moment() drops the 3rd axis [or: cubes must have a 3rd axis > 1]

[teuben]

Since moment() always drops information on the 3rd axis, I cannot read the results back. Is this an intended design, or meant to be resolvable?

[keflavich]

What information is missing? You mean the frequency axis information? I think you're right, that should be preserved in the header. Can it reasonably be put into CRPIX/CRVAL/CDELT, or will most programs misinterpret that as indicating the presence of a cube?

[teuben]

yes, NAXIS=2. I'd argue we could try NAXIS=3 and NAXIS3=1 and use an appropriate description for the compressed CRVAL/CDELT. Both MIRIAD and CASA do it that way. Sounds like a DtU17 hackday option.

[low-sky]

I think there's a larger question here. Moments are 2D images and thus should they be read into spectral-cube? By default, I would say "this is working as intended. go use astropy.fits for dealing with images" However, moment0 straddle that case where you could talk about a 1 plane image with bandwidth equal to the original image. Moment1++ isn't a good fit to this 3D cubes though.

[keflavich]

I've never been a fan of NAXIS=3 NAXIS3=1 because it is a somewhat inaccurate representation of the data; for example, optical images with wide- or narrow-band filters usually do not have any "fake" third dimension. However, I understand why you want it.

Maybe we can provide an option for writing out in "cube" or "flat" format? I would default to "flat". We'd implement this in a special-case Moment0Projection LowerDimensionalObject.

[teuben]

ok, that's what i needed to hear :-) moments are 2D images, not cubes (but as I said, both casa and miriad can do a little more with them since they do retain some useful compressed information).

[low-sky]

@teuben What sort of functionality, apart from the obvious read/write, would you be looking for in SpectralCube operating on saved moments?

[e-koch]

We already carry some meta-data along in the moments. Perhaps it can be kept there? Along with the spectral axis information, things like the spectral range used to create the moment 0 would be useful too.

[low-sky]

I agree that it might be nice to have a LowerDimensionalObject case, mostly because the SpectralCube represents the nicest wrapper to an image-with-wcs that I have worked with. I would love to have 2D images with convolve_to and spines and other nice features that are in SpectralCube. On the other hand, I'd also like for cube[37,85,32] to give a value.

[keflavich]

@low-sky that last issue sounds like a feature request. It's pretty implementable. Make an issue.