Aperture size to optimise SNR

[brandx]

Does the aperture automatically size to optimise signal/noise ratio for an image or does it require testing different sizes (including annular) to get an optimised SNR?  Does this also need to be done for all images in a sequence?  As far as I can tell once an aperture size is selected it applies to all images which may not give an optimised SNR for each

[SCO]

I have found that experimenting with different apertures, centered around the default aperture has helped get better results.  Watch the BIC though.  You want the best SNR that does not increase BIC.

[brandx]

thanks. Not sure I understand what BIC is though

**UPDATE** just found a reference https://iopscience.iop.org/article/10.3847/1538-3881/153/2/77/pdf
So play with aperature size, run for all images, create light curve and check BIC <2?

But does that imply a set size of aperture for all images?

[SCO]

BIC is the Bayesian information criterion.  It prevents over-fitting. The process is described here: https://en.wikipedia.org/wiki/Bayesian_information_criterion .  In practice, I find that the BIC starts to increase when the aperture becomes completely inappropriate.  This happens even though the SNR might keep improving.

Ed

[SCO]

Based on my tests, a set size for all images usually gives the best results, but not always.  Sometimes a variable aperture using the T1 radial profile will work slightly better, but the difference is seldom significant.  It depends mainly on the particular atmospheric conditions at the time.

[karenacollins]

Finding the best aperture requires experimentation with different apertures types and sizes, and depends on the distance of other stars from the target star (and comparison stars if they are limited).

If the target star has a nearby relatively bright neighbor, you'd likely get a best result from a hand-selected fixed aperture radius that is half (or slightly less than half) the distance to the nearby star. Use the first AIJ option and the settings above it to define the aperture size.

For well-isolated target stars, I recommend using the "Auto Fixed Apertures from multi-image T1 radial profiles" option in the latest version of AIJ (with the default cutoff threshold of 0.01). This is the third option down. This mode will cycle through all images, calculate a radial profile from each image (and extract a FWHM), and then suggest a very good guess at an optimal aperture for most situations.

The 4th option down, "Auto Variable Apertures from each image T1 radial profile" may give a better result if there are large variations in seeing throughout the image sequence", and the target and comp stars are very well isolated from other stars. This option is the most sensitive to nearby stars and will produce larger variations in the target star lightcurve as the apertures grow and shrink in size as the FWHM in the each image changes.

It is on my to-do list to write a more detailed guide for how and when to make use of the different AIJ aperture options. Keep an eye on the forum for a post in the next couple of weeks.

Karen

[karenacollins]

Also, RMS should be used to select the best aperture, but you need to define the model to measure RMS against first (i.e. the transit model for exoplanet observing). BIC is used to decide if a transit model is preferred over a flat lightcurve model, and to determine if additional detrend parameters are justified statistically. The BIC should drop by at least 14 (2 x 7 transit model parameters) to justify a transit detection, and should drop by an addition 2 to justify a new detrend parameter. Even if BIC supports adding more than one detrend parameter, I generally recommend only one detrend parameter for ground-based light curves, unless you have at least as much baseline coverage as you have in-transit coverage, and even then I'd strongly lean toward airmass detrending over other detrending. If there is a nearby star, often FWHM or tot_C_cnts detrending may work better or in addition to airmass. The bottom line is that you want detrending to "clean up" a trend (curve or tilt) and/or wiggles in the data, but not completely build a transit in data when there is no transit obvious in the undetrended data, or otherwise completely reshape a light curve.

[brandx]

Many thanks Karen I will give that a go.