Does the life of an astronomer or planetary researcher seem exciting?
Sit in an observatory sipping warm cocoa and having high tech tools at your fingertips as you diligently work and surf the wavefront of human knowledge surrounded by fine, bright people. Then – one day – Eureka! – All of your hard work and the work of your co-workers pays off, and you provide a critical piece of knowledge to humanity. A piece of knowledge that regulates a scientific debate or connects a burgeoning theory and brings everything together. Conferences … tenure … Nobel Prize?
Perhaps you envision something like this in your first year of college. But science is work. And as we all know, not every minute of working life is very exciting and enjoyable.
Sometimes it can be boring and repetitive.
It is probably not everyone's dream to start their science education by sitting in front of a computer that scrolls over photos of the surface of Mars and counts the craters. But someone has to do it. How else would we all know how many craters there are?
Mars is the subject of intense scientific research. Telescopes, rovers, and orbiters work to unlock the planet's secrets. There are a thousand questions about Mars, and part of understanding the complex planet is understanding the frequency of meteorite impacts on its surface.
NASA's Mars Reconnaissance Orbiter (MRO) has been orbiting Mars for 14.5 years. Along with the rest of its payload, the MRO carries cameras. One of them is called a CTX camera (context). As the name suggests, it provides context for the other cameras and instruments.
The powerful camera from MRO is called HiRISE (High-Resolution Imaging Science Experiment). While the CTX camera takes pictures with a larger view, HiRISE zooms in to take precise pictures of details on the surface. The couple make a strong team and HiRISE has given us more beautiful and fascinating images of Mars than any other instrument.
Craters play an outstanding role in the HiRISE archive "Picture of the Day". This is the POD for September 5, 2020. It was recorded on January 19, 2012. It shows a small crater on Mars with dunes and pink spots of chloride salts. Photo credit: NASA / JPL / UArizona
But the cameras are kind of "stupid" in a scientific sense. It takes a human to go through the pictures. According to a NASA press release, it can take 40 minutes for a researcher to look through a CTX image and look for small craters. During the lifetime of the MRO, researchers found over 1,000 craters this way. Not only are they looking for craters, they are also interested in changes on the surface: dust devils, moving dunes, landslides and the like.
AI researchers at NASA's Jet Propulsion Laboratory in Southern California have been trying to do something about it all along to find interesting things in all of these images. You're developing a machine learning tool to handle some of that workload. The tool had its first success on August 26, 2020.
The HiRISE (High Resolution Imaging Science Experiment) camera on board NASA's Mars Reconnaissance Orbiter captured this image of a crater cluster on Mars that was first discovered by artificial intelligence (AI). The AI first discovered the craters in images captured with the orbiter's context camera. Scientists followed up with this HiRISE image to confirm the craters. Photo credit: NASA / JPL-Caltech / University of Arizona
Sometime between March 2010 and May 2012 a meteor struck the thin Martian atmosphere. It broke into several pieces before hitting the surface and only produced a black spot on CTX camera images of the area. The new AI tool, an automated classifier for craters with fresh impact, has found it. Once it did, NASA used HiRISE to confirm this.
This was the classifier's first find, and in the future NASA expects AI tools to do more of this type of work, freeing the human mind for more sophisticated thinking. The crater classifier is part of a broader JPL campaign called COSMIC (Capturing Onboard Summarization for monitoring image changes). The aim is to develop these technologies not only for MRO, but also for future orbiters. Not just on Mars, but wherever orbiters are.
Machine learning tools like the crater classifier need training. 6,830 CTX camera images were fed for his training. Among these images were images with confirmed craters and others that did not contain craters. That taught the tool what to look for and what not to look for.
Mars is covered in craters, and many of them are obvious and don't require AI to find them. This is one of many fresh impact craters discovered by the UA-guided HiRISE camera and orbiting the Red Planet aboard NASA's Mars Reconnaissance Orbiter since 2006. (Photo: NASA / JPL-Caltech / MSSS / UA).
After training, JPL removed the system's training wheels and released them on over 110,000 images of the surface of Mars. JPL has its own supercomputer, a cluster of dozens of high-performance machines that can work together. The result? The AI running on this powerful machine took just five seconds to complete a task that would take around 40 minutes for a human. But it wasn't easy.
"It would not be possible to process more than 112,000 images in a reasonable time without spreading the work over many computers," said JPL computer scientist Gary Doran in a press release. "The strategy is to break the problem down into smaller pieces that can be solved in parallel."
