Skip to main content

CCD internal checks for gross errors?

28 posts / 0 new
Last post
lmk's picture
CCD internal checks for gross errors?

Ok, I rarely post to this forum because I am a visual observer, but had dabbled in CCD a while ago.

The purpose of this thread came to my mind due to the recent problem of substantial offsets observing the same variable (CH Cyg) by different CCD observers. Something like several tenths of a magnitude! Thats even worse than typical visual observers. This problem has been noted with several other variables in the recent past as well (NSV 1436).

It would seem to me that gross errors like this could be caught by the software before the results were published, if more than one comp star is in the field? The software should check the expected value of each of the comp stars based upon the others, using some kind of good algorithm. Any anomalies should then be red flagged as a problem somewhere in the reduction, comps or technique. The variable's observation should then be withheld until these internal/systematic errors are resolved and the algorithm computes the correct estimate for all comps within a certain error level.

Is this something thats already available, but not used by observers, or does this feature need to be added?

Seems like this kind of internal check of all available comps against each other would resolve most of these issues of grossly aberrant CCD observations. Of course, this implies several comps should always be used. Submitting any observation using a single comp star can be quite dangerous!

Mike LMK

FRF's picture
leak in the V filter, variable as check star, unofficial comps

Hi Mike,

I think this time there are multiple reasons for the substantial offsets and the scatter.

Most of the offset belongs to one observer (HMB), who's V filter seems to have some leak in red emission lines, as it turned out ca. half years ago in the case of T Pyx (?) - if I can recall correctly.

Several observers use 'comp 94' [000-BCG-651] as a check star, although it is an eclipsing variable: V2365 Cyg and it is clearly written in the comp star database that "do not use for CCD or PEP." (Well, they use this star only for "check star" (or 'comp2'), but I still wouldn't recommend.

There is a PEP observer who use two comp stars (SAO 31586 and SAO 31737) that are not in the VSD (and I don't remember whether they were used as comp stars at AAVSO ever). On the other hand it is not clear what a V values he is using for these comp stars since it was not mentioned in the Quic Look File. Especially since the former is a quite red star (B-V=1.48) so shouldn't be used as comp star and (therefore?) it's Vj mag listed in the HIP catalogue is ~0.18 and ~0.13 mag brighter than the VT or Hp magnitudes respectively.

hambsch's picture
V filter differences


I am the observer with code HMB.

Indeed my V filter (from Astrodon) had been compared to somebody elses V filter from the same observing location showing for T Pyx some offset by about 0.2 mag.

I have together with Tom Krajci measured standard field to get some info about the filters I use. Here ios a copy of the results from Toma analysis of the images from SA98:

First, the easy stuff. The slope of B-V versus V-v is -0.070. The V filter has a central bandpass that's a bit 'blue' of the standard. (This means your V filter will see redder stars as slightly fainter than the standard, and bluer stars as slightly brighter...but this can be corrected as part of the transform process.)

The slope of R-I versus I-i is -0.101. The I filter has a central bandpass that's a bit 'blue' of the standard.

Because both filters are slightly 'blue'...then the resulting 'span' of the V and I filters is still about the same 'span' as the standard V and I passbands, and the plot of V-I versus v-i is 1.018 ...which makes sense based on the above info.


According to this I did not think there is an issue with the filter.

I also used as comp star the one which has now an indication of being variable. Of course at the time I choose it it did not have this info. I wonder why it is not removed and the observes notified (there are not that many observers using it).

Anyway, I am not very happy that this thread is called as it is in the very first mail.

There is always the possibility that magnitudes change for the comp stars and that could be accounted for. I was also not aware of photometric filters not following the standards as it seems for Astrodons. Don will probably not be happy to hear that. If somebody knows of suppliers also supplying 50x50mm^2 real photometric filters, I love to hear that.


CH CYG PEP Comp Stars

SAO 31586 and SAO 31737 are the designated comp and check stars, respectively, for PEP measurement of CH CYG according to the starparm.dat database (2010 Sep 29 version) used by all PEP observers reporting observations via PEPObs or WebObs. If these are inappropriate, the database needs to be changed, otherwise they will continue.


TCB168's picture
Checking data


I can't comment about the variable you mention as I don't follow it but can comment about the ability to use the comp star data.

