CCD failing to catch actual variability?

Recent CCD data on AG Dra are reasonably consistent, in several filters, with several widely dispersed observers. I hope CCD observers are not throwing away merely apparent outliers; arguing against that I do see many CCD rapid time series available (far too many, I would say, in fact)

What I do see are differences between your AG Dra visual observations and others' visual observations on some (May 28, June 6, 8, 9, 10, 11, 12) but not all (May 21, 22, June 13, 20, 21, 22, 25, 27, 29) nights. It still could be real variability within a night, so perhaps you might contact other visual observers to coordinate observations at the very same UTC time, at different locations (& thus different eyes, extinction, seeing etc). This is what CCD observers do to raise confidence in photometry of rapid transient phenomena, which you may have found. And if the differences are simply observational, then you can get to the bottom of that, using synchronized observations as the arbiter.

Hi Mike,

Symbiotics with rich emission line spectra can measure several tenths brighter in V than vis. All the bright well known examples (eg: AG Dra) tend to run this way. Certainly, so called CV measures would be expected to evidence this as well. Could this contribution result in flatter V and CV curves than what you see visually ?  Certainly, I would prefer B measures if I were looking for changes like you record in vis.

Steve        

Well, There certainly are very short term variations going on with AG Dra. I had noted that previously in another topic here. However, in the past 2 weeks, I have also noted a marked constancy in AG Dra, with very little variation in short term, or over the daily time periods. My visuals are almost perfectly in line with the CCD since about mid-June!

This was quite different back several months ago, when AG Dra would vary significantly every day, frequently down near the 105 comp in brightness, and I could see several tenths variations over a minute or less time span on occasion!

Also, the V1400 Cen has gone thru a fading episode over the last 2 weeks, which is not shown by the CCD. However, it has not exhibited the rapid, short term variations I have seen in AG Dra.

Mike

Recent CCD observations of AG Dra have confirmed my reported visual fadings over the past several months. One set of time series over the past few days clearly show variations in V of several tenths, over time spans of minutes (see attached LCG plot). This is exactly the behavior I had reported before visually, but up until now, no CCD observations had shown this effect.

Apparently, the effect does not show in B anywhere to the degree it does in V or visual.

Mike

Hello

Very interesting.  If one looks at this on the LCG with a time span of 200 days for instance, the V, B, and Vis observations with 20 day averages are consistant.  There is an inverted bowl shape to them on a scale of 200 days.  Vis with 20 day means does the same and is displaced by 0.2+ mags from V as it should per Stanton Transformation.  Blue also has the same inverted bowl shape, but 1.3 mags fainter than V.  

This looks remarkably consistant.  Good Eye Mike.  Looks like all is well, and by looking at short term LC, the noise is a larger component.  

Anyone understand what causes the inverted bowl behavior on a time scale of 200 days.  The inverted bowl changes when looking on a time scale of 400 days.  Very interesting.  Plot attached.

Gary

Yes Gary, nice analysis on the longterm changes and correlations. However, I wouldn't say there is a lot of short term "noise", in the sense of random errors in the observations. When I observe AG Dra visually, frequently notice the star vary a few tenths in brightness over time spans on the order of a minute or less. I really cannot report such changes, since as a single person observing, the time span is too short to both observe and report the time to the second, so my obs are actually an average of what I see over say a minute.

The attached LCG1 plot, I zeroed in on a short span of time around the latest CCD time series, and it clearly shows this short term variation of a few tenths. However it appears mainly in V, not B! (Both filters were done by the same observer LPAC).

Generally, I seem to notice these short term variations more when AG Dra fades below visual 10th magnitude on the average, not so much when it is in its brighter state above magnitude 10.0

I notice a very similar behavior with EX Hya. That star shows even larger, more rapid variations, almost like "flickering", sometimes over 0.5 magnitude in a minute or less. So, there appears to be some similar and unusual mechanism at work that causes these effects on at least these two stars!

