@martinjuckes: I'm not sure why Jonathan muddied the waters with the "model world" concept, but this calendar should never be used with model data.
It is for encoding UTC timestamps into CF, so that they can be recovered. That's it. A model is never going to have UTC timestamps with leap seconds, so there would never be a reason to use this calendar for model results. For an adequate description, see Jim's posts -- but I'm going to try: A variable with this calendar will hold timedeltas since a UTC timestamp such that the timedeltas were computed from UTC timestamps using a non-leap-second-aware gregorian calendar. A user therefor can recover the original UTC timestamps using a non-leap-second-aware gregorian calendar. A timedelta is a span of time: (seconds, hours, microseconds) A "timestamp" is a description of a point on the time continuum in terms of year, month, day, hour, min, seconds, microsecond in a calendar A "UTC timestamp" is one with leap seconds properly applied, and thus perhaps as many as 37 seconds off from a "TAI" timestamp. HTH, -CHB -- You are receiving this because you commented. Reply to this email directly or view it on GitHub: https://github.com/cf-convention/cf-conventions/issues/148#issuecomment-437200690
