India’s AI monsoon predictions ‘inspire investment’

[NEW DELHI, SciDev.Net] The success of Artificial Intelligence (AI)-enhanced monsoon forecasting in India has accelerated work on weather prediction models in other regions, with 30 countries set to benefit from the whole project, according to US scientists.

“Motivated by the success of the Indian monsoon work, Chicago University has just received support from the Gates Foundation to benchmark existing models over East and West Africa with a focus on rainy seasons and heatwaves,” Pedram Hassanzadeh, co-director of the university’s Human-Centred Weather Forecasts initiative, told SciDev.Net.

Benchmarking is important to assess how well both traditional and AI-based models can forecast the weather and evaluate important atmospheric phenomena such as the seasonal onset and progression of monsoons.

“A lot more can be done in India and in other regions around the world, once the forecasts are properly benchmarked,” said Hassanzadeh.

“Time and resources for benchmarking aside, scaling up and fully exploring the power of the existing methods needs resources that enable real-time forecast generation and dissemination at large scales.”

This summer, 38 million Indian farmers benefited from AI-powered forecasts using a neural general circulation model (NeuralGCM), four weeks ahead of the onset of monsoon rains.

Researchers described NeuralGCM as a hybrid model that combines traditional, physics-based forecasting with machine learning to simulate the Earth’s atmosphere.

Developed by Google, NeuralGCM has stood up to benchmarking against both traditional physics-based models as well as other AI-based atmospheric models, demonstrating superior computational efficiency and strong performance on a range of weather and climate forecasting metrics, says Hazzanzadeh.

It is set to be rolled out in 30 countries globally in the next two years.

In India, the model was able to forecast a three-week pause in the progress of the monsoon, which typically makes landfall at the southern tip of the Indian sub-continent in early June before progressing northwards.

The AI-assisted forecasts gave farmers time to make critical decisions such as when to plant crops, say the researchers from the University of Chicago Institute for Climate and Sustainable Growth, working alongside the Indian government.

The AI-assisted model uses software that can be run on a laptop, making high-quality forecasting accessible to scientists and farmers. In contrast, traditional weather and climate models are notoriously expensive and rely on supercomputers to analyse complex meteorological data.

Pramod Kumar Meherda, a senior official in the agriculture ministry, told SciDev.Net: “The programme harnesses the revolution in AI-based weather forecasting to predict the arrival of continuous rains, empowering farmers to plan agricultural activities with greater confidence and manage risks.”

Michael Kremer, economist at the University of Chicago and co-director of the Human-Centred Weather Forecasts Initiative, says that disseminating AI weather forecasts has a great return on investment, “likely generating more than US$100 for farmers for every dollar invested by the government”.

Kremer believes the initiative is especially helpful for smallholder farmers, whose livelihoods are increasingly threatened by climate change.

However, Indian agricultural scientists say the new AI-based model needs to be developed further, to provide the most useful data for all those who need it.

“Sending out messages to 38 million farmers is impressive, but the message content should pair rainfall signals with data on soil moisture, vapour pressure deficit, heat stress forecasts, and crop-stage sensitivity,” said Arun Shanker, principal scientist and plant physiology specialist at the Central Research Institute for Dryland Agriculture, Hyderabad.

“For sowing decisions, a bad early onset call can mean seedling mortality, re-sowing costs, and lost season time,” Shanker added.

The researchers behind the Human-Centred Weather Forecasts initiative are working on similar programmes in other low- and middle-income countries and training meteorologists on how to use AI models effectively.

Launched this year, the programme partners five countries — Bangladesh, Chile, Ethiopia, Kenya, and Nigeria. Chicago University says there are plans to partner with ten more countries in 2026 and 15 more in 2027, broadening reach and impact to millions of farmers.

Hassanzadeh believes that the next generation of these AI weather models will have more impact.

“The current AI weather models are among the greatest achievements of AI for science so far, but I can say we are at the beginning of an AI-driven, second revolution in weather forecasting,” he said.

This piece was produced by SciDev.Net’s Asia Pacific desk.