A-Z Index | Directory | Careers

NERSC Summer Student Works Toward Machine Learning For a Rainy Day

August 15, 2022

By Elizabeth Ball
Contact: cscomms@lbl.gov

JamesDuncanposter2

James Duncan discusses his work at the 2022 summer student poster session. (Credit: Margie Wylie, Berkeley Lab)

The next time an accurate weather prediction saves you from a sudden downpour, you may have James Duncan’s 2022 summer project at Lawrence Berkeley National Laboratory (Berkeley Lab) to thank.

Duncan, a Ph.D. student in biostatistics at the University of California, Berkeley, has been working with National Energy Research Scientific Computing Center (NERSC) data architect Wahid Bhimji, NERSC Exascale Science Applications Program engineer Shashank Subramanian, and NERSC machine learning engineer Peter Harrington as part of the 2022 Berkeley Lab Computing Sciences Summer Program. Their team is developing deep generative machine-learning models for forecasting precipitation using the Perlmutter supercomputer at NERSC.

In particular, Duncan has been focusing on precipitation models, improving their resolution and incorporating an element of randomness to better mimic real-world conditions. Machine-learning models “are quite a blurry picture of precipitation compared to the ground truth,” said Duncan. “So one of the goals of this project was to try and better capture those fine-scale structures — but then also, hopefully, to predict some of the extreme states of the atmosphere like tropical cyclones and atmospheric rivers, because obviously that’s really important for disaster preparedness and safety.”

Duncan compared the improvement in resolution to predicting the behavior of the precipitation in a single large cloud, rather than a whole weather system.

The Computing Sciences Summer Program offers graduate and undergraduate students the opportunity to work with Berkeley Lab researchers on focused projects each summer.

"Working with James has been a pleasure,” said Harrington, who has served as Duncan’s mentor this summer. “We started off well since James had some experience from a previous project involving machine learning and climate science, so he already had familiarity with the dataset. This made it possible to make quick progress, and James has now developed a powerful model that is impressive in its ability to make accurate predictions on a challenging real-world task."

The summer program has ended, but the project hasn’t: after a few more experimental runs using Perlmutter, the team will write up results for publication. Duncan also hopes to continue building on this work in his own separate research, incorporating more direct predictions over the longer term.

Duncan will also continue the path toward his Ph.D., and sees himself doing further work on weather prediction models, a relatively recent interest that has proven rewarding.

“I find the data to be super interesting; it’s really motivating. It’s all around us, you can see it. It’s interesting to know the history of weather on this planet, and to see how far back you can go. I’d definitely like to keep working in this space,” he said.

He’s also noting the benefits of a possible future in the national lab system. “I worked at Los Alamos one summer during my master’s degree and really loved it there, and I’m really enjoying working at NERSC. So you know, I think national labs are a really nice place to be.”


About Computing Sciences at Berkeley Lab

High performance computing plays a critical role in scientific discovery. Researchers increasingly rely on advances in computer science, mathematics, computational science, data science, and large-scale computing and networking to increase our understanding of ourselves, our planet, and our universe. Berkeley Lab’s Computing Sciences Area researches, develops, and deploys new foundations, tools, and technologies to meet these needs and to advance research across a broad range of scientific disciplines.