Hi Steve,
I agree with all of these, as it seems does the community in general. For the record, I'd like to add another objection, that might also argue against using ISO8601 for time *metadata*. I'm not sure it's been explicitly called out yet that UDUNITS time specifications are inconsistent with ISO8601 in more than one way. For example, "1900-1-1 0:0:0" is legal in UDUNITS, but not ISO8601 (fields must be zero-padded and there must be a "T" in between the date and time). This is in addition to the different defaults to local time or UTC (actually I guess this is a property of CF, not UDUNITS?) Anyway, it seems that it might be confusing to introduce two string representations of date/times into CF, even for metadata. Are there any important cases in which ISO8601 is more expressive than UDUNITS syntax, or could we stick to UDUNITS for everything? Cheers, Jon From: [email protected] [mailto:[email protected]] On Behalf Of Steve Hankin Sent: 22 October 2010 02:06 To: CF metadata Subject: Re: [CF-metadata] time as ISO strings Hi Jon, Benno, John and other pals, Since this email thread already contains an element of informal voting I'll cast my ballot: CF is a better standard WITHOUT admitting ISO date strings as an encoding for time coordinates. My opinion is is based upon this outlook: * 'To create quality software, the ability to say "no" is usually far more important than the ability to say "yes."' (http://queue.acm.org/detail.cfm?id=1142044) Bloat and run away complexity are a continual threat to the quality of a standard as it evolves. I'd argue that the measure of whether a new feature deserves inclusion should be whether it adds useful new functionality and does so in a manner that is "clean" -- preserving the consistency and simplicity of the standard to the degree feasible. * Introducing ISO strings as coordinates adds no encoding power to CF. It raises issues of precision but then fails to address them adequately. (In fact it imports a host of ambiguities about interpretation of precision as Jon pointed out to us in the "metadata" versus "positioning" debate.) * It fails the test of consistency, since it is not applicable to virtually identical precision issues that exist for latitude, longitude and vertical coordinates. * the need to support two separate encodings (ISO and "days since xxx") for the same date/time coordinate information would force additional complexity into the standards documentation and into clients that would hope to be generic CF applications. It wound degrade interoperability for clients that did not support both encodings * it fails to address the multiple-calendar needs of CF. (creating another inconsistency) (ISO 8601 does not standardize the interpretation of dates prior to the 1582 Julian-Gregorian discontinuity ... which likely means that library codes cannot be trusted to give consistent results.) * it does not even add a useful measure of convenience. The "-t" option to ncdump already provides the needed convenience for file readers. And for file creators, a new utility that would translate ISO strings into CF time step values could probably be written in less time than this email dialog has occupied. (If this statement appears exaggerated consider that the effort to develop this utility would be paid over and over in the clients that would have to interpret this encoding.) None of this is a comment on the utility of ISO date/time strings as metadata. There are appropriate uses of ISO date/time strings in CF as non-coordinate variables and attributes. The NO vote is in regard to their use as CF coordinates. - Steve =================================== On 10/21/2010 11:57 AM, Jon Blower wrote: Hi Benno, 2010-09 is not necessarily a precise specification of a month - time zones make it a little fuzzy for one thing. Separate to this, there are parallel conversations going on in the ISO/OGC community about what time strings actually mean. A metadata person might say that "2010-09" is simply a shorthand for the fuzzy concept of "September 2010" and does not represent a precise interval (i.e. a square-wave function that is 1 during September and 0 outside). Apart from the time zone issue which blurs the boundaries, this square-wave is simply not what humans mean when, for example, they tag a report as having been written in September 2010. It just distinguishes it from version 2 of the report, which was written in November. In this context, it's just a label with some temporal meaning. These "metadata guys" are in discussion with the "positioning guys" who view date/times as precisely-defined positions within a temporal CRS. You may (or may not!) like to look at the GeoAPI mailing list, in which we are trying to figure out whether we can actually use the same Java types for both of these subtly-different views of date/times (we hope we can but haven't agreed). One might think that they are obviously the same thing, but I don't think so. You *could* modify CF so that to represent data that are "representative of September 2010", you specify a nominal date half-way through September and set the bounds to the first and last instants of September. And perhaps use a new cell_methods of "representative". But the half-way point and the bounds would be quite (very) tedious to compute in the general case (months and years are of variable length for example and depend on the calendar system). Of course, how the data is actually related to that interval is where the notion of precision might come in Actually, you've probably gathered that I also consider the notion of precision to apply to the interval itself, not just how the data relates to it. This discussion repeats a bit of the previous discussion on this list entitled "bounds/precision for time axis". I like Jonathan's distinction between the concepts of temporal resolution and representivity: