[fitswcs] Thoughts about time coordinates

[William Thompson]

Folks:

The issue of the WCS time paper has come up recently on the Fits Working Group 
mailing list, and I thought I'd share some of my own thoughts on how I'd like to 
see time handled.  The concentration isn't so much on what the final keywords 
should be, as on what kinds of functionality the paper should address, together 
with some preliminary thoughts on axis types and projections.  At least it 
should get some discussion going.

Bill Thompson


		     Thoughts about date/time coordinates

I.  Date and time values can be thought of as having the following attributes:

     1.  System:  TAI, UTC, etc.

	These are the different systems that I'm aware of:

		TAI	International Atomic Time
		UT0	Rotational time from a collection of observatories
		UT1	UT0 corrected for effect of polar motion
		UT2	UT1 corrected for seasonal oscillations
		UTC	Coordinated Universal Time
		TDT	Terrestrial Dynamical Time (TAI + 32.184 seconds)
		TDB	Barycentric Dynamical Time (TDT + periodic (~ 10 msec))
		TT	Terrestrial Time (TDT at sea level)
			(Conceptually different, but in practice same as TDT)
		TCG	Geocentric Coordinate Time (different rate from TDT)
		TCB	Barycentric Coordinate Time (different rate from TDB)

	The current WCS papers do not allow time values to differ between
	coordinate systems, i.e. DATE-*, MJD-* do not take a system code.

	I've seen the keyword TIMESYS used to designate the time system.

     2.  Format:  Segmented vs. unsegmented

	Segmented: e.g. "2005-06-10T14:18:03.804"
		   e.g. MJD = 53531, seconds_of_day = 51483.804

	Unsegmented: e.g. TAI = 1497104315.804 seconds
		     e.g. JD = 2453532.1

     3.  Usage:  Absolute vs. relative

	Time values can be used to establish the absolute date and time that an
	observation was taken, and also the relative timing within an
	observational sequence.  The accuracy for those two different kinds of
	uses can be different: e.g. the relative timing may be known to 1 msec,
	while the absolute time may only be known to 100 msec.

	Many types of data analysis depend more on relative time than on
	absolute time, e.g. time series analysis, Fourier decomposition,
	wavelet decomposition.  Requiring that time values always be put on an
	absolute basis, where they only change in the lowest significant bits
	puts an unnecessary burden on analysis software.

	We commonly use the quantity DEL_TIME (or DELTTIME) to encode the
	time in seconds relative to DATE-OBS.

II.  Possible temporal coordinate type codes

	JD	e.g. 2453532.1

	MJD	e.g. 53531.6

	YEAR	e.g. 2005.4

	TIME	Time relative to some origin.  The natural origin for the TAI
		and UTC timescales is 1958-01-01T00:00:00.  Have to think
		about this for other kinds of time.  The origin will depend on
		what projection is applied to the data.

		TIME can be in any of the time units in Paper I.

III.  Possible temporal projections

	DIF	The coordinate is a difference from some origin, typically the
		start of the observation.  The default origin is DATE-OBS, but
		can also be user supplied through PSk_1a in the same ISO-8601
		format used for DATE-OBS and other DATE... keywords.

		The DIF projection can be used with any time basis.


	UTC	Similar to DIF, except that the default origin is 1958-01-01T0,
		and the handling of leap seconds is different.  The following
		example shows how time values would increment for the two
		projections in the vicinity of a positive leap second, assuming
		that the time basis is UTC in both cases.

			TIME-DIF		TIME-UTC
						
			1998-12-31T23:59:57	1998-12-31T23:59:57
			1998-12-31T23:59:58	1998-12-31T23:59:58
			1998-12-31T23:59:59	1998-12-31T23:59:59
			1998-12-31T23:59:60	1999-01-01T00:00:00
			1999-01-01T00:00:00	1999-01-01T00:00:01
			1999-01-01T00:00:01	1999-01-01T00:00:02
			1999-01-01T00:00:02	1999-01-01T00:00:03

		In other words, with the UTC projection, every day is treated
		as being exactly 86400 seconds long.

		The UTC projection can only be used with the UTC time basis.

		An example of a type of time using the UTC projection is "Unix
		time", which is the number of (non-leap) seconds since
		1-Jan-1970.



-- 
William Thompson
NASA Goddard Space Flight Center
Code 612.1
Greenbelt, MD  20771
USA

301-286-2040
William.T.Thompson.1 at gsfc.nasa.gov