hbushouse
commented
7 years ago Email from Arnold Rots, submitted at JWST Proposal Planning workshop:
From: Arnold Rots arots@cfa.harvard.edu Date: Thu, May 18, 2017 at 10:29 PM Subject: Time Series 101 To: "dm@ivoa.net" dm@ivoa.net, IVOA List VOEvent voevent@ivoa.net, DAL mailing list dal@ivoa.net
Following up on yesterday's time series session, I thought it might be helpful to summarize the basics on time information. It really isn't that complicated. For reference and more information than you really want, you can turn to FITS WCS Paper IV: A&A 574, 36 (2015).
-
Required: 1.1 Record time stamps in JD, MJD, ISO-8601, or elapsed time (if in elapsed time, you need to give a zero point in a time stamp that is not provided in elapsed time, of course) 1.2 Provide the time scale used (TT, TDB, TAI, GPS, ET, UTC, TCG, TCB) 1.3 Provide the reference position (place where the time is measured)
-
Note the following: JD and MJD do not imply a time scale; it needs to be provided separately. JD and MJD are dimensionless, though a unit of "day" is implied. It's a bad idea to mix UTC with JD or MJD, since not all UTC days are the same length. Use the restricted form of ISO-8601: [[+|-]c]ccyy-mm-dd[Thh[:mm[:ss[.ss...]]]] No time zone characters. TDB runs on average synchronously with TT, but corrects for the relativistic effects caused by deviations in the earth's orbit from perfect circularity and constant gravitational potential.
-
Recommendations: Also provide an estimate of the uncertainty in your timestamps. Avoid UTC. It is trivial to convert the times provided by, e.g., space agencies to TT immediately when you get them and it will save headaches later on. We do it for Chandra (and RXTE). Use TT: it's the official IAU time scale, continuous with ET and the one solar system ephemerides are based on. TAI and GPS are acceptable alternatives, with constant offsets from TT. Use the same reference position for time and space and make sure it is commensurate with your time scale. For instance, when you convert to the barycenter, also convert to TDB. Beware that the barycenter is not the heliocenter. Be specific in labeling the time axis; e.g.: JD(TT;GEOCENTER) or MJD(TDB;BARYCENTER). Use proleptic Gregorian dates for ISO-8601.
-
Do never use: TJD, HJD, BJD, etc. These are not officially recognized and suggest certain metadata values, but leave considerable ambiguity as to what those metadata values actually are. Instead, specify your metadata explicitly. It avoids confusion later on and isn't much more work.
-
What if you deal with incomplete data? If you don't know the time scale and/or reference position, you can provide them as UNKNOWN and set the systematic error/uncertainty to, say, 1000 s. 100 s will do if only the time scale is unknown.
-
What else is there to know? Quit a lot, especially the so-called coordinate time scales (TCG and TCB). Because TDB runs on average synchronously with TT, but in a very different potential well, you may have realized that this requires different values for fundamental physical constants in the barycenter. That's awkward and the coordinate time scales fix that by running at different rates. Eventually these may come into more common use, but at least for my lifetime I assume we will be sticking with TT and TDB. More in the cited A&A paper.
Hope this is helpful,
Cheers,
- Arnold
stscicrawford
FITS WCS Paper IV by Rots et al. http://adsabs.harvard.edu/abs/2015A%26A...574A..36R spells out the definitions and best practices related to time specifications in FITS files, including header keywords that give time stamps, and time axes in tabular or image-based data (e.g. an image cube where the 3rd axis is time). While it is up to SDP to implement these standards for time-related keywords that they populate in level-1b products, we should make sure that any information computed by the barycentric/heliocentric time conversion script, the time columns in photometric catalogs produced from time series images, and the image/spectral cubes for time series observations is compliant with this adopted standard.