[CF-metadata] time dimension and calendar for an ice-sheet model output

From: Jonathan Gregory <j.m.gregory>
Date: Thu, 5 Feb 2009 22:11:11 +0000

Dear Constantine

> 1) time in the output file denotes "seconds (or years) relative to
> present", i.e. it for paleoclimate runs it is negative (and is used to
> extract appropriate temperature and sea-level offsets from ice-core
> data, for example) and for runs "into the future" it is positive.
> 2) a year is always 3.15569e+07 seconds.
>
> I might be wrong, but it seems that 1) suggests that any particular
> "real" reference date will be... well... unfitting, and 2) suggests
> that standard calendars are not applicable here either.

You're right, you couldn't specify a time-unit relative to a moving reference
time. But if you are using "relative to present" you can't really be precise
about what "present" means, can you? The precise meaning of "relative to
present" changes as seconds pass! I imagine you mean "approximately present"
in the same sense as years BP. That is quite vague, so it seems to me you could
just pick a reference date, define it as "present", but recognise it as an
arbitrary choice. If you are talking about -21000 years relative to present, it
doesn't much matter if it's today, last Tuesday, or 1984 that you choose. It
might be simplest if you always used the same date, so that the absolute values
of your time coordinates always had the same meaning in your files.

Alternatively, it might be more convenient if you used positive time coords
relative to an appropriate reference date, such as "seconds since -21000-1-1".
Unfortunately udunits won't process that correctly.

Your year length is exactly what udunits means by a year, so you can use the
unit of years safely.

Does your model have a seasonal cycle? If not, a calendar of "none" would be
appropriate. If it does, CF has useful facilities for describing climatological
means. You can also specify your own custom-made paleocalendar. But for these
facilities to make sense you would have to change your year to a whole number
of days, possibly with leap years.

Cheers

Jonathan
Received on Thu Feb 05 2009 - 15:11:11 GMT