|
The datamodel specifies a non-sparse covariance matrix COVAR2 which is supposed to capture the large-scale behaviour of the spectrophotometry (there's also a banded covariance matrix COVAR that captures the pixel-to-pixel errors and covariances). If we had an optical spectrograph covering grizy, then I'd think of this as a 5x5 matrix with components that roughly correspond to the e.g. <(g - <g)>)(z - <z>)> – i.e. the large scale errors in the spectrophotometry. These could come from the uncertainty in the spectrophotometric standards, from the corrections due to guiding/fibre centring errors, the mismatch between the extended object and resolve object PSFs (and thus extraction errors), and other things that I don't yet understand!
I don't have a concrete proposal for how to measure these, e.g. what is the rôle of the fibre magnitudes, but the flux calibration work seems to be the place to put this.
The COVAR2 matrix is currently 10x10 because we have c. 5 optical photometric bands and sometimes some NIR information (J) and a factor of c. 2 oversampling seemed wise.
|