-
-
Notifications
You must be signed in to change notification settings - Fork 204
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Values of ∆T = TT – UT1 overestimated for year > 2022 #519
Comments
I agree! Skyfield should, specifically:
It would be nice, while accomplishing all that, to not slow down Skyfield too measurably, even though spline interpolation will be more expensive than linear interpolation. If you would like to see my work on the problem so far, check out: https://github.com/skyfielders/python-skyfield/blob/master/design/delta_t.py You can see that I made some progress back in October, but then switched projects to polar motion — once I'd written the code to parse finals2000A.all, which also provides the polar motion offsets, I figured that I should go ahead and tackle a longstanding issue that involved polar motion. Thanks for opening this issue, it provides a useful way to track further progress. And feel free to weigh in with any design thoughts of your own towards the new spline-interpolation feature! |
I will for sure look at your work on If I got it right, the sources of ∆T in http://astro.ukho.gov.uk/nao/lvm/ are
(All these approximations are obtained by a least squares fit of the historical data available from -720 and forward.) All these expression give raise to different ∆T and have a different associated error, so it makes little sense to me to stitch them together. The user should therefore choose the most appropriate solution, depending on the application and the range of years of interest. If an "automagic" value is desired, the library should
|
I've computed also estimates of ∆T. As what regards 2. (long term extrapolation), it is obtained by integration of the corresponding LOD expression. I've only skimmed through the paper, but by comparison with the figures in http://astro.ukho.gov.uk/nao/lvm/ it seems to me that the integration constant has to be determined by continuity with the spline approximation. Moreover it seems that this expression should only be integrated forward, and not backward in time. Nevertheless here are some comparisons. The overall picture looks good: However a zoom on the 1500–2200 period (omitting parabola 1) shows that Skiefield extrapolation is not satisfactory: If we look at the 2000–2500 extrapolation this is the picture: Finally the difference between all approximations and parabola 1. (hidden because very distracting) ConclusionWhat to do for values of ∆T in the past, I don't have a strong opinion now. But for extrapolation in the future, I have no doubt that the most sensible approach would be to integrate the expression 2. of the LOD starting from the last available point in the IERS file. If you agree, I can provide a PR in this sense:
|
(I'm marking this as a "feature request" to remind myself that it will require new code, to supplement the current long-term ∆T with more polynomial segments.) |
@miccoli — I might have a bit of time in my schedule now to dig into this a bit further and more fully respond. At the http://astro.ukho.gov.uk/nao/lvm/ page, I now see:
So it looks like there are updated polynomials to use for the past ∆T. It looks like for length of day from 2020 to 2500, they now integrate this expression: lod = +1.72 t − 3.5 sin(2π(t+0.75)/14) where t is the number of centuries from +1825.0. I kind of wish they had a spline instead, for symmetry with the way they handle the years earlier than that. What approach were you thinking of — were you going to build a linear interpolation table out into the future with the integration? Or use the formula itself? It's an interesting idea to put the integrated curve out at the end of the final observation in the IERS file. Do we really want someone's code that's, say, predicting solar eclipses to change the details of its predictions with every little wiggle of the most recent day's IERS number — which then changes over the weeks that follow as they do smoothing? Or should the integration constant "anchor" be one that wiggles around a bit less? |
Are my eyes deceiving me, or do Morrison, Stephenson, Hohenkerk, and Zawilski themselves choose to position their long-term curve independently of their final interpolated value, instead of integrating forward from it? It looks to me like their long-term curve does not meet the tip of their final spline: (From http://astro.ukho.gov.uk/nao/lvm/ in the 2020–2500 section) |
This is the image captured from https://web.archive.org/web/20200218152225/http://astro.ukho.gov.uk/nao/lvm/ which I analysed for "guessing" the integration constant for the 2018 long term ∆T prediction. The 2018 expression of the LOD was lod = + 1.78 t - 4.0 sin(2πt/15) Probably my assumption of continuity was wrong: looking very closely there is a hint that the last spline point (2016) is slightly above the extrapolated curve (2010–2200). But somehow in the 2010–2016 interval both the spline and the interpolated curve appear to be very close, so it is hard to guess how the integration constant was really determined. However the graph that I obtained by continuity with the last spline point was remarkably in good agreement with the 2018 Morrison et.al. graph. But with the 2021 graph, there is no doubt that my assumption was wrong! As what regards the new situation, I had no time to properly analyse the new data but here are some quick remarks.
But for now, I have not yet elaborated on how to bridge between IERS data and long term prediction. |
Well, that took a few weeks longer than I had expected, but after investigating the options I have just released Skyfield 1.38 with what I hope you will find is a gentler ∆T curve over the next few decades. Note that I opted not to experiment with the new Let me know if you see any problems, or have difficultly getting out a curve that looks like the one above! |
Kudos for the new code, which is open to a lot of possible user customisations! First a general remark on the
and
I initially thought that the So I agree that the skyfield v1.38 implementation is a very sensible one. Just a few remarks on some minor details.
example code
|
The
I can't think of any problems offhand, as they are both good sets of numbers based on very recent analysis, but if you come across any problems I'll definitely want to hear about them.
Drat! I forgot about that one, and the graph I was using to look for discontinuities obscured that one since it was right up against the noise from the IERS daily values. As you can see above, I've committed a fix; if you'd like to try it out yourself ahead of the next Skyfield release, you can:
|
The estimated values of ∆T seem to me to be greatly exaggerated for years > 2022.
Running this example
yields
and of course it is not very likely to have an increment of ∆T of approx 2s / year in the near future. (This would amount to 2 leap seconds in 2022 and 2 leap seconds in 2023.)
I produced also some graphs to better analyse the problem.
LOD = length of day (which is the derivative of ∆T)
LOD - 86400s = difference between 1 SI day and 1 UT1 day
matplotlib program
Trend of ∆T in near future:
matplotlib program
Trend of LOD in near future
Long term trend of LOD:
I read #452, and it seems that current behaviour arises from the desire to "quickly" recover the extrapolated long term trend of ∆T (Stephenson, F.R., Morrison,L.V. and Hohenkerk, C.Y. (2016)).
But if we look at the resources in http://astro.ukho.gov.uk/nao/lvm/ it seems to me that we can expect that al least up to year 2500 the LOD will be well below the linear regression LOD obtained in years -2000 to 2000 (and this means that ∆T will also be below its long term trend.) On the contrary the Skyfield routines assume a dramatic step increase of the LOD in the near future, so that the long term ∆T will be almost recovered in the next century.
The text was updated successfully, but these errors were encountered: