Accuracy of color index in Stellarium?

[vlaiv]

This is not something directly related to Stellarium as it is not a catalog of stars but it's rather using one to display information on stars.

I wonder how accurate information in these catalogs is?

Here is what I have noticed.

While doing some calculations on color calibration of some of my old data - I'm experimenting with different approaches for further software development, I've noticed that I can't get proper color balance when using my approach.

I checked numerous times if there is error in math or approach and there seems to be none, yet I'm getting wrong color calibration. When computing transform matrix, my resulting star RGB ratios are far away from those obtained by color index from stellarium.

While trying to find out what is going on, I started suspecting color index being correct for my selection of stars - 13 stars in total around NGC7331.

I tried to find B-V index in different catalog but failed to find one with sufficient precision (found only 1 decimal place in B and V mags, so difference will be also 1 decimal place). Then I turned to SDSS which uses UGRIZ system and found some transforms online, and computed B-V varies significantly from those read out of Stellarium. I'm not overly confident in UGRIZ -> UBVRI transform that I used (there are bunch of different transforms and most specific ones require some knowledge of stellar type, I used "general" one that I suspect can have quite a bit of error in resulting values).

However, I decided to do "sanity" check on Stellarium data and found following:

This is data for three different B-V color indices (6 stars in total) that I read of my image. My image is a single stack of about 4h of data taken with OSC camera. Proper calibration was done, and data was stacked and background wiped with care (just linear fit on background pixels). I used AstroImageJ and aperture photometry to obtain values for each of R, G and B channels (even if I did not neutralize background, aperture photometry would still handle this).

First one is two stars of B-V of 0.48 - according to Stellarium, and here are my measured RGB values normalized (3 vector for RGB was normalized to unity intensity):

R, G and B from left to right respectively - first row - first star, second row second star, measured normalized RGB data.

Here is one for B-V 0.79, again two stars with said color index were chosen (according to Stellarium) and photometry on channels performed:

Here is one with 0.35 color index:

For other stars that I used, there is no color index match, but above three examples are enough.

I understand that there will be some mismatch between ratios of two stars of same B-V index due to measurement error and use of different filters, but these are supposed to be same stellar classes (or close) - each pair, and my measurement gives very different RGB ratios for each pair???

Above are not result of math so no error due to math can result in that - just a simple measurement on pairs of stars that should have same (or very similar) RGB ratios according to Stellarium, yet I'm getting very different ratios for each pair (I know that my measured RGB values will differ from sRGB RGB values for each star - hence need for color calibration, but they should be consistent for same stellar types regardless of this, yet they differ).

Any idea how I can do some sort of cross reference check and to find proper B-V indices for these stars, or maybe use another catalog like that of SDSS - but I would need either reliable UGRIZ -> UBVRI transforms or some sort of UGRIZ -> approx black body temperature transform?

 

[andrew s]

Hi @vlaiv, you could try the AAVSO Photometric All-Sky Survey . It has BV and sloan data. At a minimum you could test your transform.

Regards Andrew 

[vlaiv]

On further inspection, it looks like B-V indices in Stellarium are quite off...

I managed to find proper photometric data on one star pair - one marked with B-V: 0.48 in Stellarium using VizieR (I'm still finding my way around it).

For first of two, both UCAC4 and AAVSO/APASS DR9 give B-V index to be 0.803

For second one APASS DR9 gives value of 0.571

Both are quite off 0.48.

Will try to complete color calibration with AAVSO/APASS DR9 data. It would be great if I found a way to do it simply instead of going thru multiple browser pages to reach wanted info for each star

 

[vlaiv]

2 minutes ago, andrew s said:

Hi @vlaiv, you could try the AAVSO Photometric All-Sky Survey . It has BV and sloan data. At a minimum you could test your transform.

Regards Andrew 

Yes, it indeed seems to provide better data - will look into testing it with this.

Do you know any way to easily access this data - something like click on star and read off values?

 

[andrew s]

You have to go to the AAVSO site the navigate to the database and then the access form. You can put in an RA &Dec plus search radius and download a file. 

Not tried it myself!

Regards Andrew 

Ps you can down load the whole file.

Edited June 3, 2019 by andrew s

[Martin Meredith]

I use XHIP for the brighter stars http://vizier.u-strasbg.fr/viz-bin/VizieR?-source=V/137D where both B and V are generally to 2 dp

and Skiff's for the fainter stars http://vizier.u-strasbg.fr/viz-bin/VizieR-3?-source=II/277

which has B-V to 2 dp

and not exactly what you want (but perhaps useful in your venture at some point) is Skiff's Catalogue of Stellar Spectral Classifications 

http://vizier.u-strasbg.fr/viz-bin/VizieR-3?-source=B/mk&-out.max=50&-out.form=HTML Table&-out.add=_r&-out.add=_RAJ,_DEJ&-sort=_r&-oc.form=sexa

You may find other resources at this site too.