Although the system is powerful and a huge time saver, it cannot be operated without human error.
Victoria crater is perhaps the most famous crater on Mars. By the time the Mars Reconnaissance Orbiter landed on Mars two years after Opportunity touched down, Opportunity had left Eagle Crater and traveled the 6 km to Victoria Crater. Retrieved from NASA / JPL / University of Arizona – http://photojournal.jpl.nasa.gov/catalog/PIA08813, public domain, https://commons.wikimedia.org/w/index.php?curid=4211043
"AI cannot do the kind of skilled analysis that a scientist can do," said JPL computer scientist Kiri Wagstaff. “However, tools like this new algorithm can be your assistants. This paves the way for an exciting symbiosis of human and AI researchers working together to accelerate scientific discovery. "
As soon as the crater seeker has achieved a hit in a CTX camera image, HiRISE must confirm this. It did so on August 26, 2020. After the crater seeker marked a dark spot in a CTX camera image of a region called Noctis Fossae, the HiRISE's power took the scientists on for a closer look. This confirmed the presence of not one crater, but rather a cluster of several that arose from the objects that fell on Mars between March 2010 and May 2012.
With this initial success, the team developing the AI has sent more than 20 more CTX images to HiRISE for review.
NASA's HiRISE camera on Mars Reconnaissance Orbiter captured this unusual crater or pit on the surface of Mars. Frozen carbon dioxide gives the region its unique “Swiss cheese” -like appearance. Image: NASA / JPL / University of Arizona
This type of software system cannot yet run on an orbiter. Only a terrestrial supercomputer can perform this complex task. All of the data from CTX and HiRISE is sent back to Earth where researchers ponder and look for interesting images. However, the AI researchers developing this system hope that this will change in the future.
"The hope is that in the future, AI could prioritize orbital images that scientists are more interested in," said Michael Munje, a Georgia Tech graduate student who interned at JPL on the classifier.
This development has another important aspect. It shows how older, still operational spaceships can be supplied with energy again with modern technological power and how scientists can draw even more results from them.
Ingrid Daubar is one of the scientists working on the system. She believes this new tool will help find more craters that elude human eyes. And if possible, it will help us to expand our knowledge of the frequency, shape, and size of meteorite impacts on Mars.
"There are probably a lot more effects that we haven't found yet," said Daubar. "This advancement shows you how much you can do with veteran missions like MRO using modern analytical techniques."
"Observe a lot, best come back"
This new machine learning tool is part of a broader NASA / JPL initiative called COSMIC (Content-based On-Board Summarization for Monitoring Infrequent Changes). The motto of this initiative is: Observe a lot, it's best to return.
The idea behind COSMIC is to “create a robust, flexible orbital system for carrying out planetary measurements and monitoring changes in the Martian environment”. For reasons of bandwidth, many images are never downloaded to Earth. Among other things, the system autonomously detects "changes in unmonitored areas and provides relevant, informative descriptions of onboard images to recommend the prioritization of downlinks". The AI that finds craters is only one component of the system.
Data management is a big and growing challenge in science. Other missions such as NASA's Kepler spacecraft to hunt planets generated an enormous amount of data. In an effort parallel to what COSMIC is trying to do, scientists use new methods to comb all of the Kepler data and sometimes find exoplanets that were overlooked in the original analysis.
In 2016, a student found four new exoplanets "hidden" in data from NASA's Kepler spacecraft. The discovery underscores the challenge of managing all of the data returned from space missions. Image: University of British Columbia
And the upcoming Vera C. Rubin surveying telescope will be another data-generating monster. In fact, managing all of the data is seen as the hardest part of the whole project. About 200,000 images per year or about 1.28 petabytes of raw data are generated. That is far more data than humans can process.
In anticipation of so much data, the people who looked at the Ruby Telescope developed the LSSTC Data Science Fellowship Program. It's a two-year graduate curriculum program covering topics such as statistics, machine learning, information theory, and scalable programming.
It is clear that AI and machine learning need to play a bigger role in space science. In the past, the amount of data returned from space missions was much more manageable. The tools used to collect the data were simpler, the cameras were much lower resolution, and the missions didn't take that long (excluding the Viking missions).
And while a system designed to find tiny craters on the Martian surface may not capture most people's imaginations, it is an indication of what the future will be.
One day, more scientists will be relieved of hours of sitting going through images. You can delegate some of this work to AI systems like COSMIC and its crater finder.
We'll probably all benefit from this.