http://www.mail-archive.com/[email protected]/msg01341.html. And just for completeness we should not that ISO8601 strings are not fixed-length, nor do they have a maximum length (in contrast to what I said before, sorry). So I can see some implementation challenges in NetCDF. Cheers, Jon -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Benno Blumenthal Sent: 21 October 2010 15:43 To: Steve Hankin Cc: Jon Blower; [email protected] Subject: Re: [CF-metadata] New standard names for satellite obs data (time as ISO strings) While expressing precision in CF is an interesting issue, in this case the Wikipedia quote is using the term in a different sense than I (hopefully we) usually mean -- ISO8601 lets one express time intervals succinctly in a single string, e.g. 2010-09 to mean all of september 2010, which is not an accuracy issue, it is a precise specification of a larger interval. It lets you write 2010-09-01/10-05 as well, i.e. it is not limited to intervals that involve special notational boundaries. As Steve points out CF expresses this using a bounds coordinate, i.e. giving the precise edges of each interval. Of course, how the data is actually related to that interval is where the notion of precision might come in, which cell methods/measures addresses, perhaps inadequately for the purpose at hand. ISO8601 is quite neat in the sense that it forces one to always specify an interval, and CF software reading time bounds data and rendering ISO8601 strings would do us all a lot of good. Benno On Wed, Oct 20, 2010 at 6:34 PM, Steve Hankin <[email protected]> <mailto:[email protected]> wrote: Hi Jon, Why do you see this as an issue of date-times as ISO strings in particular? The same issues of precision are found in longitudes expressed as a degrees-minutes-seconds string compared to a floating point. Or indeed to a depth expressed as a decimal string of known numbers of digits. ("100.00" communicates different precision than "100" though both a represented by the same binary value.) CF provides the bounds attribute and the cell methods/measures to clarify (somewhat) these points. What is your proposal for improved representation of precisions? And wouldn't a general improvement in how to specify coordinate precision be preferable to a solution that applies to time, only? - Steve ============================= On 10/20/2010 9:41 AM, Jon Blower wrote: Hi all, I haven't followed this debate closely, but I've had cause to do a fair amount of thinking (outside the CF context) on the pros and cons of identifying date/times as strings or numbers. I could probably write a very boring essay on this but in summary, they are not exactly equivalent ways of representing the same information. One way in which they are different is precision. A value of "x seconds since y" has no implied precision - typically in programs we take the precision to be milliseconds, but there's nothing to suggest this in the actual metadata (anyone who tries to populate a GUI from CF metadata struggles with this). Semantically it's a time instant; i.e. an infinitesimal position in a temporal coordinate reference system. However, an ISO8601 string can have various precisions. (The string "2009-10" is not considered equivalent to "2009-10-01T00:00:00.000Z".) >From Wikipedia (http://en.wikipedia.org/wiki/ISO_8601): "For reduced accuracy, any number of values may be dropped from any of the date and time representations, but in the order from the least to the most significant. For example, "2004-05" is a valid ISO 8601 date, which indicates May (the fifth month) 2004. This format will never represent the 5th day of an unspecified month in 2004, nor will it represent a time-span extending from 2004 into 2005." I've argued before in a previous thread on this list that it would be good to be able to specify the precision of time coordinates in terms of calendar date/time fields (which isn't the same thing as providing a tolerance value on the numeric coordinate value of a time axis). I'm not saying that we should definitely allow time strings in CF, just pointing out that they have some use cases we currently can't fulfil. I'm not sure they are definitively "bad practice" in all cases. (Regarding a technical point raised below, yes, it's a pain to represent variable length strings in NetCDF, but there is a maximum length for ISO8601 strings.) Hope this helps, Jon -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Lowry, Roy K Sent: 20 October 2010 10:00 To: Ben Hetland; [email protected] Subject: Re: [CF-metadata] New standard names for satellite obs data Dear All, As others have said, I think this debate is irrelevant as there should be no need for string timestamps in NetCDF. Providing a Standard Name only encourages what I consider to be bad practice. Cheers, Roy. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Ben Hetland Sent: 20 October 2010 09:14 To: [email protected] Subject: Re: [CF-metadata] New standard names for satellite obs data On 19.10.2010 16:27, Seth McGinnis wrote: What about using 'date' for string-valued times? That was my homebrew solution when I was considering a similar problem. If I may butt in and contribute here, I usually prefer names like 'datetime' or 'timestamp' in cases like this, because 'date' is potentially confusing. It may not be immediately obvious to a future reader (or programmer) that a variable called 'date' supports points in time down to for example seconds of accuracy. (Note that string data is a big pain to deal with in NetCDF-3, because you're limited to fixed-length character arrays. You need to use NetCDF-4 / HDF5 to get Strings as a data type.) (It may not be such a practical issue with ISO 8601 strings, as a reasonable max. length can be determined, I presume.) _______________________________________________ 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