When a CCD observation is uploaded, the measurement given for the comp and check stars are just instrument values. They can be any number depending on the offset added by the software you are using and don't correlate with the catalogue value in the VSP. The only useful info would come from the difference between the comp and check values and this could be correlated to the catalogue values in the VSP. This would only work for single star comps and would not be possible if the data is submitted as an ensemble.



HTY's picture
When you use VPHOT

Just as a point of clarification, when you use VPHOT and you use an ensemble of comp. stars, the offset is added, the magnitude of the check star is computed and can be compared directly against the catalog value.


HQA's picture
CH Cyg comp stars

The two PEP comparison stars are ok; CH Cyg is a very red star (B-V = 1.6), so choosing a red comparison star can be ok, especially if that star is "red" because of interstellar reddening and not an intrinsic color.  Also, the PEP data is transformed.  We were going to put all of the PEP comparison stars into VSD, with a note indicating that they were for the PEP program, but that may not have been done yet.

Regarding Astrodon filters: the transformation coefficients that we get on other Astrokolkhoz telescopes are very small and close to the standard system.  This is across multiple filter sets and different CCD cameras, so I'm satisfied with the results.  Perhaps the one transformation attempt done by Josch and Tom was slightly in error; perhaps the camera/telescope are somewhat different than the norm.  It is always best to determine your coefficients on several nights and average; if you are determining them just by looking at single fields, then use different fields on different nights to remove as many systematic sources of error as possible.

Mike suggests one method of cross-checking observations.  There is no one correct method, each has a parameter space in which it will work and another where it will fail.  It would be valuable for all observers to double-check their observations before submission to at least catch the obvious mistakes.  We also need to work on improving basic accuracy with techniques such as transformation.

One thing that I was thinking about - VPHOT accesses VSD to list its comparison stars, but I bet that it does not look at the "notes", such as the one for V2365 Cyg.  We need some way to automatically remove these sequence stars that work for visual observers but not for CCD observers.


hambsch's picture
Red comp stars

Hi Arne,

How could you argue that our transformation attempt is slightly in error or the camera telescope system is outside the norm.

I am using an FLI ML 16803 based CCD camera and an Optimized Dall Kirkham (ODK scope from Orion Optics. So nothing special I would say.

I have now tried to use both comp stars in the range B-V 0.2 to 0.4 and more than 1 as given in the AAVSO sequence. In both cases, I get similar results for the magnitude of CH Cyg, but still the offset psersists.

I think with the help of Tom, we will get more data on different calibration fields.

What if all you guys are in error of 0.2 mag? Who knows?

KTC's picture
Gross problems - absent. Small problems?...unknown at present

Before Josch's rig was set up, if I remember right...I tested the CCD for read noise, gain, dark current, and linearity.  No problems there...matching manufacturer specifications.

I have looked over some of his images of a Landolt field or two in B V and I.  It was easy to see the gross problem of a massive red leak in the B filter...and Josch has a replacement for that which solves the red leak problem.  (And yes, we'll verify that upon install.)

The V and I filters...certainly did not have gross problems.  But my analysis was only a way to compare the effective central bandpass of his two filters...compared to standard magnitudes.  (Yes, but what about the filter 'wings'...especially for extreme-color targets?)

Are there smaller problems present?  At this time I don't know.  More testing is in order.  (Heck, more testing is almost always in order.)

I'm always learning more.  Every time I take apart a scope, test a CCD, or learn a new scripting language...I gain a little better understanding of the underlying processes at play.

We'll continue to make improvements as we test and evaluate.  Chile has lots of clear nights.

(And if nothing else...if there's an's probably constant.  That may be a 'consolation prize', but if there are overlapping data sets from various observers...this is often an acceptable workaround.  And if the offset is not constant...then even more testing is in order! ;-)

Here's a not so subtle request:  As more and more automated scopes come online...the data tsunami gets bigger.  It would be great to receive some help in testing and analysis.  (Or at least give us some ideas on how to better automate the testing/analysis...that appears to be one of the long poles in the tent.)

Thank you in advance.

lmk's picture
A better way to do the measurements?


I have now tried to use both comp stars in the range B-V 0.2 to 0.4 and more than 1 as given in the AAVSO sequence. In both cases, I get similar results for the magnitude of CH Cyg, but still the offset psersists.

