UMD Biological Sciences Ph.D. Student Wins 2026 Cherry Blossom Prediction Competition
The statistical modeling skills Wesley DeMontigny mastered while studying genomics gave him an upper hand.
University of Maryland biological sciences Ph.D. student Wesley DeMontigny won the Award for Most Accurate Prediction in the 2026 International Cherry Blossom Prediction Competition.
The annual contest—which is sponsored by the Washington Statistical Society, George Mason University’s Department of Statistics, the American Statistical Association and Columbia University’s Department of Statistics—challenges contestants to predict peak bloom dates for select trees in Washington, D.C.; New York City; Kyoto, Japan; Liestal-Weideli, Switzerland; and Vancouver, Canada. Submissions also must include a reproducible analysis, including data and code. DeMontigny’s cherry blossom forecast across the cities was the most accurate of more than 30 entrants from students, researchers and citizen scientists from around the world, earning him an award and a cash prize.
“I was driving up from Georgia with my wife to see family, and we stopped at a Starbucks when I saw the congratulations email,” DeMontigny said. “I was just really happy to find out that I won.”
DeMontigny developed his statistical expertise while researching genomics at UMD. Working with Cell Biology and Molecular Genetics Professor Charles Delwiche, DeMontigny creates statistical models that help researchers identify which genes may be missing when they assemble genomes. DeMontigny said that skill set was remarkably transferable to this competition because both cases require statistical reasoning when data is limited.
“One part of the competition that got me very excited is that you’re asked to predict the bloom date for some sites that only have three years' worth of data—which is nothing,” DeMontigny said. “It’s a small data problem, where you don’t have a lot of information, but you have to make the best use of what you have.”
To address that issue, DeMontigny first created a statistical model that ignored cherry blossom blooming altogether. Instead, he built a machine learning model to predict the weather and climate for the 2026 spring season for cities all around the world. He trained it on historical climate data from NASA and the National Oceanic and Atmospheric Administration, which provided a much more robust dataset than what was available for cherry blossom blooms. Then, using a technique called transfer learning, he applied what the model deduced about temperature dynamics and climate cycles to predict cherry blossom data.
“The overall strategy here is to take a slightly adjacent problem where I have a lot of data to inform my predictions where I basically have no data,” DeMontigny explained. “The climate forecasting model has all the information you would need to predict the climate. And transfer learning says that cherry blossom blooms aren’t so different from other climate factors, so the information to predict cherry blossom blooms is probably in the model, too.”
As cherry blossom peak bloom approached, DeMontigny got nervous that his predictions were off. In Washington, D.C., the National Park Service estimated that peak bloom would occur a week later than he predicted.
“But luckily it didn’t play out that way,” DeMontigny said, noting that his prediction for Washington, D.C. was just one day off. “I expected that my model would perform strongly, but to be the best-performing one? That was really exciting.”
