1 Efficient catalogue cross-matching – Paul Hancock (Curtin University)

Variations in the brightness of objects in the sky over time can be the result of a range of exciting astronomical phenomena – from exploding stars to black holes. Identification of these events from their light curves means that we need to first form these light curves by joining many catalogues.

This joining, or cross-matching, process presents a number of challenges – uncertainties in position, source confusion, missing data, and weird shaped sources. We currently have a few methods that work OK when the catalogues are very well behaved, but which are either very slow, or make many errors.

Creating an effective, efficient, and scalable solution would greatly assist in the ongoing research into these objects. An ideal solution would be something that could run within a database and be updated as new catalogues are added, but even a stand-alone script would be of great benefit.


2 De-noising of spectra – Helga Denes (ANU)

Decomposing complex spectra made up of multiple, potentially overlapping Gaussian components is a challenge for Galactic and extragalactic astrophysics, but is critical for understanding the properties of gas within galaxies.

Our team is currently using a code called GaussPy (Lindner et al. 2015) that autonomously decomposes spectra into Gaussian components. which improves upon previous methodologies (namely by-hand decomposition) by removing the need for human input (and subjectivity!), increasing the reproducibility, and allowing for the decomposition of large samples of data. GaussPy identifies different Gaussian components by finding inversions in the second derivative of the data, but since the data includes noise that mimics inversions, it uses a machine learning algorithm to determine the optimal parameters for first smoothing the data. For lower signal-to-noise data, the maximum confidence level of the decomposition produced by GaussPy is low (at best ~70%, Murray et al. 2017) and primarily due to the difficulty in separating the noise from the spectral signal using only smoothing.

We would like to improve the performance of the Gaussian decomposition algorithm by first devising a way to isolate the noise from the spectral features or ”de-noise” the spectra (e.g. using wavelet transforms) in order to improve the confidence level of the decomposition. Such an application will be extremely valuable to use on the large number of spectra produced by the upcoming large all sky surveys such as GASKAP and WALLABY. Additionally, an easy to use ”de-noising” software would be useful for other astronomical data.


3 (Space) Weather database and visualisation – Christopher Jordan (Curtin University)

One of the main goals of the Murchison Widefield Array (MWA) is the detection of the emission line of neutral hydrogen (also known as the 21cm line) during the Epoch of Reionisation (the dawn of the first galaxies in the universe). As this signal will be very weak,  a comprehensive characterisation of the sky, as well as the behaviour of the ionosphere is vital.

The MWA team has begun working on characterising the ionosphere, however, to be able to have accurate diagnostics as to why the ionosphere is or is not active the data will need to be compared and correlated with various weather data, including space weather (e.g. solar activity, meteor showers), earth weather (e.g. thunderstorm activity) and other possible actors (e.g. earthquakes, volcano eruptions).

This challenge is to design a tool that parses information on various types of weather to be displayed for a time and area. For example, if one was interested in what was happening around Perth on the 1st of September 2015, then this tool would tell me what had occurred: a meteor shower, solar activity, thunderstorms, earthquakes. As the MWA has a big field of view a bonus outcome for this challenge would be an app which could also display the area over which the weather acted.


4 Using Machine Learning to characterise galaxy components – Rebecca Lange (CIC)

There are many sub-components to a galaxy, e.g. bars, nuclei, thin disk, thick disk, spiral arms or shells to name a few. Each of these component is potentially a result of different formation mechanisms and will reflect distinct events in a galaxy’s evolutionary history.

However, even though some components are prominent in a galaxy’s light profile they may only contain a modest fraction of the total mass. Some components can therefore be thought of as primary (e.g. bulges and disks), and others as secondary perturbations of these primary structures (e.g.bars, pseudo-bulges, rings etc). Understanding how the fraction of mass in each component changes with galaxy type and over time is an integral part of understanding how galaxies assemble their mass and grow.

There are various fitting codes available to fit a galaxy’s light profile, however, it is inherently problematic to decompose a galaxy into its bulge and disk components and return reliable bulge-to-disk fractions (i.e. the fraction of mass found in each component). Another approach to this problem is applying machine learning techniques to the images of galaxies directly. This challenge will create and use a sample of simulated galaxy images to train machine learning models and test whether this results in a viable model for bulge-disk decomposition on actual galaxy images.


5 Looking the other way, a weather warning system for telescopes – Balthasar Indermuehle (CSIRO)

Looking the other way: New interferometric dish based telescope facilities executing unsupervised observations such as ASKAP and SKA face a multitude of new challenges in the way of equipment protection from severe weather.

Using near real-time high resolution satellite imagery as obtained from the Japanese Himawari 8 satellite, it is possible to detect convective developments early, and quickly, using the “overshooting tops” algorithm (will be provided, along with sample data). Develop a python code that identifies new cells as they develop, and track the convective cells’ centers, dimensions, determine their movement, and predict position, size, and severity for time steps of 10, 30, and 60 minutes into the future.


6 Code optimisation for the MWA tied-array beam pattern simulation – Bradley Meyers, Marcin Sokolowski, Paul Hancock (Curtin University)

The Murchison Widefield Array (MWA) is a low-frequency Square Kilometre Array (SKA) precursor radio telescope located in the Western Australian outback, approximately 800 km north of Perth. The MWA consists of 128 elements (called tiles) which each contain 16 dipoles in a 4 × 4 m grid. One of the MWA recording modes is the Voltage Capture System (VCS) which allows us to capture data in its most flexible form and then post-process it to get the desired science product.

One of our challenges is accurately calibrating the VCS data to get meaningful estimates of the brightness of objects, such as pulsars, allowing us to pursue more detailed scientific analysis. To calibrate the VCS data we currently use a simulation of the telescope’s “beam pattern” to calculate the relevant parameters for the instrument calibration. A beam pattern (or radiation pattern) is a map of the radio telescope’s sensitivity across the sky and is a function of where you are pointing on the sky, the wavelength at which you are observing, and the separation of each of your array elements. In our case, the separation of elements ranges from ∼1 m to >3 km, the observing wavelengths range from 1 – 4 metres, and the possible observing directions are effectively anywhere on the visible sky.

While there is Python code available to do this kind of simulation on a per element (tile) basis, there was previously no code that would simulate the radiation pattern when using all 128 elements combined (i.e. a tied-array beam). The individual tile simulation does not require high-resolution simulations and we can sample on a∼ 5 × 5 degrees scale and still get excellent results. However, when moving to the tied-array beam simulation, we need the simulation grid to be fine enough to resolve differences on scales less than 0.02 × 0.02 degrees. Furthermore, to accurately model the response, this grid needs to sample the entire southern sky, i.e. zenith angle θ = [0, π], azimuthal angle φ = [0, 2π]. The whole process is embarrassingly parallel, as we are effectively just computing some function F (θ, φ) over the whole sky.

For this hackathon challenge there a two possible pathways: working on optimising/speeding up either the Python or the C/C++/CUDA implementations of the calibration software.


Click here to get back to the event page.