Misalignment of multitemporal DEMs in z-axis

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

Hello,

I am doing my BA-thesis on palsa mire degradation, which are essentially permafrost cored mires. My supervisor has provided me with RBG drone data, in 2021 surveyed with the camera S.O.D.A._10.6_5472x3648 (res. 0.0249 m) & in 2023 with senseFlyAeriaX_18.5_6000x4000 (res. 0.0217 m). Additionally, I have GCPs with x,y,z marked in the field which I find in the images through recognizable features as they are e.g. metal lids, edges of concrete blocks and rock outcrops. I need these images to match perfectly in to be able to perform volumetric analysis.

I have added the same 6 GCPs with an even spread in both images, reoptimized and run the processing. However, when comparing the point cloud and cross-checking points (which should be fixed) in QGIS, I note that the DEMs differ a lot in the z-axis. So, they are well aligned in 2D, but bad in 3D. I also note that the offset is not constant, but instead that there is substantial internal variation in the error, which, again, I estimate by checking fix points. Fix points I use are rock outcrops, a W-E-running road to the north and a parallel railroad to the south. (I should be noted there is expected subsidence in all non-fixed areas in the 2023 DEM). Moreover, I note that the z-values of where I marked the GCPs themselves are different in the 2021 & 2023 DEMs, which I find strange.

How can I make them align well in the z-axis. Add more GCPs? Only do manual tie points?

Thank you!
V

Quality reports:
2021-07-25_SF_3d_report.pdf (1.4 MB)

2023-07-19_SF_arx_report.pdf (1.6 MB)

2

Hello,
I looked at your quality reports, and there are some issues. There are problems with your GCPs and with some calibrations of your cameras. I would try reprocessing but use the Alternative calibration method. This is designed more for agricultural applications, which your project probably falls under this category.

Alternative
Alternative1604×1460 272 KB

Second, I would look closer at your GCPs. It is best if you can mark between 10-15 of the best images. Try zooming in to the exact pixel of the target. If an image is blurry, don’t use it. And try to use images with the target closer to the center of the image.

After remarking, run step 1 again and see if it improves the accuracy of the GCPs.