NEWS

When Yusuke Hatanaka first arrived in Hawaiʻi a decade ago, it wasn’t to change the way scientists understood rainfall. It was just for school and a change of pace. He enrolled at Leeward Community College with no fixed major—a flexibility that he said was freeing, especially coming from Japan, where academic paths are often decided early and set in stone. Eventually, he found his way to computer science, transferring to the University of Hawaiʻi at Mānoa, and then continuing on to a master’s program where he worked with Dr. Peter Sadowski, associate professor in the Department of Information and Computer Sciences on a project developing machine-learning methods to predict the risk of adverse weather and climate events.

“Even though I am a programmer, I’ve always been interested in climate and rainfall mechanisms ever since a geography class in high school,” he recalls. “Not just the concept of rain, but why it falls where it does.”

Now a PI-CASC graduate scholar and doctoral student at UH Mānoa, Hatanaka is building on his master’s work, using machine learning to help predict rainfall trends across Hawaiʻi’s complex terrain. One way to do that is by using global models, which are a simulation of the climate that describes the state of the atmosphere in a consistent format. 

“The problem with global climate models,” he explained, “is that they’re like a blurry photo—only a few pixels cover the whole state.” 

These models, though powerful, are too coarse to show the nuances of rainfall in the smaller areas of Hawaiʻi, where rain might pour on one side of a mountain and leave the other dry. His project aims to fix that, using a technique called downscaling, which teaches a machine learning model to zoom in. 

“It’s like giving the camera more pixels,” he said.

Using historical rainfall data from statewide weather stations, Hatanaka is training models in such a way that they can extrapolate to new locations and times that do not have historical data, and translate coarse global predictions into high-resolution forecasts. 

Hatanaka hopes his research will serve as a valuable resource for communities and policymakers working to navigate the growing challenges of a changing climate.

“Improving the accuracy of precipitation patterns is an important issue because rainfall shapes the ecosystem, culture, freshwater supply, and many other defining features of the islands,” he said. “The better we understand the future rainfall patterns, the more resilient the Hawaiian islands remain against climate change.”