Mike

Hello Mike

This particular section of CCD data has a cadence of approximately one minute.  So each of these points represents the average of what happens over that time period--not unlike your reported magnitudes.  To check for sub minute flickering, we would have to change the cadence.  At somewhere near 20 seconds, scintillation will dominate.  To do this well, one would need an EMCCD which is inteded to go really fast and have low noise.  Then need to take out the scintillation--perhaps with a second setup.  Interesting problem.  Perhaps easier to get a half dozen experienced visual observers to do a simultaneous campaign.

Gary

Gary

The paper by Galis, Hric and Smelcer (2015OEJV..169....4G) does a nice job of summarizing the variations in AG Dra.  Most likely the ~200d variation that Gary mentions is 1/2 of the 355d pulsation period of the red giant component of this star.

Much of the recent CCD data for AG Dra in the AID is coming from the SRO telescope of AAVSOnet, which takes B&V monitoring images.  Each nightly BV set takes 4 minutes to complete (2xV, 2xB, plus download times), and so wouldn't resolve any sub-minute variation.  It is entirely possible to have sub-minute variations in a ZAnd system since one of the stars of the binary is a compact object.  I've heard from some visual observers that some compact objects look like they are undergoing flickering/scintillation, while the nearby comparison stars are stable.

The high cadence regime is not well studied for very many stars.  One of the problems was highlighted by Gary Walker - scintillation.  This affects all observations, but is really noticeable for small aperture telescopes (like the BSMs) and/or short exposures.  If short exposures are not needed, you can reduce the scintillation by averaging multiple short exposures or using a neutral density filter to allow longer exposures.  If short exposures are needed to obtain fast cadence, then you either need to use a bigger telescope (which is why most high cadence observations have been done with ~4m telescopes), or find some way of synchronizing multiple telescopes spread over a large ground area.

My limited visual experience is that it takes several minutes to make an estimate, comparing the target vs. several comps to decide which pair to use and then to do the interpolation.  Therefore, if Mike is using a similar technique, his estimates take about the same length of time to produce as does the CCD system.  CCD measures should therefore show the same 0.5mag variation from night to night that Mike is finding, yet I don't immediately see that for the SRO dataset.  As others have mentioned, a more concerted effort would be required to confirm or deny such variation.

I don't give a lot of weight to the two recent time series on AG Dra from the same observer, as the mean magnitude differs from other observers and there is no confirmation from a second observer.  I'm more inclined to believe clouds or scintillation at this point.  That variation may be real, but because it differs from all of the rest of the submitted observations, I'd feel a lot better if it were confirmed.

Perhaps the recent campaign by Dr. Galis (AN 572) will give more insight into the short-time variation of this interesting system!

Arne

The observer has told me that the two nights of AG Dra time series were saturated, and will be removed.  Therefore, there is no current high cadence CCD data in the AID that would confirm or deny rapid variability for AG Dra.

Arne

Hi Mike,

Saturation is an interesting beast, especially with the anti-blooming gate CCDs, such as was used by the observer in question.  Let's say the true peak value of a star is 80K counts, which is then truncated by the range of the analog/digital converter to 65K.  The first thing you will see is, if the target is saturated but the comps are not, the star will be fainter than reality (65K vs. 80K).  That is obvious from the light curve (I think the observer has deleted their observations by now, but they were 0.6mag fainter than other V-magnitude measures).

The second thing is, as the seeing or transparency varies, the comp stars will change in brightness.  However, the target star will remain at its saturated brightness.  Therefore, the magnitude difference (V-C) will change instead of remaining the same.  So the second effect of saturation is a large scatter in the differential brightness of the target star.  That was also the case in the reported time series.  Note that this can even seem like a periodic variation, depending on the length of the time series and how the seeing or transparency changes.  For example, as you follow a star to higher airmass, the comp stars will get dimmer, while the saturated target stays bright, so you would see a slope in the resultant differential light curve.  Something like that can sometimes also be a clue, if the slope follows the shape of the extinction curve.

So just because you see large scatter in a reported time series, don't always believe this unless the star is known to have short-period variability, or a second observer confirms the variation.

Arne

Arne, The data was already deleted from AID, but I saved the LCG plot I made before. Knowing Pablo's location and time of the observations, I can figure out AG Dra was between 25 and 30 degrees altitude in the NW direction from him. So, there shouldn't be a serious change in airmass over the 30mins of his observations, though there could have been clouds or fog or local pollution (he is in the east Los Angeles basin). His comp star data is no longer available, so we cannot know if he used fainter ones or not.

