Dear Jonathan and Jim,
I'm o.k. with placing UTC or GPS as part of the calendar attribute
(rather than in the units). Your arguments in favor of that option make
sense to me now.
Jim, I think the second bullet (*) point is important and makes sense to
me. I hope this doesn't mean we're in some disagreement.
In order to compare variables representing measurements taken
simultaneously, but with one variable stored following the UTC system
and the other following the GPS system, you could do the following:
1. Determine the offset between the UTC reference time-stamp and the
GPS reference time-stamp (note even when the apparent reference times
are identical, there can be an offset of up to 16 seconds)
2. For the UTC variable, subtract the offset (found in step 1) from all
the time coordinate values. Also change the name of the calendar to
gregorian_gps. And finally, change the reference time for the time
units to be the same as the units used for the GPS variable.
Now both variables will have time coordinates that follow the same GPS
calendar (also with a common reference time). You can then, of course,
convert the elapsed time to a time-stamp of the form 2015-01-01 0:0:0
following the rules of constructing a Gregorian calendar (without leap
seconds).
If one of the variables has been collected according to UTC, but the
elapsed time has been stored without accounting for leap seconds (i.e.,
calendar = "gregorian_utc_nl"), then the procedure is more complicated:
1. For each time-coordinate add the number of leap seconds that
historically have been applied under the UTC system between the
reference time and the elapsed time. [This converts to gregorian_utc.]
2. Follow steps 1 and 2 above to convert to gregorian_gps.
If anyone is really worried about differences in time of a few seconds
when comparing a model to observations, then things become more
complicated because the actual length of a model day is exactly 86400
sec, but the earth's day averages closer to 86500.002 seconds. The
elapsed time between a sunrise occurring in 1980 and one occurring
exactly 30 year later will be shorter by 15 seconds in the model. To
keep the model in sync with the real diurnal cycle, we'll have to insert
leap seconds into elapsed time. Alternatively, we could throw away the
leap seconds (and the data associated with those times) in the
observational dataset and yield a time-series that would stay in sync
with the model's diurnal cycle.
I would summarize this as:
1. the length of a real earth day (as gauged by differences in elapsed
time between the beginning and end of the day) is the same whether the
elapsed times are recorded using the gregorian_gps or gregorian_utc
calendars. The UTC system makes sure, however, that time-labels stay in
sync with the earth's diurnal cycle, whereas the GPS system allows the
phase of the wall clock to shift relative to the earth's diurnal cycle.
2. the length of a model day (exactly 86500 sec) is a little shorter
than a real day, and this should be indicated by setting calendar =
"gregorian" calendar (without suffixes), or one of the other original
calendar attributes (like "julian" or "noleap").
3. For observational data, use of calendar="gregorian" indicates that
the reference time and elapsed time might be wrong by as much as 16
seconds (but they also might be correct).
Please let me know if my understanding is incorrect (if you can take the
time to do this).
cheers,
Karl
On 7/10/15 7:01 AM, Jim Biard wrote:
Jonathan,
I think I'm just about completely in agreement with you about meanings
and such! I think your second bullet (*) point about algorithms is not
necessary, but you have worded it in such a way that I'm OK with it.
Can you agree to changing gregorian_nls to gregorian_utc_nls? I know
it's longer and all, but I think it's better to be precise. This
calendar assumes UTC without consideration of leap seconds, and I
don't want someone to confuse it with a more generic 'no leap seconds'
case. (As a side note, you could have a similar - if unlikely -
scenario where someone encoded GPS timestamps using a leap second
aware algorithm to get wrong elapsed times. And as long as the inverse
of the same algorithm was used to decode to timestamps, correct
timestamps would be produced. I guess that calendar would be
gregorian_gps_ls. And I'm not suggesting that we add that calendar!)
It's also still not to late to go with one or more new attributes to
cover the time system instead of making more calendars, but I'm OK
with where we are.
Grace and peace,
Jim
On 7/1/15 10:42 AM, Jonathan Gregory wrote:
Dear Karl
Thanks for your contributions. The points you are raising are similar to those
which Jim and I have been debating for some time, gradually moving towards
mutual understanding! I think the position we had got to is quite economical.
