Recent Changes - Search:

Doastro

The image subtraction procedure uses several scripts and modules in Python and Fortran 95, connected via f2py and using OpenMP for the parallelization of the most important tasks. We use several python modules specially designed for astronomical usage such as astropy, as well as some R routines via rpy2. The main script is contained in the file doastro.py, but there are several other files which are relevant. The structure of the pipeline is shown below.


Image subtraction procedure structure

The workflow starts with the observations in the telescope, which once finished are automatically transferred to our computer cluster in about 30 seconds. The raw data is then preprocessed by a calibration pipeline (e.g. the DECam community pipeline) in order to remove instrumental effects from the data. We then remove cosmic rays from the data using the public CRBLASTER code, and extract sources and generate catalogues using the public SExtractor code.

With the source catalogues we then perform astrometric and flux calibrations between pairs of images to be subtracted (usually the latest image with respect to a reference image) and between our reference image and a public catalogue of stars in case the reference image is obtained from the incoming data (in our case we use as reference the second image of the first night).

When a new image arrives and the astrometric solution is found we query a public catalogue of minor planets for moving objects that could be present in the position of the new observation.

With the astrometric solution found we then project the new image to the coordinate system and grid of the reference image using a Lanczos windowed sinc kernel for the image resampling.

Then, after subtracting the sky emission from both images, either the reference or resampled image is convolved to have matching point spread functions (PSFs) using a novel algorithm which performs the convolution between the image and a non-parametric kernel with variable pixel size, dividing the image in different regions to take into account spatial PSF variations.

The PSF and grid matched images are then subtracted to look for transient candidates. Every region around pixels with a signal to noise ratio greater than 2.5 in the difference image is classified as either real or not using a random forest classifier previously trained with real DECam data.

When at least two difference images return a true candidate in the same location in the image and when at least one of the two images shows a positive flux difference we assume that we are dealing with a new non-moving optical transient. When a moving transient is present in the reference image it can cause simultaneous detections in several image difference pairs, but all of them should be negative.

When a candidate passes the previous spatial and temporal filter, all the data obtained at the position of the object is revisited in order to build a light curve of the candidate. We then obtain a Lomb-Scargle periodogram of the light curve to try to find short period variable stars that could mimic fast, young supernova explosions.

All the information about moving and non moving candidate is shown in a webpage which highlights their location in the individual CCDs and different fields. Only after visual of a candidate obtained in this fashion we will trigger different follow up observations.


Webpage showing all candidates in a given CCD: empty squares are candidates that appear just once (usually moving candidates in the new image), green circles show candidates that appear more than once and which only have negative flux differences (usually moving candidates in the reference image) and yellow stars show candidates that appear more than once and which show at least one positive flux difference (assumed to be a real new, non-moving transient candidate)