I wish data wouldn't be so quickly deleted, until the source of any errors can be traced down.

OTOH, I hope he could simply do another run using exactly the same setup, just cut the exposure time or use a ND filter or something to get the counts into the optimal range!

Mike

Mike I usually setup my scope for several targets and let it run till sunrise, I made a mistake, my system is too much for these bright stars, upon closer inspection they were fully saturated..(my bad, I take full responsibility, should've known better) My system is just too much for these stars, if I expose for the target then the comps will be too faint as reported by VPHOT simple as that.

I mistankenly thought that if vphot accepted my reports then it would be ok meaning that they would be rejected if they were all outliers,  I know now that's NOT the case and I have summarily deleted the AG Dra observations. Even with the B filter AG Dra is saturated. Arne's description of the reason for the variability was as usual spot on , there was variability but it wasn't on the target, the variability came from the comps and check star as the airmass changed for various reasons. 

My system is a Celestron C14 Edge HD, with an STF8300 and an FW8 filter which includes a V,B and CBB filter.

To fix that problem I have now piggybacking a Williams Optics Z61 doublet (61mm) with a modified Canon 7D Dslr, and I will be taking the upcoming DSLR course so that I can do photometry on bright stars.

It is what it is, a heck of a learning process for me and mistakes will be made but when called on it I will take responsibility, learn from it and move on.

Lesson learned, I now I move forward.

Thank you Arne for you boundless knowledge and your  kindness. You Sir are a class act.

Best!

PS I'm checking all my data and should I find more I will immediately delete the observations from the database and fix it..

Thanks for your great efforts Pablo! Yes, been quite toasty in CA, though little change from nominal at our other place up in the high desert ;) I plotted your recent, properly exposed, data in LCG1, attached here. I used a very short time scale on the x-axis (~1 minute for the individual tick marks) and it shows your V observation cadence is more like 8 secs?

But, the important point is the short term variations you found, of nearly 0.2 magnitudes, are very much like what I observe visually on this star! Sometimes I see variations up to 0.3 mag, and other times the variations seem to disappear, and the star seems pretty constant over these short intervals of a minute or less.

However, this still fails to explain why many other CCD observations do not catch the larger variations up to 0.5 mag or more, which occur over several days. These are quite obvious by visually observing, and shown in the LCG, but apparently not "tracked" by most CCD?

Hopefully, you can keep observing this star, this same way, on a regular basis, Pablo!

Mike

Mike

The V Cadence was 2 seconds and some 3 seconds and 1 second...It may have taken the system 8 seconds to download etc but the EXPOSURE was 1 to 3 seconds with most of them at 2 seconds.

Glad you found them useful

Best!

Pablo

PS I'll try to give you plenty of a heads up on Facebook when I observe it again..

Since I can only  control the exposure in order to achieve the interval between observations /the sampling interval or sampling frequency, that's what I refer to as cadence. The 8 seconds makes sense based on my system. I don't have USB3 or Firewire for the camera download thus I can't shorten the 'cadence" anymore.

Cheers!

The two mini time-series by Pablo and Frank are nice datasets for error analysis.  I'll start with Frank's V-band set.

Frank had 49 20-second exposures, and used the traditional comp (000-BCY-343, 117) and check (000-BCY-345, 120) method.  For that series, the mean target magnitude was 9.813, with a standard deviation of 0.007 magnitudes.  This is worse than the S/N=740 value reported by AIP4WIN, but most S/N reports are theoretical estimates and basically just give you guidance.  Frank is using a C14, and with 20-second exposure, scintillation is pretty minimal.  His (C-K) values were -0.260 +/- 0.009, again about right for this fainter set of stars.  The extreme values (that set the peak-to-peak range for the dataset) were 0.015mag fainter and 0.018mag brighter than the average, or about 2 sigma.  This is perfectly reasonable for a 49-element dataset.  This shows that Frank is getting sub0.01mag precision for 10th magnitude at 20seconds, not bad.

