Lily Larsen '26 on Accessibility, Neural Nets, and Cold Emailing NASA
By Ray Perry
This summer, Lily Larsen ‘26 worked with something truly out of this world — specifically, satellite data.
Larsen interned with the University of Alaska Fairbanks, a satellite facility contractor for NASA. She worked at their DAAC, a data distribution center, which transmits commands to satellites and downloads and processes the return data.
The team she worked with was focused on processing InSAR data, which is a specialized kind of radar data that uses light waves to track deformation of the Earth.
InSAR data can track “even a millimeter of deformation,” Larsen said, making it useful for geophysical analysis, from natural disaster research to natural resource management.
“A volcano that’s nearing eruption might start to swell [as] magma fills the magma chamber. Only centimeters [of swelling] until it erupts, but these satellites can actually track that,” Larsen explained. “Or [with] earthquakes, you can see the waves of the earthquake in these satellite images. [Or] to track how much groundwater is being used—you’ll see the ground cave down in places where people are using water sources.”
However, InSAR data is difficult to process. Raw data is recorded as an image, with different levels of deformation displayed in bands of color — making it difficult to determine how much deformation exists in three dimensions.
“Ideally we’d just have an image where a high value is a high amount of deformation,” Larsen said. “[It] can take a few hours of processing [to] a full week of processing for a single image.”
When she joined, the research team was training a machine learning model to process the images faster than typical programs could, which would make large quantities of data far more accessible. While Larsen did work on machine learning, doing quality testing and fixing bugs, helping with machine learning was only a small part of the internship.
The main part? Accessibility. “I made a documentation website and packaged it as a library, made it something that’s actually usable to the general public [and] to academia as a whole,” they said. “Before this, other scientists couldn’t use it if they wanted to.”
Larsen met with officials from the United States Geological Survey (USGS), Arizona Department of Water Resources, and NASA’s Jet Propulsion Laboratory (JPL) to discuss the applications of the tool.
“You have to convince the broader scientific community that this tool is worth using and that the effort of learning it is worth it,” Larsen said. “I was trying to [decrease] the effort of learning it so that it might seem more worth using.”
Larsen hopes that the machine learning model can be run as part of regular data processing for InSAR satellite data in the future. Beyond that, they also hope that eventually it will be used by the United States Geological Survey to study earthquakes.
“We’re closest to [using] it to create very large sets of earthquake data for people to study,” Larsen said. “[Or] creating data sets of volcanoes [that] are having these deformation events.”
Additionally, NASA and the Indian Space Research Organization (ISRO) are collaborating to launch a satellite called NISAR in early 2024. NISAR will use InSAR to collect deformation data, circling the planet every 6 days on average. With this amount of additional imaging, the NISAR satellite could expand the scope of InSAR research to include disaster monitoring.
“It might be possible to combine [NISAR data] with other satellites to do very rudimentary monitoring work, like natural disaster monitoring for volcanoes,” Larsen said. “But that’s really far out and the model would need to be much more sophisticated.”
Reflecting on the experience, Larsen said that the coolest part was “meeting with people from JPL.”
“As a kid who was always a space nerd, it’s like wow, I’m talking to someone from JPL right now,” they said.
For students looking to find summer research positions, Larsen shared an (admittedly daunting) trick: cold emails. She found this position by reaching out to a parent’s friend about summer internship opportunities, and then cold-emailing the sources the friend directed her to.
“At universities that conduct a lot of research, [professors] would often love to take on a student as an intern, but they don’t have a set pathway for creating applications for student positions,” Larsen said. “Cold emailing people can be a really successful strategy.”