2139 RIQ FCa : ADI algorithm with RAVC or CVC coronagraph

[astronomyk]

See MET-2237, MET-2139

[gotten]

RIX:

This study is very clear: is there any plan to do a similar study for the APP ?

"and in a second step optimally-scaled time-localized medians are subtracted in annular rings.": I do not understand this part... what is it used for ?

"Subsequently the derotated and PSF subtracted images are averaged to give a final stacked version": have you considered stacking in median instead of mean ? Would there be any gain ?

About the construction of the SNR: are there any consideration to use a STIM map instead of the Mawet-2014 t-student map ? Note that if high-pass filtering is applied prior to the ADI subtraction, the t-student should get closer to the Gaussian.

Figure 78 caption, "Single frame of image stack with mean PSF subtracted"": i believe it is the median instead of mean

RIX answers

This study is very clear: is there any plan to do a similar study for the APP ?

Yes, we are working on a Python (median subtraction) ADI code (based on VIP_HCI) to process simulated METIS APP data that will help us with implementing the corresponding APP pipeline recipe in C.

"and in a second step optimally-scaled time-localized medians are subtracted in annular rings.": I do not understand this part... what is it used for ?

In the original ADI definition by Marois (2006) they did a further annular optimization to improve noise (I was being overcomplete in mentioning this in the DRLD). In the medsub routine of VIP_HCI (https://vip.readthedocs.io/en/latest/vip_hci.psfsub.html#module-vip_hci.psfsub.medsub) this is the difference between the default mode='fullfr' and mode='annular'. Only the simple fullframe median will be implemented for METIS.

"Subsequently the derotated and PSF subtracted images are averaged to give a final stacked version": have you considered stacking in median instead of mean ? Would there be any gain

In this context I mean median-combined however it would be easy to provide this as a setting [default median / mean]. I didn't consider any potential gain here. (perhaps in some digitization noise regime?)

About the construction of the SNR: are there any consideration to use a STIM map instead of the Mawet-2014 t-student map ? Note that if high-pass filtering is applied prior to the ADI subtraction, the t-student should get closer to the Gaussian.

The METIS ADI pipeline is mostly there for quicklook reduction at the telescope and to provide the datacubes and metadata needed for HCI experts to use their own optimized code. (imagine awesome futuristic routines from the year 2029 and beyond here). The METIS ADI code should be straightforward to implement and easy to maintain so functionality should be kept to minimum and more specialized approaches are left to the users. We follow the default definition of VIP_HCI (https://vip.readthedocs.io/en/latest/vip_hci.metrics.html#module-vip_hci.metrics.contrcurve). If necessary we can discuss which specific noise estimation method to implement for contrast curve / detection / snr maps with the HCI lead Olivier Absil during PIP FDR. An optional FFT-based high pass filtering (as suggested) might be useful to improve detections and easy to implement.

Figure 78 caption, "Single frame of image stack with mean PSF subtracted"": i believe it is the median instead of mean

Agreed, I need to change this to median.

Action Item 2237

Update the document to state that the optical algorithm will be examined again at a later date. Options to both stack images using median and mean will be available in the pipeline.

Clarification: this is about the "optimal" algorithm to determine the contrast and SNRs. We will baseline Mawet et al 2014 following VIP_HCI.

[gotten]

Todo list:

Simulations + processing + writeup

• complete APP + ADI python prototype code based on simulated data from Scopesim + HEEPS (more difficult)
  Text correction in section 10:
• add description of prototype code to DRLD Section 10 and put code on Prototype github.

Text corrections in Section 6

• emphasize that our baseline is Mawet et al 2014 and more advanced methods are left to the user
• change mentions of mean to median (as default). Allow mean,median,sigmaclip for final collapse of datacube. Use just median to estimate average PSF.
• add possibility to use percentiles with the contrast (and SNR ?) curves

[gotten]

In the process of making APP prototype code and fixing HEEPS + ScopeSim.

[gotten]

HEEPS+Scopesim with APP and APP prototype code works. Need to add in 2D maps and put it on github. Also add writeup to DRLD.