Martin

[vlaiv]

Thanks, I'm actually using that - here is my current procedure - not quite one click but it seems to work.

In AstroImageJ I did plate solve via nova.astrometry.net. Right click allows for object to be found in SIMBAD database. There is VizieR photometry widget on right where you can search within couple of arc seconds from a given coordinates.

This gives search in photometric data with a table at the bottom - where I find AAVSO/APASS data and follow the link for particular record - there it provides B and V to three decimal places.

 

[vlaiv]

This is much better now

Color correction seems to be able to solve even me mixing up R and B channel (wrong bayer pattern)

Here is list of 16 stars - left are RGB triplets of reference values (unity vectors), and right are RGB triplets of measured and corrected RGB values. There are some small deviations, but in general I would call this good correction:

Some of the stars used have fairly high uncertainty in measured B-V index, and there is probably some error introduced by B-V index -> temperature conversion (I did not use actual BVR filter curves to find temperature for corresponding B-V index, but rather simple formula found online).

[dph1nm]

This might be useful for you

https://archive.stsci.edu/prepds/atlas-refcat2/

It is probably the most accurate stellar catalogue around. You will still need to convert to B-V though!  https://www.sdss.org/dr12/algorithms/sdssubvritransform/ should do the trick - I would be pretty confident in these transforms.

NIgelM

[vlaiv]

2 hours ago, dph1nm said:

This might be useful for you

https://archive.stsci.edu/prepds/atlas-refcat2/

It is probably the most accurate stellar catalogue around. You will still need to convert to B-V though!  https://www.sdss.org/dr12/algorithms/sdssubvritransform/ should do the trick - I would be pretty confident in these transforms.

NIgelM

I'll need some help with that if you are up for it?

There are number of transforms listed and is there a way to derive one general formula?

Some of transforms are not quite logically consistent and that is confusing me. For example:

Jester et al. (2005) ugriz ->UBVRI star transforms:

- two different conditions are listed, but Rc-Ic < 1.15 is redundant. There is no need for it to be listed if transforms cover all star types - U-B condition is enough. If on the other hand Rc-Ic is genuine condition - transform set is not complete and does not cover all stellar classes

- I'm trying to calculate UBVRI values and proper transform depends on U-B condition that I yet don't know! Which transform should I use? Solution might be - use both and validate original condition, but there is off chance that in some cases either both will satisfy original condition or none will satisfy original condition - will need to check if it is possible. It is actually not possible - since there is discontinuity and not overlap - calculate and check seems like plausible strategy.

Jordi et al. (2005)

- two different transforms are given for same expression (which one to use)

- There is mixture of letters which confuses me (UBVRI and ugriz seem to be mixed on lhs - is this correct)?

I'm also having trouble distinguishing between UBVRI and UBVRcIc (from what I gathered they are the same thing?) and ugriz vs u'g'r'i'z' - which are two slightly different systems (small difference in response curves, otherwise very similar)?

[vlaiv]

2 hours ago, dph1nm said:

This might be useful for you

https://archive.stsci.edu/prepds/atlas-refcat2/

It is probably the most accurate stellar catalogue around. You will still need to convert to B-V though!  https://www.sdss.org/dr12/algorithms/sdssubvritransform/ should do the trick - I would be pretty confident in these transforms.

NIgelM

Actually, now when I examined the data in catalog, I don't need to do transforms to find B-V as catalog itself contains estimated temperature for Mag>17 and temperature range of 3000-10000K

Quote

We have used these three broad bands to infer stellar effective temperatures, Teff, for all sources brighter than G=17 mag with Teff in the range 3000-10 000 K (161 million sources).

Taken from https://arxiv.org/abs/1804.09374

And readme of catalog has following entries:

Although this does not address concrete topic of this thread (finding color index), it does solve my problem - finding stellar temperature for colorimetric calibration. I think some download and trimming of the data is in order. There are stellar positions as well which will be a good basis for plate solver

[andrew s]

Not sure how important this us to your calibration but in spectroscopy we have to find catalogues with spectra that have not had interstellar reddening removed or if it has apply the reverse transform. 

If it is important have a look here http://www.spectro-aras.com/forum/viewtopic.php?f=6&t=207&p=2159&hilit=Berardi#p2159

Regards Andrew 

Edited June 4, 2019 by andrew s

[vlaiv]

24 minutes ago, andrew s said:

Not sure how important this us to your calibration but in spectroscopy we have to find catalogues with spectra that have not had interstellar reddening removed or if it has apply the reverse transform. 

