As a child in Petrozavodsk, Russia, Dmitrii Kochkov loved solving geometry puzzles with his parents. In high school he participated in math and physics competitions. By the time he reached graduate school, he had turned toward machine learning and quantum physics to solve tough problems. The now 34-year-old joined Google Research as an AI resident in 2019, using machine learning to create programs that could solve interesting equations. His career then turned toward weather and climate change.
Weather is governed by fluid dynamics, described in part through partial differential equations. These equations underpin the computer models that meteorologists use to forecast daily and weekly weather. The model then applies global weather data to tell us whether we should expect rain, sunshine or extreme heat. But some processes, such as cloud formation, must be approximated in the models, which can lead to errors and biases. Kochkov and his teammates built NeuralGCM (for “general circulation model”) to replace those approximations with machine-learning predictions trained on past weather data. The system can predict weather conditions on par with the best models (up to 15 days out) and reproduce past temperature patterns as accurately or more so than current gold-standard models.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Christie Hemm Klok
NeuralGCM is starting to show results. Researchers at the University of Chicago have used it to forecast the start of monsoon rains in India up to one month in advance, providing crucial information to millions of farmers. Kochkov’s team is creating a newer version of the model that is easier to use, which will eventually allow scientists to study how climate change is altering weather extremes and water availability. This comes at a time when funding for this kind of research is unstable. “Enabling people to do the best work they can with given resources seems more important than ever before,” he says.
This article is part of “The Young American Scientists,” an editorially independent project that was produced with financial support from Regeneron.