I think with the help of Tom, we will get more data on different calibration fields.

What if all you guys are in error of 0.2 mag? Who knows?


Well, I never meant the title of this thread to disparage any particular observer. Its simply a statement of fact that there are substantial offsets between CCD observers for quite a number of stars, not just CH Cyg or T Pyx. I am sure if other stars were also examined like this, a similar problem can be found.  There is nothing new about this problem, I have brought it up several times in the past several years.

HMB might conceivably be more correct than everyone else, but I think common sense would tend to weigh against that conclusion. Esp. considering HMB's data is even further offset from all the visual observations.

Clearly transformations issues (or lack of) is responsible for many of these problems, esp. the red stars. And there may be more subtle causes too, as Tom mentions, and which I am not familiar enough with to comment on.

But, I think anyone would have a hard time arguing that offsets of several tenths of magnitude is not a "gross" error for CCD. Its even bad for visual. CCD really needs to come closer to its potential high level of accuracy. Hopefully, Arne's CHOICE classes will soon start to make some significant headway towards fixing these long standing issues.

One suggestion I can make - Apparently many CCD measurements are made using just one comp star, and it may even be several magnitudes different than the target. Visual observers could never make decent estimates using such extreme extrapolations. Certainly CCD is a lot better in this respect, esp. if transformed, but still it seems that any extreme extrapolations is asking for trouble. Maybe CCD observers can adopt the same technique used by visual observers - bracket the variable using one comp above, one below, and keeping the difference in magnitudes reasonable (< 1 mag or so). Then, the software could compare the extrapolated measure of the variable coming in from both directions, and see if the results are close or not. Even taking the simple average of the two bracketing measurements would have a marked increase in the accuracy of the result. (For the same reason the Runge-Kutta numerical methods are several orders of magnitude lower in errors than the simple Euler method).

Well, since i do not know the internal details of all the reduction softwares out there, maybe some already do this way? Does VPHOT do bracketed estimation too?

Mike LMK

HQA's picture
comp star magnitudes

LMK suggests that using comp stars that are "magnitudes" different from the target star will potentially cause problems with CCD observations, and recommends bracketing the target star with two or more comps to avoid any such problem.

First, I need to repeat that there are nice stars to measure, and there are pathological ones.  T Pyx had a huge Halpha line; CH Cyg is a symbiotic star, with a spectrum that mostly looks like an M star with molecular bands, along with a strong, but narrow, Halpha feature.  A good spectrum of CH Cyg can be found at Christian Buil's site: (scroll down to the bottom to see spectral plots).  We must remember that the V filter is not a square response function, but more like a Gaussian, with a peak near 550nm but a long red tail that stretches out to Halpha.  Because of this, anything that does not look like a smooth black body, and anything with an emission line that lies within the bandpass, will be measured differently by different people.  The same thing happens to the eye, which is why some observers will always measure fainter or brighter than others on the same star.

Assuming that all other problems have been corrected (and remember, CCD observing really is harder than visual observing, so give those observers who use digital techniques some credit), then you can get observer 1 lined up with observer 2 with some effort.  First and foremost is to transform your data - that gets you most of the way onto the standard system.  Then, for highest accuracy, you really need to synthesize what your response would be to the typical spectra of a given object - that is, you need a template of what its spectrum looks like to provide an extra "transformation" to match multiple observers.  Most researchers don't go to this extreme (the only one I know is Ulisse Munari with his ANS consortium), but instead just ask observers to get good coverage of an object so that they can just shift one observer's light curve with respect to another and mesh them by just using a constant offset.

So for pathological cases (and I consider a symbiotic star to be pathological), you will always get some "scatter" caused by simple offsets between observers.  For CH Cyg, what we currently see is a combination of such offsets along with more mundane errors and systematic effects, such as basic transformation, different magnitudes for comp stars, etc.  The light curve generator is not adequate for displaying the estimates in the detail necessary for analysis; it is designed to do a generic job for multiple filters, visual observations, etc.

Back to the original statement... a CCD detector is extremely linear; extrapolating with it is NOT the same as extrapolating with the eye.  There are many fields where a bright comparison is just not available, and so using fainter ones, but exposing properly to keep both target and comparisons in a linear range (not saturated, good signal/noise) can often be done with high accuracy.  I do try to bracket any target in every way that I can (magnitude, color, spatial layout in the field, etc.) to randomize any possible systematic error, but it is not always possible.  Something as bright as CH Cyg doesn't always have a brighter comp star for such bracketing.  So Mike's suggestion of bracketing is good, and using multiple comp stars is also good, but are fine points of the discussion.  Get rid of the easy errors first.