Let me try to relate it to what you suggest.
Your main suggestion is to allow UTC or GPS to be stated in the time units,
whereas Jim and I have been discussing putting it in the calendar attribute.
The latter is what I prefer because it seems to me that the distinction between
UTC and GPS is quite like the differences among all the model calendars, and
with the real-world calendar. The calendar attribute indicates two things:
* The calendar (or time system) in which the reference time is stated.
* The algorithms which should be used to decode the time coordinates into
timestamps in the calendar of the reference time, and to encode timestamps in
that calendar to time coordinates with the given reference time.
These points distinguish UTC and GPS times, and they similarly distinguish
UTC from 360-day-calendar times, for instance. So it's a more uniform approach
to use the calendar attribute in all cases. Also, parsing the time units att
is less convenient than inspecting the calendar att, which is a single word
with only a small number of possible values.
1) CF not try to accommodate folks using "wrong" software.
The case in point is using software without leap seconds to encode UTC
timestamps. I think it's too harsh to call it "wrong". It pretends that UTC
is something slightly different and other-worldly, and that makes its elapsed
times a bit inaccurate. But it's a perfectly reasonable thing to do. As I
said in a recent email, I expect that many or most climate and NWP models,
when using the "real-world" calendar to deal with real-world weather and
climate (assimilating and comparing with real-world obs), have time coords
which are encoded from UTC timestamps but not using leap seconds i.e. "wrong".
This seems sensible to me since they certainly don't depend on precision to
within a few seconds, and it's easier. So why not? It's not the job of CF to
tell people what to do, and we should accommodate it so long as it's well-
defined and has a use-case.
2) we relax our requirement that udunits be able to handle the time
coordinate because it won't recognize and interpret "UTC" and "GPS".
udunits can't handle model calendars anyway. We use udunits only to define
the format of the time units, not to depend on it for decoding and encoding.
There is also a use-case for real-world times where the data-producer is not
precise about whether the ref time is UTC or GPS, or whether the coords were
encoded with leap seconds. It seems that we should distinguish this case from
the precise and deliberate use of 86400-second days for UTC. I think this is
a rather specific case of imprecision, not the same as inaccuracy in coord
values in general (for which we don't have a convention at present - no-one's
asked for one yet), and that it relates specifically to the calendar. So I
still prefer the proposal we've already arrived at for four values of the
calendar att for the real world:
gregorian: Not specifying whether it's GPS or UTC or how encoded. Elapsed
times and decoded coords are therefore imprecise.
gregorian_nls: UTC ref time and time coords encoded from UTC timestamps without
leap seconds, which are inaccurate elapsed time, but can be decoded accurately.
gregorian_utc: UTC ref time, time coords are elapsed time, decoding to UTC
timestamps if UTC algorithm is used.
gregorian_gps: GPS ref time, time coords are elapsed time, decoding to GPS
timestamps if GPS algorithm is used (which is actually the same algorithm
as for gregorian_nls i.e. 86400-second days).
Best wishes
Jonathan
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
--
CICS-NC <http://www.cicsnc.org/> Visit us on
Facebook <http://www.facebook.com/cicsnc> *Jim Biard*
*Research Scholar*
Cooperative Institute for Climate and Satellites NC <http://cicsnc.org/>
North Carolina State University <http://ncsu.edu/>
NOAA National Centers for Environmental Information
<http://ncdc.noaa.gov/>
/formerly NOAA’s National Climatic Data Center/
151 Patton Ave, Asheville, NC 28801
e: [email protected] <mailto:[email protected]>
o: +1 828 271 4900
/Connect with us on Facebook for climate
<https://www.facebook.com/NOAANCEIclimate> and ocean and geophysics
<https://www.facebook.com/NOAANCEIoceangeo> information, and follow us
on Twitter at @NOAANCEIclimate <https://twitter.com/NOAANCEIclimate>
and @NOAANCEIocngeo <https://twitter.com/NOAANCEIocngeo>. /
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
