“TraP”, a transient detection pipeline, gets its first public release

We’re happy to announce that the TraP, an image-plane transients-detection pipeline, has reached version 2.0 and been made publicly available for the first time, along with an accompanying paper (to be published in Astronomy and Computing). The TraP project was born out of the UvA LOFAR-transients group, with close collaboration from the 4 Pi Sky team over the past few years.

The TraP is what’s known as a source-cataloguing pipeline, which means that it processes images one-by-one, extracting source positions and flux intensities from each image, then attempts to match up these measurements into lightcurves spanning the entire dataset. If something new pops up, or if a source shows significant variability in its lightcurve, the TraP will flag this up for further investigation. This approach has been used before (e.g. for the CRTS project), but the TraP is novel for a few reasons. Key among these is fact that TraP can be run on a plain-old stack of images, without the need for a carefully preprocessed ‘deep’ image (an image of unusually high quality made by summing the best images from a dataset) to seed the catalogue beforehand. This makes it possible to run the TraP on a sequence of images as they are observed in near real-time. It can also collate data from across a range of frequency bands, which is vital in the current era of wide-band observatories and multi-observatory observing campaigns.

A Banana screenshot

A screenshot from Banana, the web-based graphical interface for browsing TraP results.

Additionally, the TraP project has embraced some ways of working that are still quite unusual in astronomy. For starters, the graphical interface used to give an overview of results is a Django-powered web-interface. This means that while all the software and results are stored and run on a heavy-duty central server, end-users can still browse through the results in a intuitive manner using the web front-end, Banana (don’t ask). The TraP has grown into a significant software-project by most standards (~20,000 lines of Python and SQL code, ~350 unit-tests), with a diverse group of contributors. To ensure things kept running smoothly we employed comprehensive unit-testing, continuous integration, issue-tracking and code-review, which should put the project in good stead for a more open development model.

The TraP has already been put to good use on data from the LOFAR RSM, ALARRM, and JVLA-CHILES surveys (see our projects page). Hopefully this open release will allow profitable use with many more datasets.