Project leads: Liza Sazonova (University of Waterloo), Michael Balogh (University of Waterloo)
Point of contact: Liza Sazonova (liza.sazonova@uwaterloo.ca)
Rubin project code: CAN-CAN-S5-1
Relevant working groups: Galaxy Morphology
Project status: Active
The structure of galaxies is a crucial piece of the puzzle for understanding their evolution. Galaxy structure can be described visually, parameterized using a model such as a Sersic profile, or quantified with non-parametric morphology measurements (concentration, asymmetry, Gini, M20, etc.). This project will develop software to measure quantitative morphology of galaxies observed with LSST quickly and robustly using machine learning.
The data volume of LSST will be too large to fit each galaxy using traditional methods. The modern solution to this is to use machine learning to classify galaxies, e.g. into mergers and non-mergers. However, classification models are trained for one specific task, and cannot be used for a related but different task without re-training. The advantage of quantitative morphological parameters is that they can be used for a wide range scientific tasks, including classification, without a need to re-measure the parameters.
There are two important issues a new quantitative morphology pipeline needs to address. First, non-parametric morphology measurements depend strongly on both image depth and seeing. While LSST will achieve depths of ~30 mag/arcsec2 by the end of the survey, data from different releases will have different depths, making morphology measurements inconsistent. LSST data will also have a much higher seeing compared to space-based data. Second, the large data volume does not allow to measure galaxy morphology directly using standard tools, which take up to ~10s of computational time per galaxy.
To address these issues, we plan to develop an ML model to predict quantitative morphological parameters from LSST images independently of image resolution and depth. Our work plan is as follows:
Test commonly used morphological parameters & ensure they are well-defined
Train an ML model to predict morphological measurements from imaging without direct computation
Train a second model to predict unbiased measurements, independently of resolution & noise, without direct computation
The first step of the project is to test quantitative morphology parameters and ensure they are robustly defined and perform well under different noise levels. We discovered that asymmetry (Conselice et al. 2000, 2003) is not well-behaved and does not recover true asymmetry of the object on average. If asymmetry is re-defined, we are able to measure it independently of image depth out to SNR/pixel ~ 1, without any additional modelling.
Upcoming presentation at the 14/03 Galaxies telecon.
The code developed during this project will be published and available on GitHub.
This work is very complimentary to a project by Dr. Vazquez-Mata to perform automatic morphological classification of galaxies.