LHires III shifts with long exposures - how to handle?

[Peter Somogyi]

Dear LHires III/L200 owners,

I am opening this topic to discuss how to process calibration shifts of long (multiple hours, e.g. 12 x 10 minutes) of exposures.
As I have experienced so far, many of us approaching it several different ways, with no mainstream resolution I'm aware of (whilst we produce such spectra in masses).

Via private discussions, I've met these approaches:
1) neglect - because:
a) it's a hobby / just doing short exposures (such that I understand)
b) let the database owner / professional notice the skew, when checking gathered spectra seriewise (drop/fix)
c) checks it later if ever happened (e.g. tellurics around H-alpha), drop/fix when error above threshold
d) does think it usually never happens with the owned LHires III (I have met 2 people in private mails of this thinking, in the last 2 years).
Antiproof: see the references below.

2) those who handle it in a way
a) ISIS -> Wavelength registration (and do calibration only for 1 exposure)
b) calibration before first, after last exposure - then use their average, accept the skew as a resolution degradation
c) calibration between each exposures (automatic) - a few of us able to do (including me):
i) use sum/average of these calibrations
ii) process individual exposures as spectra in 2D, average them in 1D as a final step (huge manual effort with ISIS)
iii) https://www.astrophoto.at/spectrocalc.html

3) "lateral" (http://www.astrosurf.com/buil/isis/tuto ... ion_us.htm) - special use only, we may possibly never used

For long exposures (= topic of this thread) like 1+ hours of exposures, obviously 2)+ is the way to go.
So far, my favourites have been 2.a) and 2.c.ii), of which I have these problems:
2.a) is useable only when spectrum is high-SNR, and/or have enough spikes to fit on (error is hard to calculate, ISIS does a complex approximation here).
Earlier discussion: http://www.spectro-aras.com/forum/viewt ... ?f=8&t=458
Since error is unknown (can be any - saw strange behaviours), I'd never use it for a precise RV measurement (e.g. CCF) - as the best precision is only the 1st spectra's CCF precision + fitting errors of further spectra. However, for measuring EWs (and RVs in a viusal way) it's still acceptable.

2.c.II): sum or avergae of spectra in 1D manually, causes in ISIS the fits headers to be wrong, need to be edited manually (the DATE-OBS field is what I often mistype, as cannot be copy/pasted in ISIS - unlike all the other fields). I even miss the possibility of a sigma clipping of such an average (correcting spikes, e.g. comsic ray filter often fails when its hits the spectrum - so that excluding that single spectrum at that wavelength region's pixel would be desired, something IRAF can do).
Note: in case of my Nova Cas 2020 serie (made a few times 8 hours), even the Y of spectra have moved (nearby star for guiding + my polar alignment isn't the best), whilst ISIS not finding the Y automatically so need to adjust it manually. One more reason to do it per-exposure.

This sum (or avg) in 1D is not how the way ISIS works by default (same can be true for BASS), so this earlier discussion on aavso is broken (of course, till method 2.b you don't meet such a problem):
https://www.aavso.org/batch-spectra-upload
I suggest continuing discussion here (it was offtopic on aavso). Misunderstanding must have been, as I told I am averaging the results in 1D one by one (using ISIS's sum feature of many spectra, not the images!).

Although 2.b) is better than 1), any of the 2.c) is even better than 2.b), and I consider the 2.c.ii/iii) method even better (something "unfortunately" I can easily produce).

This year I've been introducing too much manual mistakes, and for this reason I'm thinking about switching processing software.
I appreciate mentioning other alternatives, too (e.g. Audela step by step in English, for exactly this use-case with per-exposure calibrations).
For 2.c.iii) (SpectroCalc), I'm curious if anyone is using it, and uploads for which database.
Or, it would be nice only to resolve the fits headers problem in the ISIS sum feature (should not be hard).

References:
http://www.spectro-aras.com/forum/viewt ... f=25&t=459
http://www.spectro-aras.com/forum/viewt ... ift#p14222
http://www.spectro-aras.com/forum/viewt ... ion#p10924
...