If it is important I can point you at the relevant discussion.

Regards Andrew 

Huh, did not think about that. It really depends on what one wants out of color calibration - should it display colors of a target as seen from earth (outside of the atmosphere), or should it display colors of object "in clear line of sight" - in case of galaxy like we are standing some distance away from that galaxy for it to both appear large enough and bright enough.

I guess most people would choose second option - this means reddening removed, but good point, maybe at some point in time I should have both options in software

Interesting question, and a food for thought. It started my mental process and I ended up thinking about GR and gravitational shifts .

Do we account for position of the star in the sky when doing spectroscopy? Is this effect noticeable? I'm talking about difference between stars that are towards galaxy center and ones that are on the opposite side in relation to our Sun. In first case light will be "escaping" gravity potential of much of the galaxy - so it will be red shifted, while second case should be blue shifted as light is "falling" towards the galaxy?

[andrew s]

@vlaivif you want to get very subtle then there is always atmospheric transmission to take into account 🤔  

[andrew s]

7 minutes ago, vlaiv said:

Do we account for position of the star in the sky when doing spectroscopy? Is this effect noticeable? I'm talking about difference between stars that are towards galaxy center and ones that are on the opposite side in relation to our Sun. In first case light will be "escaping" gravity potential of much of the galaxy - so it will be red shifted, while second case should be blue shifted as light is "falling" towards the galaxy?

Radial velocity dominates any GR effect. RV causes small shits seen in line positions. You have to account for the motion of the earth but once that is done you can follow pulsations and other shift e.g. material ejection of a stars photosphere. With SN you get both RV shifts and delays due to the metric expansion.

Regards Andrew

[vlaiv]

8 minutes ago, andrew s said:

@vlaivif you want to get very subtle then there is always atmospheric transmission to take into account 🤔  

I don't think that is an issue with this approach that I'm using - it is supposed to correct for both atmospheric and instrument response.

In most images that we are concerned with, air mass is pretty much constant (except very wide field photos with short focal length lens - but these will have different issue - too dense star fields to do this sort of calibration). Transform should account for extinction vs wavelength in pretty much the same way it does for the fact that instrument is recording photon count and color is defined in energy flux (wavelength dependency between the two).

[dph1nm]

On 04/06/2019 at 13:20, vlaiv said:

I'm also having trouble distinguishing between UBVRI and UBVRcIc

Sigh - astronomers are very sloppy these days! Strictly, RI should refer to the Johnson system and RcIc the (Kron-)Cousins system. They are (quite) different filters sets and numerically produce colours which may differ by several tenths of a magnitude. However, I suspect many people now use RI to mean the Cousins system. Anything based on Landolt stars (e.g. Jordi et al) is RcIc.

On 04/06/2019 at 13:20, vlaiv said:

two different conditions are listed, but Rc-Ic < 1.15 is redundant. There is no need for it to be listed if transforms cover all star types - U-B condition is enough. If on the other hand Rc-Ic is genuine condition - transform set is not complete and does not cover all stellar classes

I think this is just a of what subset of stars were used to derive the fits. So if you add the words "Fitted to" in front of the descriptions it makes more sense e.g. "Fitted to all stars with Rc-Ic<1.15". You will probably find the differences for stars with U-B<0 are very small (at the level of 0.01-0.02 mags).

On 04/06/2019 at 13:20, vlaiv said:

- two different transforms are given for same expression (which one to use)

- There is mixture of letters which confuses me (UBVRI and ugriz seem to be mixed on lhs - is this correct)? 

Again I think this is a question of subsamples of stars used for the fits. I would stick to the general one rather than the populations specific ones. The mixture of upper an lower case is correct - just done for convenience I think - some of the transforms are for colours (e.g. V-I) and some for the difference in magnitudes between similar bands (e.g. r-R).  Hopefully the equations are all consistent!

Behind all this is the fact that these transforms are not really linear, and ideally you should also be using second order terms or higher. But mostly people just rely on linear transforms, with maybe a change of slope somewhere.

NIgelM

[vlaiv]

9 minutes ago, dph1nm said:

Behind all this is the fact that these transforms are not really linear, and ideally you should also be using second order terms or higher.

I did wonder about that. At the moment I'm doing linear transform between recorded triplet and calculated value in linear RGB space (or XYZ - same thing because there is linear transform between the two). Simple least squares fitting.

I wonder if I could improve fitting by using higher order transform? It would obviously require more calibration stars, and I have a concern about skewed ratios. Affine transform ensures that vector length ratios between two spaces is preserved. I'm not doing this for any sort of measurement - rather it is for getting "true" color, but I wonder if non linear terms will do some funny hue shift or similar.