Continuum Subtraction

A common task with data cubes is continuum identification and subtraction. For line-rich cubes where the continuum is difficult to identify, you should use statcont. For single-line cubes, the process is much easier.

First, the simplest case is when you have a single line that makes up a small fraction of the total observed band, e.g., a narrow line. In this case, you can use a simple median approximation for the continuum.:

>>> med = cube.median(axis=0)  
>>> med_sub_cube = cube - med  

The second part of this task may complain that the cube is too big. If it does, you can still do the above operation by first setting cube.allow_huge_operations=True, but be warned that this can be expensive.

For a more complicated case, you may want to mask out the line-containing channels. This can be done using a spectral boolean mask.:

>>> from astropy import units as u  
>>> import numpy as np  
>>> spectral_axis = cube.with_spectral_unit(u.km/u.s).spectral_axis  
>>> good_channels = (spectral_axis < 25*u.km/u.s) | (spectral_axis > 45*u.km/u.s)  
>>> masked_cube = cube.with_mask(good_channels[:, np.newaxis, np.newaxis])  
>>> med = masked_cube.median(axis=0)  
>>> med_sub_cube = cube - med  

The array good_channels is a simple 1D numpy boolean array that is True for all channels below 25 km/s and above 45 km/s, and is False for all channels in the range 25-45 km/s. The indexing trick good_channels[:, np.newaxis, np.newaxis] (or equivalently, good_channels[:, None, None]) is just a way to tell the cube which axes to project along. In more recent versions of spectral-cube, the indexing trick is not necessary. The median in this case is computed only over the specified line-free channels.

Any operation can be used to compute the continuum, such as the mean or some percentile, but for most use cases, the median is fine.