VASTER - ASKAP Intra-Observation Transient Search Pipeline

VASTER is a noval pipeline designed for the efficient detection and characterization of intra-observation transients in data from the Australian Square Kilometre Array Pathfinder (ASKAP) telescope. This pipeline streamlines the entire process, from data retrieval to candidate selection, providing valuable insights into transient astrophysical phenomena.

See details in the paper Wang Y. et al. (2023)

Some flowcharts can also be found in this Google slides

Example outputs: SB11676

Features

Workflow

1. Automated Data Retrieval: VASTER can automatically download ASKAP data from CASDA based on given SBID, simplifying the initial data acquisition process.

2. Model Creation: The pipeline generates a deep sky model from calibrated visibilities to serve as a reference for data analysis.

3. Model Subtraction: VASTER performs model-subtraction to isolate deviations from the deep sky model, potentially revealing transient sources.

4. Short Image Creation: Users can specify timescales (e.g., 10 seconds or 15 minutes), and VASTER creates short residual images for further analysis.

5. Statistical Map Analysis: The pipeline generates statistical maps (chi-square map, peak map, standard deviation map, and Gaussian map) based on the short residual images, aiding in the identification of transient objects.

6. Transient Selection: VAST can select transient candidates based on the generated statistical maps.

Outputs

For each transient candidate, VASTER generates the following outputs:

• candidate catalogues (csv)
• model image cutouts (png)
• animations created from model-subtracted snapshot images (gif)

Installation

git clone git@github.com:askap-vast/vast-fastdetection.git

Package Requirements

• python >=3.9
• numpy >=1.23.1
• scipy
• astropy >=5.3.3
• aplpy
• matplotlib <3.6
• scikit-image
• astroquery
• xmltodict https://github.com/conda-forge/xmltodict-feedstock
• python-casacore

Quick Usage

Running in slurm supercomputer system - only tested for Shanghai SKA regional prototype

bash run_everything.sh <sbid>

It will then automatically download data from CASDA and then submit a bunch of sbatch jobs

Rnning in bash

bash run_everything_for_bash.sh <sbid>

It will create a new folder named "SBxxxx" under current directory. Within the new folder, it will generate the following sub directories

• data/ (to store visibilities)
• models/ (to store time-averaged model images)
• images/ (to store model-subtracted short images)
• candidates/ (to store final candidates including csv and png)
• fitsfiles/ (intermeidate folder)
• scripts/ (to store scripts that will be used)
• logfiles/ (to store logfiles)

Step-by-step insturction (bash shell)

1. prepare data structure

python tools/get_everything_ready_for_bash.py <sbid number> <path to save results>

It will create data structure under defined path, with sub directories mentioned above.

2. download visibilities

bash <path>/scripts/download_selavy.sh
bash <path>/scripts/bash_CHECKDATA.sh 

All of the following steps are stored in /scripts/bash_PROCESSING_beam??.sh, i.e.

bash <path>/scripts/bash_PROCESSING_beam??.sh

Read the following if you want to do it manually.

3. untar visibilities

tar xvf <path>/data/scienceData.....tar

4. fix askap soft flux scale and beam positions

python tools/askapsoft_rescale.py <input visibilities> <output visibilities name>
python tools/fix_dir.py <visibilities name>

5. create sky model and subtract

casa --logfile <path>/logfiles/<name of logfile> -c imaging/model_making.py <input visibilities> <output name>

6. create model-subtracted short images

casa --logfile <path>/logfiles/<name of logfile> -c imaging/short_imaging.py <input visibilities> <output name>

7. candidates selection

python select_candidates.py <time-averaged model image> <deep_source_selavy_catalog> <folder_short_fits_images> <beam_number> <output folder> <output_name>

Step-by-step instruction (cluster)

This instruction assumes the pipeline runs on a computer cluster, so commands like sbatch <script name> are used. If it is run on a local machine, copy the line inside the script and run it with python.

Prepare the scripts for the pipeline

If running the pipeline on a cluster, adjust the runtime of MODELING, IMGFAST, and SELCAND as appropriate.

python /your/folder/to/pipeline/tools/get_everything_ready.py <SBID> <output_folder>

Download the visibility data Download the data one by one for 36 beams.

bash /your/output/folder/scripts/bash_GETDATA_beamXX.sh

Download selavy components and mosaiced fits images.

bash /your/output/folder/scripts/download_selavy.sh
bash /your/output/folder/scripts/download_mosaic_images.sh

Untar the data

cd /your/output/folder/data
tar xvf scienceData.XXX.YYY.ZZZ.beamXX_averaged_cal_leakage.ms.tar

Repeat this step for all 36 beams.