Thank you for reading it, and I wish a happy new year for all of you!
Peter

[Jack Martin]

Peter,

My Lhires suffers from the same problem, you are not alone.

So, I use SPOX, and do not touch the Lhires at all. Sometimes I see a wavelength calibration shift, sometimes I don't. It seems works to some degree.

For us, everything in Spectroscopy is a compromise, do you want to get acceptable spectra or spend hours fiddling, I never never got it to work 100 %.

Our equipment will never be as stable as Spectrographs in the big observatories.

Question is, What is the causing the wavelength calibration shift, and how can it be stabilized ?

I am interested in solving this issue, at source, a modification of some kind, that would eliminate this instability.

Not sure the software can fix this, I use BASS.

Regards,

Jack,

Essex UK

[Peter Somogyi]

Thank you Jack for describing your experience!

I never intended to fix the skew itself, and the method 2.c.ii) that I'm doing for a while is already treating that error sufficiently for my needs - reducing the resolution drop to a single exposure's timeframe.

My main problem left here, is only the amount of processing work of this method with ISIS - too much manpower with many mistake possibilities (fits header). This is also why I don't encourage others to tread this path.
Since you have SPOX, you could also do this (guiding gets tricky - needs a deeper setup), however I'm not sure whether if BASS has any support for feeding in N+1 calibrations together with N exposures. Please fix me, if it does!

All the best,
Peter

[Tonis Eenmae]

Peter, thank you for nice list of possible options. I'd comment about them.

If you are for superb quality radial velocity changes or work at the limit of the setup (in terms of wavelength scale accuracy), there are straightforward options:

• The most ordinary method would be bracketing your individual spectra by arc spectra. If it is easy to take arc spectra, I'd go into that direction. If you're sure that the telescope + instrument are not experiencing sudden changes of gravity vector (read: slews, flips), it's relatively safe to interpolate the constant of dispersion function (see that link in your 3) ). Re-identifying en masse is really a desirable tool for all of that.
• If your resolution is high enough (i.e. in the case of LHIRES or some echelle), measure radial velocity or wavelength shifts of isolated telluric lines. Apply that correction to the dispersion function polynomial constant. A modification of that for lower resolution (R<5000?) is: use sky emission spectrum if present and accessible. There is good number of isolated night sky lines with known wavelengths. You can apply your arc dispersion curve directly to extracted night sky spectrum and estimate shifts from night sky lines OR (even more accurate!) construct dispersion curve of night sky emissions.
• Use correlation with telluric template (to be done before heliocentric correction or subtracting heliocentric correction value from correlation result)

If your're not up to very high accuracy radial velocities, it may be possible to use DIB-s (diffuse interstellar bands), there are a number of them and they are pronounced in stars close to the plane of Milky Way. Just shift all the spectra so that DIBs overlap the best. There are few sharpish ones around Halpha, some in visible spectrum and in UV there is Ca K line. Sometimes it is possible (and sometimes even necessary) to combine couple or more methods.

To check the quality of dispersion curve, one can also observe few radial velocity standard stars (there are plenty of bright ones). Analysis of them (e.g. correlation using model spectra, both full spectrum at once and step by step wavelength wise) can/may reveal interesting details about the live of a spectrograph...

I'd use summing/averaging of 1D spectra (well, actually what I'm doing routinely, sometimes with tens of spectra) AFTER all of the calibration steps that can change wavelength scale.

Wishing clear skies,
Tõnis

[Robin Leadbeater]

Hi Tonis,

Bracketing with lamp spectra is the standard routine with the LHIRES and is essential because of the poor stability but even this is not enough as the flexure is not linear with sudden jumps. (In my case some of it comes from the grating holder. One of my holders is much more stable than the other). Another way of getting high precision RV measurements is to superimpose lamp lines on the spectrum using a lamp mounted in front of the aperture. Using this technique I have been able to get to 0.5km/s precision with the LHIRES.
See this presentation (at 18:35)
https://britastro.org/video/13862/14769

The limit then is variations in the line profile shape due to the position of the star on the slit (beyond that a fibre feed is needed, not just for stability but to produce a stable line profile shape)

This does not solve the problem of loss of resolution though when adding sub exposures. The solution as Peter suggests is to treat each sub exposure separately. ISIS does not handle the fits header well though when adding spectra and so it needs to be edited.

