Re: How to keep FITS header when stacking?

Posted by karenacollins on Aug 13, 2017; 5:37am
URL: http://astroimagej.170.s1.nabble.com/How-to-keep-FITS-header-when-stacking-tp725p740.html

Hi Luca,

Your chi2r does seem quite high. Do you have the CCD noise settings specified properly in Aperture Settings for your CCD?

(these numbers are for one of my CCDs and shouldn't be interpreted as generally correct for all CCDs and gain settings)

Have you calculated by hand a representative scintillation noise to see if it is on the right order of magnitude for the difference in your expected uncertainty? Of course the shorter the exposure time, the more the scintillation component dominates. Following the Southworth paper you referenced, you will have significant gains in photometric precision if your field is not crowded and you can perform defocused photometry.

All that said, I'd be happy to analyze your dataset to understand why the chi2r is so high (and I might learn something along the way). We could start by you forwarding me your measurements table directly (in confidence) to kcollins atat cfa dotdot harvard dotdot edu, if you are interested. If that leaves open questions, we might need to investigate a way for you to transfer your calibrated images.

Regarding MCMC, I am referring to full system modeling tools such as EXOFAST that are typically used by professional astronomers to model the full system dataset (multiple light curves, multiple RV datasets, Doppler tomography, and measured stellar parameters). In those cases, AIJ is used to formally calculate the errors based on the CCD equations specified in the AIJ paper, and then the MCMC in the global fitting tools (not AIJ) should scale the errors as needed. AIJ itself does not implement MCMC at this time.

For now, you can always quickly load the AIJ measurements table into a spreadsheet program as a tab delimited file and scale the uncertainties as needed, if you are not using a separate tool such as EXOFAST to scale the errors.

I am not arguing that AIJ ideally calculates uncertainties, but it is adequate for most transiting exoplanet work.  However, I think it is a good idea to provide an option within AIJ to allow error scaling. I'll do my best to add that feature when I get time. I will also probably provide an option to allow AIJ to calculate errors to include contributions from scintillation. However, it will likely be in the December time frame when I have some time off from my daily job demands to get it finished. To do this properly, I'll need to make a separate panel for the user to enter/verify noise calculation data (aperture size, airmass keyword name in fits header, telescope altitude [which can default to the one shown in Coordinate Converter], and exposure time fits keyword name) {notes to self when implementing}.

To answer your question "I'm also curious about why you think that the scintillation noise would change based on the proximity to the target star?", the point of differential photometry is that the affects of the atmosphere are partially cancelled out, and the closer the target and comparison stars on the sky, the better the cancellation. I don't have data to prove it, but I expect that same principle applies to scintillation noise as well.

Regarding the Southworth reference (https://arxiv.org/pdf/0903.2139.pdf (eq. 13)), my quick-look interpretation is that they are only calculating the photometric error for the target star, because they claim that calculating the error for the target star is good enough and that the contribution from the comp stars does not need to be included:

(after equation 15)

This is indeed not always the case in my experience, and is highly dependent on the field and availability of good comp stars. It is probably as incorrect as AIJ not calculating scintillation noise in many cases.

Regarding the reference to "https://tinyurl.com/y8dge5gn": No, AIJ does not implement that  detrending algorithm. I believe it works best with very long datasets, rather than short single night observations. We use something similar in the KELT data reduction pipeline (i.e. ~5,000-10,000 exposures over many years). You could approximate this in AIJ by detrending using one or more comp star light curves (i.e. rel_flux_Cxx, where xx = 02, 03, ...) as detrend parameters (in addition to airmass, etc).

Karen