Pablo took 29 2-second V-band exposures, if I read his post correctly.  With his C14, this should have about 0.01mag scintillation.  However, his mean target value was 9.884 +/- 0.040 magnitudes.  I don't know what his signal/noise was for each measure.  The extreme values are 0.083mag too faint and 0.081mag too bright, about 2 sigma, pretty normal.  His check star (120) was measured at 11.972 average and 0.059mag standard deviation.  So the target star deviation was somewhat less than the check star deviation, probably about right for the two magnitude brightness difference.  But both target and check have large standard deviations, which is indicative of low signal/noise.  Pablo, you might indicate how bright both the target and check stars were; roughly what signal/noise was reported for both; and what stars you used for your ensemble.

I don't see any evidence of variability in the target star over this very short interval of time, either for Frank or Pablo.  All variation can be easily explained as random noise, and none of the variability shows any trends: bright points may be immediately adjacent to faint ones.

Again, to demonstrate the variability that Mike Linnolt is reporting, you would have to observe on the same night and the same time as Mike, and for a longer time series.  Preferably you would have two CCD observers observing simultaneously so that you could correlate any changes.

Note that I am not dismissing short-term variability.  I am just saying that the current datasets are not sufficient proof, and to demonstrate such variability is a much more difficult task than you might naively think!

Arne

I attached the info you asked for Arne, I believe it was a full moon, hot and thin clouds rolling in and out , extreme light pollution (Street light illuminating my Observatory, I sometime use a Film set black silk on a C stand to give some shade, but it's not always possible). I always add CALLSTAT=BDF manually because I use Astroimage J to calibrate and I just don't know how to add the calibration info on the header, so far nobody else knows how to help me accomplish that..(neither here nor there I know, I just threw this here in case somebody knows how to do that!)

Pablo

From Pablo's reply, the signal/noise ratio for the target in the 2second V-band exposures was 203 (good!), but the comparison stars, 2 magnitudes fainter, only had SNRs in the 20-70 range.  That was most likely the major part of the standard deviation during the short time series, plus some effect from the light clouds.  Short exposures see clouds much more than longer exposures, which tend to smooth out the transparency variations.

Frank Schorr analyzed the B-band time series from both of you.  His 30sec B exposures had standard deviation of 0.040mag, while your 30sec B exposures had standard deviation of 0.053mag after throwing out one outlier.  The difference is likely due to sky transparency, so it looks like you are doing things mostly right, Pablo!  Both of the B time series are consistent with no variation, just random scatter.

It is always easier to test your system and learn how to improve the quality of your photometry by looking at known constant stars, rather than try to compare two time series on a variable.  While Kepler finds nearly everything is variable, its precision limit is orders of magnitudes below anything possible from the ground.  Therefore, the best constant star targets are most likely Landolt standards.  There are a couple of good fields up in the evening, such as one centered on SA110-503, which gives you 8 stars to play with.  The AAVSO calibration fields like M67 are also good candidates for precision testing.

Bright lights impact photometry less than you might think.  As long as your star is well above the (bright) sky background, good photometry is easy.  Bright lights are much easier to deal with than clouds!

Arne

I am not convinced that the variations seen in Pablo's V observations are due to errors. The questions that bother me about this explanation are:

1. If clouds are the culprit, you would expect just as much variation in the B exposures as the V.

2. Its very unlikely from statistical variation alone, that several observations out of his small set would show 2 sigma deviations (only 5% should) about the mean!

3. If the variations in V are due to SNR "noise" of them or the comp stars, then you would expect even more variation in the fainter B, but that is not the case. The brighter measurements show more variations.

So, given the situation at the moment, I am unwilling to accept LPAC's V observations are erroneous.

Mike

Frank

My 30 second SNR for the B filter was about 360,by the way I will be transforming all my data as M67 becomes visible in my sky. Last years coefficients are no longer valid since I changed my setup.

Take care.

Pablo

Unfortunately, for purposes of investigating this issue, my more recent visual observations have shown it has become less variable, both on the very short term as well as over periods of several days, compared to back in June/July. Though my coverage of it has been less intense recently too, due to several factors, including weather and location in the sky for my latitude...