Partial constraints for GCPs

I am monitoring glacier change with monthly measurements, and currently need to align each flight manually due to the lack of precise ground control points. Because of the way the landscape changes, it would be relatively easy to establish repeatable, precise lat/lon coordinates for certain features even without precise geolocation, but altitude may change significantly between measurement runs.

Conversely there is a glacial lake with known altitude but it would be almost impossible to use lat/lon out there. So I was wondering if the GCP control point format could be extended to allow something like

<geo_x> <geo_y> NAN <img_x> <img_y> <image_name>
NAN NAN <geo_z> <img_x> <img_y> <image_name>

Update:
I've read more of the source and I would guess the first option (unknown altitude) may be easier to accommodate. The cost function is calculated in image space, and one unknown component would be equivalent to the distance of the specified image location to a line rather than the reprojected point. In the second case (elevation only) there appears not to be enough information.

Start by reading the GCP control-point format handling and the image-space cost function described in the issue. Determine whether unknown altitude can be supported without changing the existing model, and compare that with the elevation-only case. Done means the selected partial-constraint form is implemented, validated, and covered by appropriate tests.