Characterisation draft

[louys at alinda.u-strasbg.fr]

Dear Anita , dear DM members,

Thank you Anita for this extended document.
here are a few comments , I have copied only the bits I want to discuss:


Quoting Anita Richards <amsr at jb.man.ac.uk>:


As you know , there are 2 strategies to build up the XML schema: one 
that gathers the axes first , and then the properties along each axis, 
like coverage; another one that groups the metadata by properties and 
describe them in a sequence along the various axes.
For the moment the Axisfirst strategy is more developped. But in both 
cases , we will need to distinguish Axis on one hand and the properties 
(Coverage, ...) on an another.
So I would suggest to separate the requirements of Axes from the 
Coverage ones like this:

AXES ( in the schema , described by CharacterisationAxis, AxisFrame, ...)

>
> A description should provide at least one Coverage axis ("Axis
> Frame", "Axis").
>
> The unit and coordsystem 'must' be given for each Axis present (these
> may be relative to an internal reference only, e.g. (x,y) spatial
> coordinates. In such a case the Location and Bounds 'must' be given on
> that axis).
>
> All three of the Space, Time and Spectral Coverage Axes 'should' be
> given\footnote{some might be considered irrelevant for simulated data, or
> not
> conventionally provided e.g. for old spectra with no time stamp}.
> The Observable Axis 'should' be given. \footnote{its omission may seem
> reasonable if publishing e.g. the coverage intended for a future
> survey}.
In this case, the Axis can be defined , but the Coverage could be empty 
, or rather undefined or unknown ???

> Other Axes e.g. Velocity 'may' be given, with their Coverage.


COVERAGE

>
> The value 'must' be given for either the Location or the Bounds on
> each Axis.
> Values 'should' be given for both Location and Bounds.
>
> If Location is not given then for some Axes the default Location can
> be the mid-point of Bounds\footnote{in some cases this might be
> complicated (e.g. some spatial coordinates) or impossible}.
>
> If Bounds are not given then in some cases defaults are possible
> e.g. if a spatial axis has Coordsys ICRF the default would be all
> sky \footnote{ a more restricted coverage might be derived once there
> is a link to Observation and the telescope location.}
>

> Support 'should' be given, otherwise it defaults to Bounds, if present.
	Here we have defined Support as beeing the area or interval where we 
know measurements have effectively be taken. If an application follows 
this definition, for example, computing statistics on pixel values 
within the Spatial.Support, the result with Support defaulted to Bounds 
will be inconsistent.
I would prefer that Bounds should be defined if Support is given.
May be too restrictive for Data providers???

.....

OTHER AXES

> These relate to a specific COVERAGE Axis Frame but themselves contain
> Location (e.g. typical or reference value), Bounds etc.
>
> If there are many areas of Support within the coverage, the Resolution
> and Sampling Precision refer to the inside of each Support area and they
> are assumed to be consistent within each Support area (if not, each area
> must be described separately).

I fully agree with this : for the moment we don't know precisely how to 
build the schema for complex regions of diff. Sampling, Resolution, 
etc...

...
>
> If Resolution is present:
> You 'must' give resolutionRefVal (i.e. Location)
> You 'should' give resolutionBounds (default is resolutionRefVal)
>
If the resolutionBounds is not known, I would avoid to give any value 
and let the application interpreting the Xml description , to look for 
resolutionRefVal which is a 'must' field. Implicit default values may 
be badly documented and generate side effects.

>
> If SamplingPrecision is present:
>
> You 'must' give a samplingPrecisionRefVal (i.e. Location) for the
> samplingPeriod.
> You 'should' give a samplingPrecisionRefVal for the sampleExtent.
>
>

In the schema, because samplingPrecisionRefVal is in fact a vector: 
(SamplingPeriod, SamplingExtent) we 'must' have a value for the 
SamplingPeriod and we 'should' have a value for the SamplingExtent.

Then we 'should' give a value for samplingPrecisionBoundsSamplingPeriod
and we 'should' provide values for  samplingPrecisionBoundsSampleExtent.

> ACCURACY
>
> Each Axis Frame 'should' have associated Accuracy values for the
> precision of measurements along that axis, divided into statistical
> and systematic uncertainties. For each measurable quantity:


.....

> You 'should' give the ErrorRefVal (typical value i.e. Location)
> You 'may' give the ErrorBounds for uncertainties which vary along the
> domain of the axis
> You 'may' give an ErrorMap (as a URI) to describe the variation of
> errors with location.
>
I thought there could be cases where the ErrorBounds is more usefull 
than the ErrorRefval, when the Error have extreme values and the 
average error as ErrorRefVal does not bring very much.
So it could be that ErrorBounds is provided , and not ErrorRefVal. not 
realistic???

FLAGS

I agree with all the other parts of the document.


Cheers, Mireille.