# [COVERAGE] was: Re: [QUANTITY] Plea for pragmatism

Patrick Dowler patrick.dowler at nrc-cnrc.gc.ca
Fri Oct 31 12:40:17 PST 2003

```Coverage, or "bounds" as we call it in CVO, is a 1st order summary of the
full blown description. However, the model here doesn't make sense to me
because it does treat things uniformly when they are uniform and treat
them differently when they are different.

The WCS is typically ~2d, so the spatial coverage/bounds has to be a 2d
construct. CVO uses polygons, but other  shapes would be fine depending on how
exact one wants to be... the idea is that you want to tell the difference
between "in" and "out" and whether two things overlap/intersect.

Spectral and temporal axes are 1d, and a 1-d bound is an interval (loValue and
hiValue below). They are not error values on the refValue. What is "time of
observation"? The  observation starts at loValue and ends at hiValue, no??

In the CVO model, each axis (spatial, spectral, temporal) has a  "bounds"
object and a "sampling" object. The bounds are polygon or interval
(currently). The sampling object has several parts: number of bins, bin size,
resolution, and fill factor. One can compute the Nyquist ratio from bin size
and resolution, which is what Alberto was referring to about data being
undersampled). So, all axes have the same sampling description and the bounds
description dpeends on the dimensionality. If you split up the two spatial
axes in an attempt to have 4 x 1d axes, you could use intervals for bounds
everywhere, but then you are essentially putting an axis-aligned bounding box
around the polygon, which is a worse characterisation of the WCS, for no good
reason.

My thoughts, based on our experience actually trying to model this in general,
put it all into a database, actually describe different types of data this
way (WFPC2 images, 2QZ spectra, ROSAT fields, CFHT 12K images, etc)
and then query it in a uniform and general fashion.  Yeah, there are things
I'd do differently, but not much differently :-)

On Friday 31 October 2003 03:18, David Berry wrote:
> Doug,
>
> > Here is a first cut at the component data models needed for data
> > characterization.  This is based on further development of our
>
> Cambridge
>
> > etc. discussions.
> >
> > Coverage / Characterization Component Data Models
> >
> >     sky coverage
> >         coverage on the sky, if applicable
> >         e.g., a circular or rectangular region or aperture on the sky
> > 	Can be used to estimate the WCS but the full WCS should be
> > 	defined elsewhere.
> >
> >     time coverage / bandpass
> >         time of observation
> >         refValue, hiValue, loValue, fillFactor
> >         refValue is the mean time of observation, e.g., mid-point
> >
> >     spectral bandpass
> >         range of spectral frequencies in data
> >         id, refValue, hiValue, loValue, units, fillFactor
> >         id is user-defined bandpass name, e.g., "V", "SDSS_U",
>
> "K-Band", etc.
>
> >         refValue is the characteristic frequency of the bandpass
> >
> >     spatial bandpass
> >         range of spatial frequencies in data
> >         hiValue, loValue, units, fillFactor (no refValue)
> >         loValue is also known as the spatial resolution
> >
> >     flux bandpass
> >         range of flux values in data
> >         hiValue, loValue, units, fillFactor (no refValue)
> >         loValue is also known as the limiting flux or magnitude
> >         hiValue is saturation limit or maximum flux
>
> ...
>
> > Did I leave anything out?  Polarization perhaps, but that might be
>
> better
>
> > dealt with elsewhere.
>
> Where does spectral resolution come in this?
>
> Is the idea that a "coverage" component gives a summary of the ranges
> taken by significant parameters of the observation? For instance,
> spatial coverage is just a summary of the WCS (i.e. you could work out
> the spatial coverage by using the WCS to check every pixel). In this
> case, we need to make clear what happens to coverage when the
> observation
> is modified (e.g. taking a cut out will modify the spatial coverage,
> filters may change the effective spectral bandpass of an image, etc).
> So does coverage describe the data set "as it is now" (i.e. post
> processing) or as it was "in the beginning" (i.e. pre processing) -
> presumably the former.
>
> Also, in the same way that the spatial coverage is just a summary of the
> WCS, presumably the same could be true of other aspects of coverage. For
> instance, the "spectral bandpass" coverage  may be a summary of some
> more
> sophisticated spectral bandpass component. Thus it would be
> inappropriate
> to put a sophisticated spectral bandpass model within the coverage
> component.
>
> In general. I'm a bit unclear about the difference between the term
> "bandpass" and "coverage" in your list. "Spatial coverage" and "spectral
> bandpass" both seem to be similar in that they both describe the
> sensitivity of the observation over their particular domain; "spatial
> coverage" gives the sensitivity of the observation as a function of
> spatial position, and "spectral bandpass" gives the sensitivity of the
> observation as a function of spectral position. On the other hand,
> "spatial bandpass" seems to be more like a resolution.
>
> An alternative way to generalise the "coverage" problem, may be as
> follows... A "coverage" has an (optional) component for each domain of
> interest; spatial position, spectral position, time position, data value
> and spatial frequency. Each of these component has the following
> sub-components:
>
>    sensitivity: The sensitivity of the observation as a function of
>       the domain (spatial position, spectral position, etc).
>
>    resolution:
>       Typical increment between independant values in the domain
>
>    sample size:
>       Typical increment between between adjacent values in the domain
>
>    error
>       Typical uncertainties associated with values in the domain.
>
> Not all of these sub-components may be applicable to all domains - for
> instance I'm not sure what the resolution of the data value domain is
> ("error" would describe the data value uncertainty and "sample size"
> would describe any quantisation).
>
>
>
> David

--
Patrick Dowler
Tel/Tél: (250) 363-6914                  | fax/télécopieur: (250) 363-0045