Photometry with Astropy photutils : Reviewers please

I assume that you are doing DSLR photometry, and that you do preprocessing to de-Beyer, apply bias, dark and flat field corrections off line, using some other software. If these assumptions are not correct, some of the following comments will be rendered inappropriate.

My first concern is about the comparison star selection process. It appears that by going directly to Simbad you are bypassing the AAVSO's standard comparison star sets. (No big deal.) On the other hand, the selection process does not appear to consider spectral match between the target and comparison stars and check star, and (of lesser concern) the lack of consideration of estimated errors in the photometry in choosing them.

Another concern is the fixed choice of a 6 pixel radius for the measurement aperture, and the apparent lack of a background annulus. My own practice is to look at the profile of the target signature (and, at least initially, of the other stars) to select an aperture that maximizes SNR while avoiding including any extraneous background stars, and ensuring that none of the stars are in saturation. My experience is that a single aperture simply does not work over the range of seeing and defocus conditions encountered in the real world. So, I suggest you implement a visualization routine to allow display of the ADU profiles and SNRs to facilitate proper choice of measurement aperture. For that you will need to implement the background annulus, also variable, to provide a noise estimate.

I notice that you do not correct for, or even calculate, air mass. IMO, that is a serious deficiency.

There is some question in my mind as to whether the residual on the linear regression is the correct measure of error. Perhaps someone else would like to clear this up? (In my own practice, the magnitude is estimated based on a zero residual (i.e., the target is placed on the curve) and the error is given as the standard deviation of the multi-image measurement process.) Speaking of which, it appears that your script is for a single image. Of course, it would not take much to loop over an ensemble of images and use the result to estimate magnitude.

Finally, the report needs to comport to AAVSO requirements. You need to indicate air mass and filter. Results should be reported to three decimal places only to avoid the mis-impression of unrealistic precision.

Hi Derek,

I quite enjoyed what you have done :-) Very nice! I'll add also couple of ideas/suggestions that could be useful for you:

• just for sake of simplicity and reliability, I'd use circular annulus for sky estimation, mode or median value (if you can choose) would be good
• to ensure the best sky measurements, creating frame mask for stars and using that mask for sky annulus measurements should give you the best estimate of local sky - photutils has a function for that
• determine/estimate FWHM automatically (fitting?) and define your aperture radiae based on that (e.g. stellar aperture 1.5xFWHM, sky annulus from 3-5 FWHM, your may like different numbers) - that helps somewhat with varying seeing/trailing
• always estimate uncertainties as well as you can (i.e. include your ensemble members, target(s) etc ), IMHO there was a function for that. It is often said in exact sciences that without uncertainty estimate there is no measurement...
• you may like to convert all magnitudes to unit exposure time (e.g. 1 second):   mag+2.5*math.log10(exposure)   - then it is easier to compare e.g. results when you have taken frames using multiple exposure times, also to create airmass-instrumental magnitude graphs for multiple nights etc etc
• if your target and ensemble stars do not have dramatically different colours, you could just combine ensemble stars into "one" comparison star, most probably a correct way would be adding up aperture sums (in your tables) and deriving magnitude from that.

I also wanted to ask if I may use that notebook for a student's astronomy course to explain basic photometry?

Best wishes,
Tõnis

Hi Derek,

Good work!

For the past couple of years, I've been using a Python Jupyter notebook that I wrote to perform aperture photometry on CCD time series. Like you, I also use the WCS for each image to determine the positions needed for the photometric apertures. I have a few initial comments, in no particular order.

