Ry Cutter
Rapid identification of transients that evolve on short timescales is important to understanding the physics and to testing current models surrounding these astrophysical phenomena. In the advent of the wide-field survey era, solutions to automatic real-time transient identification are more pressing than ever. These telescopes will see tens of thousands of sources with every exposure; given the size of the images and number of stars captured, finding transient quickly presents a real challenge. The Gravitational-wave Optical Transient Observer (GOTO) is a high cadence, wide field of view, survey telescope, and presents the perfect opportunity for the development of tools for the rapid identification of transients. In this poster, I will detail the challenges when searching for transients in wide-field data and describe the development of an image subtraction pipeline package, ZOGY in Parallel (ZiP), that can stack, align, and subtract data all in real-time. The benefits of this package over other methodologies for large fields of view are articulated. I show that to correctly align wide-field images, higher-order transforms are necessary. I also establish a new PSF modelling technique that employs Zernike Moments, highlighting that they can create spatially varying PSF kernels quickly. Furthermore, I demonstrate it is possible to optimally subtract large images on sub-minute timescales; offering a practical solution to identify transients and alert follow-up facilities within minutes.
We are now in the era of wide-field survey Astronomy. Finding transients quickly for broadband follow-up is more important than ever. The most effective way to find changes in the sky is with image subtraction. With an unprecedented number of stars per exposure, computing subtractions in real-time presents a new set of challenges.
The Gravitational-wave Optical Transient Observer, GOTO, is a wide-field survey and is the pathfinder for the next generation of high cadence wide-field surveys. Therefore, GOTO marks the opportunity for the development of tools for the rapid identification of transients in wide-fields.
[An image of the GOTO with its 8-scope configuration]
Image Subtraction has two main requirements that are made significantly harder with fast, wide-field, optics. The first is source alignment, which is the process of making sure stars line up on the same pixels in both images. The second is Point Spread Function (PSF) modelling.
A wide-field requires higher-order transforms to successfully align the whole image, and the PSF is not static across the entire image, necessitating multiple models. This poster will show the development of a python package designed to build subtraction pipelines for wide-field surveys.
[An example of image subtraction]
Slide 2. Title: Aligning Wide Fields
Wide-fields cannot be aligned simply with linear mapping. To fix this, a regular affine transform is used on the centre of the image, and a spline in employed to account for the higher-order transforms needed to correctly align the images. This new method is called Spali2.
[O one of GOTO’s original subtracted images is given, showing how wide-fields cannot align the whole image]
Two key metrics need to be assessed to test alignment performance. Flux conservation tests how much noise is introduced to the source flux through the transformation. Spali2 has shown to be able to transform images without introducing noise to source flux. The other metric is difference in source position. Of the methods highlighted, Spali2 is able to align the most sources successfully.
[Two plots are given, one showing that the flux uncertainty from the Spali2 transform is low; the other, highlighting that Spali2 has the most sources aligned within a pixel in both axes.]
Slide 3. Title: Modelling the PSF quickly with Zernike Moments.
Zernike Polynomials work on an orthogonal basis and form a complete set, meaning that they can recreate arbitrarily complex surfaces. Zernike Moments are the projection of an image onto a set of complex Zernike polynomials. Using a spatially varying function, multiple PSFs can be modelled from one image.
[Two figures: Zernike Moments recreating an image of a kitten and complex PSFs from a single image]
For GOTO, each kernel can be built in under 10 seconds.
[The last figure highlights the modelling speed for different kernel sizes for different polynomial orders.]
Slide 4. Title: ZOGY in Parallel
Using a parallelised algorithm combined with the ZOGY functions, an entire GOTO image (8000 x 6000 pixels) is subtracted in under 30 seconds. This is 8 times faster than the original ZOGY implementation and over twice as fast as ZOGY’s competitor, HOTPANTS.
[A figure showing ZiP finding transients down to the seeing limit, where HOTPANTS fails]
This python package is called ZOGY in Parallel (ZiP). The ZiP Package contains parallelised co-addition functions for stacking, Spali2, and Zernike Moment PSF modelling. This allows users to make an entire real-time subtraction pipeline all within the ZiP framework.
Combining this with machine learning, GOTO can flag transient candidates within a minute of an image being taken. Presenting a new paradigm for rapid transient follow-up.