David Plotkin

University of Chicago

It can take a long time for a computer models to make hurricanes – especially rare, highly intense hurricanes and typhoons.

David Plotkin wants to change that. The Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient is out to improve weather models so high-performance computing (HPC) systems can more quickly and accurately simulate big storms. That ultimately could help scientists better understand how a warming climate will affect tropical storm intensity.

“Usually we study how hurricanes form just by running really big computer models, and the strategy has been to run bigger and bigger ones,” says Plotkin, a doctoral student in earth sciences at the University of Chicago. “We’re trying to be a little bit more systematic and a little bit more clever about how we do that.”

It’s important work. Climate researchers think that as the planet warms due to greenhouse gas emissions, hurricanes and typhoons are likely to be more powerful “and the strongest ones produce all the damage,” Plotkin says. To minimize human impact, it’s “important to predict what’s going to happen with those strong storms under global warming, and in order to predict that, we need to know how they form.”

Climate models aren’t great at producing hurricanes and typhoons, also known as tropical cyclones. Because they happen irregularly – especially the biggest ones – researchers might have to simulate decades of weather before their algorithms produce a cyclone. “You end up wasting a lot of computing time, waiting around for the model to form the storm. We’re trying to save that time by systematically biasing the model into being more likely to form storms but in a way where we can keep track” of how the model is tweaked so the results are still relevant to the real world.

Plotkin: science image

To do this, Plotkin and Dorian Abbot, his doctoral advisor, use action minimization, a mathematical technique borrowed from molecular dynamics (MD) models, which calculate chemical reactions and protein structures. Both kinds of models try to find the likeliest path for a transition: for MD, from one compound or protein shape to another; for climate, from calm seas to violent storms. Plotkin and his colleagues build artificial climate model trajectories that lead to storms, then minimize how these pathways deviate from true model dynamics.

Researchers use the computation time and power that action minimization saves to increase the model’s resolution, calculating what’s happening at points spaced, say, 30 kilometers apart instead of 50. The resulting simulation includes more, and more precise, physics for a detailed outcome.

In the end, Plotkin hopes the models will help scientists better understand how these violent storms form and how they rapidly intensify. That information also will help further improve the computer models.

“The models are probably missing some parts of the physics that are usually not very important but become important at hurricane conditions,” Plotkin says. Multiple factors, such as how energy is transferred between the ocean and the atmosphere, “change pretty drastically at hurricane conditions, so our models might not be doing a good job with that.”

In test runs, the action minimization approach generated fairly realistic large-scale simulations of how wind speed increases over time as a storm forms. Plotkin is working on making the code usable for high-resolution modeling.

Plotkin got his first exposure to modeling tropical cyclones during his 2014 practicum at Argonne National Laboratory. With Robert Jacob, he worked on improving simulations of hurricanes and typhoons in global climate models. Though the summer experience hadn’t directly affected his thesis research, Plotkin says, it taught him how to use global climate models and “it helped convince me that there were interesting and worthwhile questions there to be asked.”

As the son of a software-engineer father and a biophysics-professor mother in Boston, Plotkin grew up surrounded by science. He likes his research because it’s a good intersection of applied mathematics and “something that is relevant in a real-world kind of a way.”

Plotkin will face the real world when he graduates in late 2017. He’s considering positions that would employ machine learning and data science skills. “If that’s related to climate, that would be awesome,” he adds, but he’s also considering medical research.

Image caption: Rapid intensification (RI) of tropical cyclone surface winds as simulated by the WRF model using action minimization. Consistent with observations, RI occurs over a 48-hour timescale and results in a doubling of the maximum surface wind speed.