[CF-metadata] CF calendars (was: problem with times in PSD dataset)

From: Jonathan Gregory <j.m.gregory>
Date: Mon, 10 Dec 2012 09:21:05 +0000

Dear Jon

> Just to repeat a remark that Steve Hankin made whose implications have not been explored in this discussion: different countries adopted the Gregorian calendar at different times. (Greece didn't adopt it till 1923!) So what is considered a valid Gregorian date varies from country to country (and some of those countries don't even exist any more, or at least the boundaries have changed...)

> 2. The non-proleptic Gregorian calendar is extremely problematic for historical observations as well as for models (astronomers use the Julian calendar consistently for this reason).

Yes, that's right. Nonetheless I don't think we can abolish the real-world
calendar, despite its ambiguities, because it's the one we really use! It is
not realistic to expect users of CF for historical obs to convert their dates
to the Julian calendar, I suspect, even if we say they should.

My proposal for the strict_gregorian calendar is to limit the danger
of the *default* calendar by forcing the calendar to be specified in cases
where there's a high risk of problems e.g. with a reference date of 1-1-1, so
that people are made aware of the danger. We could instead make it mandatory
always to specify the calendar, as has already been suggested i.e. not allow
a default at all. Would you prefer that?

As you say, we should be clearer about what the real-world calendar means, in
cases where users really want to use it. Adopting the earliest Julian-Gregorian
switch-over date is consistent with udunits and has some logic. It means that
dates would have to be converted for historical obs from countries which did
not convert at first. People working with historical data are aware of
different calendars being in use, so that seems more likely to be done than
converting to an other-wordly Julian calendar. Alternatively we could leave
it undefined, as it is now, but I don't think that's satisfactory. There must
be instrumental data affected by late switch-over e.g. in Greece and Russia. We
ought to say which calendar CF expects to be used for this.

udunits doesn't say what calendar the ref date is in because udunits supports
only the real-world calendar. I agree CF should be clear about this and the
obvious rule is that the ref date is in the calendar you are using. As you
say, that would be sensible. Any other choice would be rather bizarre.

The use of units of time which are calendar-based (months and years) is
deprecated in CF. That's because those units are not helpful, as you say,
unless you have calendar-aware software to interpret them, and some additional
rules on how that interpretation should be done e.g. what does 31 Jan 2012
+ 1 month mean? I think this is a different issue from calendars, since it
applies to any calendar which has variable-length months.

Best wishes

Jonathan
Received on Mon Dec 10 2012 - 02:21:05 GMT