution more in line with the DEM values.