Choong Ling Liew-Cain
Gather.town id
BD01
Poster Title
Constraining stellar population parameters from narrow band photometric surveys using convolutional neural networks
Institution
Mullard Space Science Laboratory, UCL
Abstract (short summary)
Upcoming large-area narrow band photometric surveys, such as Javalambre Physics of the Accelerating Universe Astrophysical Survey (J-PAS), will enable us to observe a large number of galaxies simultaneously and efficiently. However, it will be challenging to analyse the spatially resolved stellar populations of galaxies from such big data to investigate galaxy formation and evolutionary history. We have applied a convolutional neural network (CNN) technique, which is known to be computationally inexpensive once it is trained, to retrieve the metallicity and age from J-PAS-like narrow-band images. The CNN was trained using synthetic photometry from the integral field unit spectra of the Calar Alto Legacy Integral Field Area survey and the age and metallicity obtained in a full spectral fitting on the same spectra. We demonstrate that our CNN model can consistently recover age and metallicity from each J-PAS-like spectral energy distribution. The radial gradients of the age and metallicity for galaxies are also recovered accurately, irrespective of their morphology. However, it is demonstrated that the diversity of the data set used to train the neural networks has a dramatic effect on the recovery of galactic stellar population parameters. Hence, future applications of CNNs to constrain stellar populations will rely on the availability of quality spectroscopic data from samples covering a wide range of population parameters.
Plain text (extended) Summary
Measuring the age and metallicity (Z) of stellar populations within galaxies helps us to understand their evolutionary history. Upcoming large-area narrow band photometric surveys, such as J-PAS, will enable us to observe a large number of galaxies simultaneously and efficiently. However, it will be challenging to analyse the spatially-resolved stellar populations of galaxies from over 1 trillion data points. The resolution of J-PAS narrow-band data is 50x worse than CALIFA high resolution spectral data, and the lack of spectral lines in NBD make it much harder to determine stellar population properties. With J-PAS and J-PLUS expected to gather 1.5TB of data per night, we need to have a quick & efficient method of data analysis.
We use a convolutional neural network (CNN) to analyse our data as the application of trained CNNs to new data is very quick. We trained the CNN with the synthetic J-PAS-like narrow-band data created from CALIFA spectra (input features) and the accurate age or Z derived from CALIFA spectral data (output label), and tested the CNN with the unseen data.
Our dataset is formed of 19,727 CALIFA spectra from 190 galaxies. We use two methods for splitting the dataset into training and testing. For Set A, we take 25% of the data from each galaxy to test and train with the rest. This is the ideal case, where the histories of the galaxies in the training and testing sets are similar. For Set B, the training and testing sets come from different galaxies. This is the realistic case, where we know nothing about a galaxy before we observe it.
We made plots of the recovery of age from the CNN for Set A and Set B with the spectroscopically determined ages are plotted against the CNN age predictions. These plots showed that recovery for Set A was significantly better than for set B, with standard deviations of 0.03 dex and 0.14 dex, respectively. Results for Z were similar to those for age.
In conclusion, this proof of concept study shows for the first time that convolutional neural networks are able to reproduce ages and metallicities from narrow-band photometric data. The increased accuracy in predictions from Set A implies that galactic evolutionary histories can be very different for galaxies of the same age. For better convolutional neural network predictions, we need more high resolution spectral data from a variety of stellar populations for our training dataset
We use a convolutional neural network (CNN) to analyse our data as the application of trained CNNs to new data is very quick. We trained the CNN with the synthetic J-PAS-like narrow-band data created from CALIFA spectra (input features) and the accurate age or Z derived from CALIFA spectral data (output label), and tested the CNN with the unseen data.
Our dataset is formed of 19,727 CALIFA spectra from 190 galaxies. We use two methods for splitting the dataset into training and testing. For Set A, we take 25% of the data from each galaxy to test and train with the rest. This is the ideal case, where the histories of the galaxies in the training and testing sets are similar. For Set B, the training and testing sets come from different galaxies. This is the realistic case, where we know nothing about a galaxy before we observe it.
We made plots of the recovery of age from the CNN for Set A and Set B with the spectroscopically determined ages are plotted against the CNN age predictions. These plots showed that recovery for Set A was significantly better than for set B, with standard deviations of 0.03 dex and 0.14 dex, respectively. Results for Z were similar to those for age.
In conclusion, this proof of concept study shows for the first time that convolutional neural networks are able to reproduce ages and metallicities from narrow-band photometric data. The increased accuracy in predictions from Set A implies that galactic evolutionary histories can be very different for galaxies of the same age. For better convolutional neural network predictions, we need more high resolution spectral data from a variety of stellar populations for our training dataset
URL
Email: choongling.liew-cain.18@ucl.ac.uk; Twitter: @ChoongLingAstro
Poster file