Cheers
Robin

[Jack Martin]

Peter,

Put your question to John Paraskeva creator of BASS at;

https://groups.io/g/BassSpectro/topics

Happy New Year,

Jack

Essex UK

[Peter Somogyi]

Tonis,

By practice, we use tellurics only for checking spectra, mostly in case of high SNR + smooth continuum (cooler stars have spikes everywhere, rendering it unuseable). They are too low by SNR for a general use, and dependent on sky quality. However, DIB and tellurics certainly useable at low resolution (e.g. < R~500).
The background sky I usually get, is useable only at certain conditions (lowest resolution e.g. R~500 with full spectrum - bright enoguh pollution with proper lamps, e.g. not LEDs), otherwise they are mostly absent or weak in case of my typical use (R~3000 - 18000, in a narrow band of LHires).

Robin,

Thank you for confirming my experiences, at least now we can gather up and show up a common need for software developers!

I agree to that sometimes there can be sudden jumps during a single (longer e.g. 20-30 minute) exposure. In this case, I can easily spot this jump (always looking at RMS report of ISIS about the averaged calibration) and decide whether want to drop it. Way much better than dropping the whole sequence, maybe the only target for a night.
Since I cannot foretell when such a jump happens (processing only 1-2 days later), I'm always taking calibrations between spectra (15-25 seconds typically).

I did watch now your video at 18:35, though already seen your link about it - and tried to implement quickly (for project: http://www.spectro-aras.com/forum/viewt ... =12&t=2261 ) but got not enough time to implement it good enough (small surface + bright enough source).
This mode is typically useable only for this special type of project anyway, otherwise hard to explain for databases or remove these lines without artifact.

Jack,

I am registering now to that group, and will shoot a question soon.
No rush here, since this is a known problem for at least 8 years, and only a few of us investing extra efforts. If no other option, I will try SpectroCalc (another learn curve).

Clear skies,
Peter

[Robin Leadbeater]

Thank you for confirming my experiences, at least now we can gather up and show up a common need for software developers!

I am not convinced BAAS is the way to go. ISIS does processing in a professional way. We just need it to handle the fits header better when combining spectra. Spc Audace might be an alternative but I have not used it. Benji understands how professionals work and processes spectra in a rigorous way

This mode is typically useable only for this special type of project anyway, otherwise hard to explain for databases or remove these lines without artifact.

I agree, it is not a practical solution for most cases

Cheers
Robin

[Robin Leadbeater]

I am interested in solving this issue, at source, a modification of some kind, that would eliminate this instability.

I saw this problem immediately I built my kit in 2006. There is no simple fix as it is inherent in the design and comes from a number of possible sources. I eliminated all the ones I could (making sure everything is tight, increasing the strength of the springs on the grating holder and remote controlling the lamps.) but it is still not perfectly stable. (I am sure Shelyak would not use the same design if they were starting from scratch.) Spectrographs can be designed to be stable. The ALPY is an example of good design where there is no stress on the important components and it does not move at all. It is such an important parameter that spectrograph manufacturers should test for thermal and mechanical stability and publish the results in the specification.

Cheers
Robin

[Robin Leadbeater]

I eliminated all the ones I could (making sure everything is tight, increasing the strength of the springs on the grating holder and remote controlling the lamps.) but it is still not perfectly stable.

I have just checked how well mine is working. Using the good grating holder (the bad one has play in the bearing) it moved 4um during a 2.5 hour exposure last night so not too bad (it degraded the resolution ~10%)

Cheers
Robin