In Moscow Doug asked what science users wanted from SEDs and I think the question deserves more attention. Here are some priorities drawn from use cases which have arisen from AstroGrid and VOTech science cases and user workshops, to get the ball rolling.
Firstly, in general an SED is any sequence of flux density measurements along a spectral axis - there are already more precise VO definitions but the main point here is that high-resolution regularly-sampled spectra _can_ be thought of as SEDs but in many cases more dedicated spectral tools and standards are more suitable. However high-res spectra can be mixed with individual broad-band photometry points in an SED (with some caveats, see below).
SEDs may come from a single data source or many.
The more fundamental requirements come first, but this does not mean that the tool components have to work in this order:
The sources need to be cross-matched. That typically gives a table with many sources, with column labels for each waveband. I believe that current tools need this to be split and transposed with one table per source, with one wavelength (etc.) column with many values and one flux column with many values. (can be done using STILTS)
The units need to be homogenised. This can be done easily to moderate accuracy (e.g. VOSpec methods, STILTS etc) in some regimes/some units; in the case of optical magnitudes or X-ray flux densities (and even nominal band centres) it may be model-dependent and is often largely manual at present. This is being worked on by several EuroVO partners (AstroGrid, ESO et al.). In general the current outputs are good enough for e.g. diagnostics using SEDs stretching from X-ray to radio, but not necessarily for <5% accuracy e.g. for calculating photmetric redshifts from optical data, unless the input data are all from a homogenous source.
The SED needs error bars, if possible to take into account conversion uncertainties.
For data taken with many instruments, aperture matching may or may not be a consideration - e.g. for a star, is it unresolved in each data set? Might ther be more than one source in low resolution observations?
Are there any issues in including high-res spectra, whether or not it contains lines?
Finally, ESO are aworking on a tool to also include data extracted from images, and AstroGrid are working on a model-fitting tool (using e.g. the SVO models)
thanks
a