busko at stsci.edu
Wed Apr 23 10:38:46 PDT 2003
I would like to complement the contents of my write-up on spectral
data structures with an important point.
In the Specview design we almost completely swept under the rug the
issue of spectral resolution. This is something I should have added to
my write-up, but forgot. The reason is that originally Specview was
conceived as basically a visualization tool, with no data analysis
capabilities. And most of my recent work on it has emphasized the
When we added the ability to fit models to data, and now when we are
adding other data processing and analysis capabilities to it, we face
the problem of how to handle resolution issues when the input data
itself doesn't carry explicit resolution information. For instance,
we need to know what is the effective resolution of each sample when
computing a chi-squared metric between the data and a spectral model.
When using the current version of Specview to fit a model to a
composite data set made out of data pieces generated by different
instruments, the chi-squared computation may be more or less biased
depending on the differences in resolution between the instruments.
We have a plan to address this issue in a future release of Specview,
but my aim here is to emphasize this important aspect of spectral data
that, in my view, should be an integral part of any VO spectral standard.
More information about the dm