Bradley, a tenth grader in Ardent AI Scientist, won a Grand Award in Earth and Environmental Sciences at the 76th Regeneron International Science and Engineering Fair, held in Phoenix, Arizona. The fair brought together more than 1,700 students from more than 67 countries, regions and territories. He advanced to ISEF from the Buckeye Science & Engineering Fair.
The project
Coral reefs support more than a billion people, and rising ocean temperatures are making bleaching more frequent. Current monitoring often detects bleaching stress only once it has become fatal to the coral.
Bradley's project, Machine Learning for Coral Bleaching Diagnosis and Forecasting Using Images and Environmental Data, approached the problem in two phases.
In the first, he trained two convolutional neural networks that told healthy coral from bleached coral with 88% accuracy, and compared five machine learning models on environmental data. The comparison produced the finding that shaped the rest of the work: reef images give the most precise diagnosis, while environmental data is most useful as an early warning.
In the second phase he built on both. A new neural network grades coral on a five-level severity scale with 92.9% accuracy and a quadratic weighted kappa of 0.97, separating degrees of bleaching that earlier yes-or-no models could not. Five forecasting models predict bleaching events up to twelve weeks in advance with 75% accuracy, which gives reef managers time to act. He built the system entirely on free computational tools, so researchers with limited resources can use it.
A judge at the Buckeye Science & Engineering Fair wrote that the project "would be credible as an undergraduate capstone or early-stage graduate contribution."
Two years, two ISEF projects
Bradley joined Ardent AI Scientist in summer 2024, as he entered ninth grade. He brought a strong foundation in math, physics and Python from his Ardent classes, and no research experience. He was one of three students in the program's pilot cohort.
His first project used NASA light-curve data to detect exoplanets. Comparing AI models, he found that a simple decision tree reached more than 99% accuracy while a neural network struggled on the small dataset. That project qualified for ISEF. The coral system, his second, won a Grand Award.