On a mountaintop high in the Chilean Andes, where the atmosphere is very dry and there is little light from civilization, a telescope aims at the night sky. Mounted on its lens is a custom-built 570-megapixel camera that scientists hope will help them better understand the mysteries of dark energy.
Dark energy makes up nearly 70% of the universe and is believed to be responsible for speeding its expansion. But scientists have only been aware of dark energy’s existence for about twenty years, and there is still much they do not understand.
To help them find out more, the Dark Energy Survey (DES) employed the Dark Energy Camera (DECam) to observe 5000 square degrees of the Southern sky and recorded images of hundreds of millions of galaxies in an effort to find patterns of cosmic structure that may reveal the nature of dark energy.
From raw data to science-ready
The DECam operated on behalf of DES for roughly one hundred nights each year from 2013 to 2019, producing close to 2 terabytes (TB) of data each night. Totaling 758 nights of observation over six years, that adds up to an enormous amount of data.
To transform this massive data dump into science-ready images and catalogs, the collected data is sent to the National Center for Supercomputing Applications (NCSA) where a team of scientists and technologists prepare it for use.
NCSA first cleans up the images, removing noise from the camera and atmospheric artifacts like bright trails left by satellites. They also stack the images, layering multiple observations to make the equivalent of a long exposure and create catalogs from all of the objects detected in those images.
“If you can only look for one minute, it’s not much,” says Matias Carrasco-Kind, senior scientist at NCSA and data release scientist for the Dark Energy Survey. “But if you look for one minute many, many times, you can build that up and look much deeper.”
The resulting images include many different astronomical objects, from galaxies to stars, asteroids, and even dwarf planets. In the latest collection of data, there were almost 700 million objects.
“At the beginning of the survey we only had a few basic tools to access the data,” says Carrasco-Kind. “But when more and more data arrived and our collaboration grew, we realized we needed to do something because we wanted to provide more streamlined and easier access to our datasets.”
While the DES data release infrastructure was created out of necessity, it soon became clear how valuable it was. “It’s been growing,” says Carrasco-Kind. “We’re actually hiring more people to help because it’s been so useful. We’re adding new services and leveraging our collaborations to increase its impact.”
“The main goal is dark energy and dark matter,” says Carrasco-Kind, “but there’s all this other science you can do with the data because there’s so much information there that needs to be digested.”
For example, galaxies evolve differently in different environments. An isolated galaxy will not develop in the same way as a galaxy in a crowded environment. Comparing whether there are, say, more spirals in the past or the present or the other way around can contribute a lot of understanding.
To support these inquiries the infrastructure will feature a signature search for galaxies, currently being developed. Using an AI image-matching technology, researchers can find galaxies of a similar type. They just drop a picture of the desired galaxy into the infrastructure, and it will return a ranking of the top ten candidates that look the most similar.
And that’s not the only way researchers can interact with galaxies. An active visualization allows users to explore the thousands of stars and galaxies that have been imaged by the DECam using a portal developed by collaborators from LIneA in Brazil.
“You can actually scroll around, move up and down, and zoom into specific objects. It’s very convenient,” says Carrasco-Kind. “Visualizing the sources from an image at the same time and being able to zoom around them is a powerful tool.”
Other upcoming visualizations will allow researchers to improve the classification and labeling of these unstudied galaxies—as spiral, elliptical, irregular, etc. Once a few hundred galaxies have been labeled manually, a machine-learning model can be trained on them and then applied to rest of the set.
Ultimately, the researchers are trying to understand what the universe is made of. An accurate census of existing galaxies will help them on their way.
Learning from other fields
As more DES data is processed and released, Carrasco-Kind and his team continue to improve and expand the infrastructure.
One step was presenting the DES Data Release Infrastructure at the Gateways 2019 conference last September. Hosted by the Science Gateways Community Institute, Gateways 2019 brings together scientists and technologists who build and host gateways containing all kinds of scientific information.
For Carrasco-Kind, the conference was an opportunity to interact with other scientists, get feedback about the infrastructure, and learn about other platforms.
“There are a lot of common needs across different fields,” says Carrasco-Kind. “Usually when you develop, you’re enclosed in your field. But this way we can learn from the other scientific communities. They may be facing similar problems and have some solutions that we haven’t thought of, which is encouraging.”
Romancing the stars
Carrasco-Kind is from Chile, and while he grew up in the capital city of Santiago, he recalls childhood trips into the countryside where the sky at night was very dark.
He was impressed by being able to see small satellites and the Milky Way. He got his first telescope as a teenager and a few years later decided to pursue astronomy as a career. While the romanticism of looking through a telescope has since been overtaken by computation, Carrasco-Kind has no regrets.
“It’s been very, very fun to build this infrastructure,” says Carrasco-Kind. “I know it’s been useful for many people, and it’s enabling a lot of different science cases, which is awesome.”