Skip to content

lightcurver Documentation

logo

What is lightcurver?

lightcurver is a package / pipeline leveraging STARRED for precise photometry of blended sources in a large number of epochs. It specifically tackles the following type of problem:

  • The dataset has a large number of frames.
  • The frames are oversampled wide-field images (think modern sky surveys, such as the upcoming Vera Rubin LSST).
  • Only a small portion of each frame if of interest (think supernova embedded in a galaxy, or lensed quasars).

The above type of problem has historically been hard to solve: obtaining high quality light curves for such datasets has mostly been a manual process, taking up to a few months of an investigator's time per light curve. Doubt would also always remain: have we extracted as much signal as possible from the data, or could the signal-to-noise ratio of the curves be improved if we re-process everything just one last time? Are there systematics errors in the normalization that bias the shape of the extracted light curves?

lightcurver aims at making the process at least semi-automatic (set it up once for a few frames, then let the pipeline automatically handle any future frame), and builds up on experience of manually reducing such datasets to offer a no-compromise photometric precision. The end goal is being able to claim that the photometric uncertainty obtained in the light curves is dominated by the noise levels in the original frames, and not by normalisation or deblending scatter or systematics.

Example result

The two figures below show lightcurver outputs for a dataset captured by the ESO 2.2 meters telescope. (1)

  1. ESO program 0106.A-9005(A), PI Courbin, La Silla Chile

The first image shows a calibrated cutout of the object of interest, a lensed quasar, from one of the wide-field images of the dataset. The second image is the STARRED model, a high resolution image cumulating the signal of all the frames. The last image is a Hubble Space Telescope image for comparison.

example_deconv

The other product being the lightcurves of the point sources:

example_lightcurve

The method

lightcurver will prepare, for each frame, a precisely flux-calibrated cutout of the region of interest, together with a ready-to-use PSF model. The cutout pixels can then be modelled with STARRED, in a so-called "joint-forward-modelling". The process fits a high resolution model, with a constant pixelated component and a variable-flux point source component, to all available epochs at once, minimizing one global chi-squared. This allows us to both obtain precise light curves of the point sources modelled with the PSF, and the high resolution model that cumulates the signal of al the frames of the region of interest.