KTC's picture
What about Sloan filters for 'pathological' photometry?

We must remember that the V filter is not a square response function, but more like a Gaussian, with a peak near 550nm but a long red tail that stretches out to Halpha. 

Are Sloan filters better in this regard?  Their response is closer to a 'box'.  (And, I have not heard yet of reports or complaints of any Sloan filters that suffer from red leak.)

Roger Pieri
Eps AUR offset

There is an example of offset between Eps AUR observers in the eJAAVSO Karlsson's paper. This is 20 observers using either PEP, CCD or DSLR, various transform techniques and very different comp stars. The peak/peak offset is +/- 12 mmag with a sigma of 7.5 mmag. So bad ? 

Clear Skies !


HQA's picture
Sloan filters

Sloan filters have much steeper wings, and so in that respect, give more consistent results between observers.  There have been red leaks in these filters as well, especially the "Gen 1" u' filter (Astrodon's Gen 2 filter addresses this problem).


WGR's picture
Looked at LCG


I just looked at the LCG for the past 100 days for CH Cyg.  What I see here is the following:

1.  I see Visual Scatter of  0.7 (with p-p = 2)

2.  The CCD Scatter is 0.2 mags on what appears to be 2 sets of data, offset by about 0.2 mags.  This could be as has been pointed out, transformations or sequence values.

3.  Visual data and CCD data should not necessarily plot on top of each other, there should be an offset--See Dick Stanton's work of a couple of decades ago.

4.  I don't see how one would know which of these are "Gross" errors, based only on the submitted data?  Any suggestions?



WGR's picture
"Gross Errors" in CH Cyg Found


The "Gross Errors" in the CH Cyg light curve peaked my curiosity.  I did 2 things:

1.  First I Transformed the HMB data give the Coef provided and the B-V = 1.6 of the object.  THis made the HMB data and the other data come together by about a 0.1 mag of the 0.2 mag in question.  A small improvement.

2.  Second, I looked up Stanton paper and transformed the Visual data to V mags.  This turned out to be = 0.21 * (B-V) = 0.4 mags.  Attached is a plot of the curve with the Transformation.  This is a big improvement.  The CCD data basically plots right on top of the Transformed Vis data--as it should be. 

3.  I did not have transformations for the other v observers.  If they will provide Tb, I will be glad to transform the other data.  Suspect that it will make the V CCD Data converge even more--but don't know that. 

4.  So I don't see any more Gross Erros in this plot.  Perhaps we can now get off this CCD vs Vis thing, and hopefully both the CCD and the Vis observers have learned something--namely Vis data and v data should not necessarily plot on top of each other. 


WGR's picture
Plot of Xformd data

Sorry, but I am afraid I am unable to either paste the Plot into this reply, nor include it at a file attachment.  If  someone know how to do this, I will be glad to include the plot.  Its quite informative.  If you send me you email address, I will be glad to send it to you.




WGR's picture

I figured out how to include the plot of Xformed CH Cyg Data




lmk's picture
Raw vs. transformed

Nice plot Gary! If you throw out the extreme outlyers of the visual data, the correlation of everyone looks really great.

But the reality of the database is that its a mix of mostly untransformed with some transformed data. Clearly, that makes the raw data look pretty messy. As far as I know, no visual observers "correct" their estimates using the Stanton relation before submitting. I think AAVSO policy is not to do that.

Even though the Stanton study is a good one, and the 0.21 coefficient is a reasonable average fit, there is quite a bit of spread between individual visual observers. So, it may not be a transformation that should be universally adopted as yet.

Why the HMB data is still offset after the transformation remains a mystery. I hate to sound like picking on this one observer, but digging deeply into what he is doing may uncover something to explain why his data is offset consistently from other CCD and visual with this star and others as well.

Mike LMK

WGR's picture
Raw vs Transformed

Hello Mike

You are correct, actually in this case, none of the data in the past 100 days was transformed, and it is customary for the AAVSO database.  Transformed measurements are encouraged, and we have talked about doing that in the future.  This is true for Vis as well as BVRI.  I believe that PEP is transformed--can someone confirm that?