A folder of the same name without the tar extension can be seen in the data folder after the process completes.

Rescale and fix the data

sbatch /your/output/folder/scripts/slurm_FIXDATA_beamXX.sh

Deep modelling

sbatch /your/output/folder/scripts/slurm_MODELING_beamXX.sh

A .fits image in the form of SBID_beamXX.image.tt0.fits should appear in the models folder.

Short timescale imaging

sbatch /your/output/folder/scripts/slurm_IMGFAST_beamXX.sh

Short images will be saved in the images folder. You can check the progress in the .output file from the logfiles folder if running on a cluster.

Candidate selection

sbatch /your/output/folder/scripts/slurm_SELCAND_beamXX.sh

Chi-square and peak fits images of each beam will be saved in the candidates folder. Lightcurve, deep image and snapshot animation of the candidates (if any) are also saved.

Final candidate list

python /your/folder/to/pipeline/get_overall_table.py <SBID>

Change the base_folder to the pathname of the candidates folder. base_url is used when the lightcurve, deep image and snapshot are saved to an online server.

A SBID.csv file will be produced at the end.

Paramter customisation

Short timescle imaging setting

10-second images are generated using the task tclean from CASA. Paramters can be adjusted in /vast-fastdetection/imaging/short_imaging.py to accommodate for the scientific goal. Relevant parameters may include:

imsize controls the size of the image in unit of pixels and cell sets the angular size of one pixel. Increase cell when imaging a larger image to reduce runtime.

uvrange sets the uv-range of data to be imaged. Greater value represents more compact object.

gridder and wprojplanes determine the type of gridder used and the amount of w-values employed for W-projection. gridder = widefield and wprojplanes = -1 accounts for the w-term and generates spatially accurate image but requires longer runtime. gridder = standard and wprojplanes = 1 ignores w-projection and produces inaccurate image, especially when the source is away from the beam center, but the imaging is roughly 5 times faster.

Refer to CASA documentation for further details on tclean.

Candidate selection threshold

The sigma-level of the chi-square map and peak map can be adjusted in /vast-fastdetection/run_all.py. The limit of candidates plotted is also set in that python code.

Detailed instruction for candidates selection

The select_candidates.py scripts can automatically build a cube, generate final significance maps, select candidates and generate final products using optimised parameters.

If you are interested in modifying some parameters, please see below.

from vastcube.cube import Cube, Filter
from vastcube.plot import Candidates

Generate a (significance) cube

Load a bunch of short images

Save the location of a series of short images into imagelist. Note the images should be in a correct order (e.g., with time ascending).

Create a Cube class for the following processing.

imagelist = glob.glob('/folder/to/your/images/*.fits')

cube = Cube(imagelist)

Remove bad images

Remove images with rms larger than two times median RMS level.

cube.remove_bad_images()

generate a cube

The default way in run_cube.py is to form a cube directly

cube.icube()

If can convolve a spatial kernel with each image, to smooth the noise level and improve the detection.

cube.icube(ktype='gaussian', nx=19, ny=19)

The kernel type can be modified through ktype='gaussian' (a 2D gaussian kernel), ktype='psf' (dirty beam)

The kernel size can be modified through nx and ny (pixel values)

The kernel HWFM will be automatically calculated through fits header information (i.e., the synthesised beam size).

Note this smooth process will take much more time (~40min).

The generate (smoothed) cube is saved in cube.sigcube

Select transients candidates through a matched filter

Build a Filter using the generated cube

f = Filter(cube.sigcube)

Do a Gaussian (or other kernel) smooth on time axis

f.fmap("gaussian", width=4)

• Chisquare map - "chisquare"
• Gaussian map - "gaussian"
• Peak map - "peak"
• Standard Deviation map - "std"

Save the output map to fits file

f.to_fits(fitsname, imagename=imagelist[0])

Find local maximum (candidates)

c = Candidates(chisq_map, peak_map, std_map)

Or include Gaussian map using argument gaussian_map=<gaussian_map_name>

c.local_max(min_distance=30, sigma=5)

The detection threshold can be changed using sigma (in log space).

min_distance is the minimum distance (pixel) of neighouring local maximum

Remove potential artefacts

We need deep source catalogue information to removew potential artifacts - currently it supports aegean format table. But it can also change to selavy format table.

c.select_candidates(deepcatalogue)

Save final results to a csv/vot table

c.save_csvtable(tablename, savevot=True)