1. I think that the magnitude error is underestimated. The linear regression gives equal weight to the stars, regardless of their SNR. The uncertainty will increase as the SNR decreases, which becomes a problem if your ensemble and target have a wide range of magnitudes. Also, this method of error estimation assumes that the spectral characteristics of the comp stars are similar to those of the target star. For example, if your comp stars are in the range 0.4 < B-V < 0.8 and the target star has B-V = 1.0, the relative redness of the target would probably result in a larger error than the comp-star residuals would indicate.
2. I use PythonPhot to do the actual photometry. I never really tried the aperture photometry routines within photutils. One of the nice things about PythonPhot's aperture photometry routine is that it allows you to specify a background-measurement annulus so that your star flux is background-subtracted.It also gives you reliable uncertainties on your star magnitudes.
3. If you were to modify your code to analyze a time series, I've found that it is computationally inefficient to call wcs = WCS(fits_header). Instead, I wrote a custom function that creates an empty WCS object and then populates it with the relevant WCS keys from the fits header. This is shown in the Astropy WCS documentation. If I remember correctly, this sped up my for loop by a factor of 5.
4. If you already know the equatorial coordinates of your target and comp stars, you don't necessarily need to do the source extraction unless you're worried about proper motion, inaccurate catalog coordinates, or a poor WCS solution. In practice, I rarely encounter these issues. As a safety check, I simply plot the photometric apertures over a sample CCD image to verify that they are positioned correctly.
5. I think that a better way of matching sources to target/comp stars would be to eliminate the nested for loop, find the Pythagorean distance to the nearest source, and to use np.argmin() to return the index of the matched source from your sources table. This should be faster than the current method, which I believe could also produce a false match if there are multiple sources within 4 pixels.

Colin

Derek, I am trying to learn photometry with Photutils and have your notebook, but trying to run it, I could not locate 

FITS_FILE = '/home/dokeeffe/Pictures/CalibratedLight/2017-10-02/KIC08462852-2017-10-02-20-56-22Light_005.fits'

Did you remove it?  could you put it back or send it to me?

Thanks, Richard Post

rspost@comcast.net

Hi Derek,

I've been using your excellent Jupyter notebook as a tutorial on photometry using astropy and photutils. Its very clearly organized and it has been a great learning tool!

In order to get it to work locally, I've been using DSS fits images.  I was wonder if I could have a copy of the image you used in order to duplicate the results exactly:

'/home/dokeeffe/Pictures/CalibratedLight/2017-10-02/KIC08462852-2017-10-02-20-56-22Light_005.fits'

Thanks,

Leo
____________________________

leo.fernig@gmail.com

(Vancouver, BC)

Your notebook replaces - dec with positive dec....

Hello Derek,

Thanks for sharing your Jupyter notebook on ensemble photometry using astropy and photutils. I'll definitely take a look and leave any feedback or suggestions I have.

I would also like to note that I recently started delving into the world of cryptocurrencies. It was quite exciting! I initially struggled to understand the difference between tokens and coins, but found a really helpful article called "tokens vs coins" that cleared things up for me. If you are interested in learning more about this, you can read the article, you should Go to Source.

We look forward to exploring your notebook and discussing it further!

Best regards, Harvey

Hello Derek,

Thanks for sharing your work!

I'm working on a similar project using AstroPy.  Let me share some thoughts on the subject:

• As it was already mentioned, it worth using annulus aperture tho estimate the background, which is just ignored now.  Photutils provides the necessary classes.  As a rough replacement, the image median could be used (better sigma clipped variant from astropy.stats).  It would allow for SNR estimation.
• Having SNR implemented, it is possible to choose the optimal aperture radius. For that, photometry for a range of radii is tabulated and SNR for each star is plotted against the radius.  Typically the SNR plot has  more or less pronounced maximum.
• SNR would allow estimating the uncertainty of the target instrumental magnitude (roughly 1/SNR)
• To estimate the uncertainty of the result, the following procedure is needed:
  • using squared residuals, calculate the uncertainty of the slope and the uncertainty of the intercept for the regression (see, for example, discussion here: https://sites.chem.utoronto.ca/chemistry/coursenotes/analsci/stats/ErrRegr.html)
  • propagate uncertainties for the slope, the intercept, and the instrumental magnitude trough the regression formula.

--

Best Regards,

Dmitry