The Stanton value of 0.21 is indeed an average of 46 observers as I remember.  Its probably a good average, but some observers could be quite different.  In fact, I looked at transforming all the Vis data to see if the scatter is reduced, using individual coeficients.  However, there is a problem:  Of the 459 Vis measurements submitted over the past 100 days, 243 of them had a wrong comp star designated in the data base.  There were 58 random errors, hard to tell.  There were 104 obsedrvations using the 74--but I could not find that label in the B size chart or the PT/sequences going with it.  I suspect that observers are using old charts--there is a 73 star that I suspect that they mean.  The same condition exists for 32 observations using the 84 label.  Again no 84 but there is an 85.  Probably old charts, but I don't know that.  If I were using this data for a research project, I would have to either follow up or throw out half the data!  That's not good.  Observers work hard to get this data and they like to have it used.  I urge all observers--including ccd observers, to make sure you are using the up to date chart for CH Cyg.


I also believe that the ccd data other than HMB should be transformed to see how it aligns with the other data--to tell the complete story.  I suspect that their coeficient has the opposite sign from the HMB transform, and that would help align the data.  If observers of CH Cyg send me their Tb or Tv, I will add that to the chart.  In particular, the prolific observer JM, I would love to transform your data and see what happens.


This comp star issue also affects the ccd data.  Imagine what to do when you look at the color of the comp star "ENSEMBLE".  Again, just throw it out, or assume that the average color of the ENSEMBLE is some value and use that for transformation. 


The combined data looks good to me.  I agree it can be better, and scatter is something that I spend a lot of my research time looking into.  Thanks Mike for bringing this up, shows how we can all do better.






hambsch's picture
Ch Cyg data


from my Southern remote obervatory, Ch Cyg is at very high airmass. So all observations are done at airmass larger than 3. This seems to be the reason that the data differ from other observers where the star is much higher above the horizon. For Ch Cyg, I could remove my obervations as other observers cover it in depth. For other stars that might not be the case.

I would like to know however from LMK what other star he meant where I differ according to him from other CCD observers (except T Pyx).

There were the issues with S Dor and HD 269006, where the data found in internet (VIEZER) for the I band where off compared to what other observers used. I got in contact with the respective observer and am now using the same comp stars giving a similar result. Hence for these two stars the issues with different magnitudes especially in I band are solved.



WGR's picture
Extinction drops out for same field?

Hello Josch

I could be wrong, but with differential photometry, the extinction terms drop out, so I don't think thats the answer.  I still would like to see some other observers of CH Cyg send in their Tb and Tv transformation coef.




Mark Blackford
Mark Blackford's picture
Differential extinction

Hi Gary,

under "normal" CCD differential photometry observing conditions the effects of differential extinction are very small and can be safely ignored. However at high airmass the target and comp stars must be very close together and have very similar colour index.

Josch is observing at airmass over 3 and CH Cyg is very red. I doubt the comp stars are close in colour so differential extinction may well be a big contributor to the difference in reported magnitude. Cheers,


WBY's picture
Differential Extinction


I agree. The second order extinction term starts to have a perceptible effect at high airmass if there is a significant difference in color between the comp and target star.

v0 = v - k'v*X - k"v*(b-v)*X

or the equivalent for a color index

c0 = c - k'c*X - k"c*c*X.

With a very red star and one that is "pathological" with a strong emission line in the tail of the filter curve, it is very likely that comp stars will vary significantly in color. So you probably need to do your differential photometry after extinction correction using the second order term. Most of us don't do that. We depend on Mt-Mc+Rc (Rc is the sequence "reference" magnitude) cancelling out the effect of extinction.  The other thing that you might have to worry about is that with a strong emission line and strong absorption bands in the spectrum, your "normal" second order coefficient for nice stars may not apply. It would be interesting in this particular case to see what effect including a second order term would have. Unless you have this scripted in, it makes the process much more intensive since you need to do it manually using a spreadsheet.

WGR's picture
Include second Order Ext, but what sign?

