Earlier in 2025, 38 million farmers in India received a first-of-its-kind weather forecast about the monsoon onset four weeks in advance. Led by the Indian Ministry of Agriculture and Farmers’ Welfare in collaboration with the University of Chicago’s Human-Centered Weather Forecasts (HCF) Initiative, a multi-disciplinary team of climate scientists, artificial intelligence (AI) experts, economists and social scientists rolled out an AI-based weather forecast programme to provide long-range forecasts about the monsoon arrival across India. This period of continuous rainfall marks the beginning of the rainy season in India, which is a crucial time for growing both staples and cash crops for more than 100 million farmers across the country.
Farmers usually rely on past knowledge, word-of-mouth updates, and government bulletins and advisories to anticipate when the rainfall might reach their specific region once onset has been declared. However, predicting the Indian monsoon has been a challenge for many decades. “Monsoons are notoriously complex continent-sized patterns of wind that persist for months and yet interact with large numbers of small-scale clouds that each last only a few hours,” explains William Boos, an atmospheric scientist and monsoon expert at UC Berkeley, and a collaborator on the project.
Monsoon onset was identified as a crucial factor for Indian farmers by another team of researchers at the University of Chicago. They found that providing a monsoon forecast a month in advance resulted in substantial savings and losses averted for farmers in the Indian state of Telangana, who tailored their investments for planting based on this information. The results from the experimental study substantiated that long-range forecasts can help farmers adapt to climate change1.
However, running physics-based numerical weather prediction models requires a lot of computing power, a highly skilled workforce to make them work for specific use cases, and most models provide reliable forecasts only a few days in advance. In this respect, recent advancements in AI weather forecasting have been a game-changer.
“This is really a revolution that started in early 2022 when two papers2,3 showed that a deep neural net trained on ERA5 can have forecast accuracy comparable to the best numerical weather models,” says Pedram Hassanzadeh, an AI expert and extreme weather researcher at the University of Chicago. “And once trained, these can be 100,000 times faster to run [than numerical weather predictions]. You could run it on a laptop and do the best weather forecast.”