ELSA: Euclid Legacy Science Advanced analysis tools

Intro

Euclid is an ESA space telescope launched in July 2023, designed to understand the nature of dark energy and dark matter. To achieve this, Euclid is observing over a third of the sky with high resolution imaging and spectroscopy, which will establish “the” reference map of the extra-galactic celestial sphere for decades to come. The giant archive produced will be a goldmine to study the history of the formation and growth of galaxies over the age of the Universe, driving answers to many fundamental science questions on the co-evolution of galaxies and supermassive black holes, the interaction between stars, gas, and galactic nuclei in galaxies at cosmic noon, and excelling in the discovery of rare objects including gravitational lenses.

However, the richest gold veins are also the most difficult to exploit: the tools developed for Euclid’s primary science will not be enough to open the rich legacy for the astronomical community. We therefore created ELSA to explore new methodologies and create cutting-edge pipelines, tools and algorithms. Our ambitious goal is to push the boundaries of spectroscopic analysis to the limits, uncovering hidden details of even the faintest and rarest galaxies measured by Euclid. We will leverage state of the art machine learning to efficiently handle the high- dimensional data and reveal the underlying physical processes they encode. This will need dedicated computing resources and highly motivated researchers versed in the most advanced techniques, that will work with our team of leading experts in the field of galaxy evolution to reveal the treasures preserved in the Euclid vault. Our machine learning will be supplemented by citizen science, enormously extending the reach of ELSA’s impact. ELSA will be a forge of knowledge and advanced tools that will not be confined within the boundaries of our teams, but shared with the whole scientific community and beyond to foster new projects and unforeseen discoveries.

The image above is from ESA's Euclid launch site (copyright ESA - European Space Agency, CC BY-SA 3.0 IGO), while the background image for this website is Euclid's early release observations of the Perseus cluster (copyright ESA/Euclid/Euclid Consortium/NASA, image processing by J.-C. Cuillandre (CEA Paris-Saclay), G. Anselmi, CC BY-SA 3.0 IGO).

Team

ELSA's work is distributed into five work packages, shown schematically above. An overarching management work package is led by the Coordinator, Dr. Margherita Talia. The work package on one-dimensional spectra is led by Dr. Lucia Pozzetti. Work package 3, on spatially resolved spectra, is led by Dr. Ben Granett. Our work package on machine learning is led by Dr. Sotiria Fotopoulou. Finally, our work package on dissemination, engagement and citizen science is led jointly by Prof. Stephen Serjeant and Dr. Hugh Dickinson. More information on these work packages can be found under their dedicated sections on this website.

List of team members

1D Spectra

The objective of the 1D Spectra work package will be to develop specific analysis tools, outside of the standard official Euclid pipeline, that will be able to handle thousands to millions of spectra, for an optimal and thorough exploitation of the Euclid spectroscopic galaxy surveys for both cosmological and galaxy evolution studies (i.e., studies of intermediate-high redshift scaling relations). We plan to apply the widely used approach of "spectral stacking" to boost the overall S/N ratio of a homogeneously chosen sample of galaxy spectra. The resulting stacked spectrum allows for the detection of spectral characteristics and continuum that are weaker than the instrument sensitivity due to the constructive interference between distinct spectra's signals and the incoherent addition of noise. Further, we plan to extract a wide range of physical parameters, such as star formation history, dust emission, and metallicity developing and using codes, besides the official Euclid pipeline, that combine photometry and spectroscopy. This allows for a more thorough understanding of the physical and chemical characteristics of observed objects.

The image above is from Gabarra et al. 2023.

2D Spectra

The objective of the advanced two-dimensional spectroscopy work package is to enable advanced analyses of the NISP 2D slitless spectroscopy dataset in the Euclid Deep fields by providing tools and reduced data products to the project and to the scientific community. The new tools developed by this work package will produce value-added data products that will be made available to the community.

The image above is an example of slitless spectroscopy with HST, from Brammer et al. 2012.

Machine Learning

The objectives of the machine learning work package focus on delivering machine-learning models that capture the information content (unsupervised methods & outlier analysis) of the Euclid dataset, and methods that will deliver physical parameters in a robust, and time-efficient manner (morphology, star formation rates). We will deliver a visualisation dashboard that gathers heterogeneous data stored in the Euclid and other archives for maximal exploitation of the data.

The image above is an illustration of the relative performance on a clump detection task of our Zoobot-based Faster RCNN models (3rd and 5th columns) relative to a model that has an identical architecture, but has a backbone trained on the well-known Imagenet library of terrestrial images (2nd and 4th columns). The first column shows the aggregated annotations for the same galaxies that were provided by volunteers via the Galaxy Zoo: Clump Scout citizen science project. Our Zoobot-based models recover more clump locations and more accurately match the annotations provided by citizen scientists. The first three columns show images from the SDSS survey. These are similar to the data used to train our model, although they were not part of the actual training set. The 4th and 5th columns show images from the Hyper-SuprimeCam (HSC) legacy survey. The Imagenet-backed Faster RCNN model performs poorly on HSC data. Although Zoobot has been trained as a classifier using HSC data, our Faster RCNN model was not. Using Zoobot as a backbone lends our model the flexibility it needs to perform well when applied to data from a different survey without retraining.

Citizen Science

The objectives of the work package on dissemination, engagement and citizen science are both one-way information dissemination to all our stakeholder communities (in common with typical Horizon project communication work packages) in order to execute our communication and engagement strategy, and a more innovative and research-driven two-way engagement, dialogue and debate in particular with the science-inclined public through the medium of crowdsourced data mining (i.e. citizen science).

The image above shows a direct image (left) and dispersed spectrum (right) of an example galaxy from the HST WISP survey, as presented to citizen science volunteers. The white arrow shows the bright light produced by an emission line. Image Credit: Vihang Mehta.

Internal

Within the Euclid consortium official Slack workspace, the project can be contacted in the #swg-legacy-elsa channel.

ELSA has an internal wiki, available to authorised users of the internal Sharepoint site. Consortium members can record their presentations to non-ELSA audiences in this Sharepoint spreadsheet.

Other internal Euclid resources can be found at the Euclid Quick Start Guide.

Meetings

ELSA kick-off meeting: Bologna, 12-13 February 2024. Please visit the meeting website for the schedule and logistical information.

ELSA annual meeting: Goldney House, Bristol, 13-17 January 2025 (image above). Schedule here.

Contact

Within the Euclid consortium official Slack workspace, the project can be contacted in the #swg-legacy-elsa channel.

Other internal Euclid resources can be found at the Euclid Quick Start Guide.

Public contact details for the work package leaders (if available) are available on their personal webpages, for which links are available under the Team tab.

Social media: Bluesky, Mastodon, LinkedIn, Github, Twitter

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';