Hello Mark and WBY

I agree with Mark and WBY, that the second order extinction is appropriate for CH Cyg, with B-V = 1.6 and Air Mass = 3.5.  As a mater of fact, with a second order extinction coef I found in Henden and Kaitchuck of -0.041, this effect works our to be .23 mags--assuming a B-V of 0.0 for the comp star.  However, it seems that this is a brightening effect, which would take the high air mass data and separate it more from the low air mass data.

The reasoning is as follows.  Assume that CH Cyg and the comp star are both 7th mag, and that the comp star is B-V = 0.  So the secondary extinction of the comp will be zero.  The secondary extinciton of CH Cyg will be something like the number above.  The secondary extinction of CH Cyg will make it appear fainter than the outer atmospheric value.  So to correct it, one has to apply the transformation so that it becomes brighter--ie subtracted from the instrumental data.  Am I missing something?

I was really hoping that this would make the 2 data sets come together ever more. 

In addition, we have to remember that the low air mass data (aprox 1.25) will have 0.08 mags of secondary extinction, which would also have to be subtracted from the low mass data.


WBY's picture
Sign of Second order extinction

assuming that X is the same for both stars as it is for all practical purposes in most amateur CCD fields of view (< 1 degree).

the vt0-vc0 = vt-vc -k"*((bt-vt) -(bc-vc))*x using spreadsheet arithmetic. k" in the example in appendix G of Henden and Kaitchuck is a negative number (-0.41). And it stands to reason that it should be because extinction affect bluer stars more than redder ones. Therefore

-k" is a positive number if k" is negative

(bt-vt) -(bc-vc) is a positive number if bt-vt = 1.6 and bc-vc = 0.0


vt0-vc0 +vcref is a more positive number (dimmer) than it would have been without the 2nd order extinction (vcref is the sequence magnitude of the comp used as the reference).

Therefore the effect of the second order extinction where the comp star is bluer is to make the target star dimmer.

Therefore, If I am not turned around, this moves the magnitude in the right direction.

Brad Walter, WBY

WGR's picture
Second Order Ext


I agree with;

 Therefore the effect of the second order extinction where the comp star is bluer is to make the target star dimmer.  We are saying the same thing--as long as you mean "is to make the target star dimmer in the instrumental mag plot."

So to transform the data, to get the out of atmospheric magnitude, we need to take the observed dimmer magnitude of the target star, and subtract the 2N order ext number from the data to correct it.  This makes the transformed data brighter than the instrumental data.

The only problem here, is that the high air mass data on CH Cyg is already brighter than the surrounding data.  So it seems that the second order extinction makes the problem worse, not better.


Am I turned around? 

Perhaps the only solution is to do a full blown transformation of all the data, both high and low air mass and see what we get.  Based on the incremental analysis, it does not seem to be the total answer.





HQA's picture
2nd order extinction, etc.

As Brad mentions, 2nd order extinction in general makes red stars brighter than they should be, so properly accounting for 2nd order extinction will move a red star to fainter magnitudes.  Since Josch is observing at 3 airmasses, this seems the most likely reason why his data don't line up with others after transformation.  One worry, however, is that Josch indicated he tried using both "blue" and "red" comp stars and got the same result.  I also recommend against observing at 3 airmasses unless there is no alternative - work on the stuff that is easy!

Regarding transformation, Josch's coefficient of -0.07 would normally seem large, except that he is using the KAF-16083 sensor.  This is a blue-enhanced sensor, but not microlensed, and so has a different spectral response curve.  For the sensors used at AAVSOnet, here are typical values of V vs. (B-V) using the dielectric filters:

KAF-1001E  -0.008

KAF-6303E  -0.016

KAF-0261E +0.014

KAF-0402ME -0.017

KAF-16083E -0.040

The 16803 tends to have a bluer response (more like the Sony HAD sensors), so the transformation coefficient is more negative (redder stars have to be made brighter to match their standard values).  Remember that the Astrodon filters only come with one bandpass shape; they are not tailored for your individual CCD, and so that is one reason why there are differences between telescope/camera systems and why we need to transform in the first place.

If you want to see what properly transformed CCD data is capable of, look at which shows the results of transforming supernova light curves from a half-dozen amateur-class telescopes.  I think we should strive to get this same quality of data from our observations. Even if it means fewer total submitted estimates, each one will be more accurate and better useful for scientific research.


Log in to post comments
AAVSO 49 Bay State Rd. Cambridge, MA 02138 617-354-0484