AI-Enabled Weather Forecasting Reaches Millions of New Farmers in India

A CEGA-funded pilot and randomized evaluation informs the scaling of long-range monsoon forecasts to 38 million farmers in India by Precision Development and the Government of India

At the beginning of each agricultural season, farmers must make a number of important production decisions: when to plant crops, how much to plant, and what to plant. The outcomes of these decisions depend heavily on external factors such as weather and market prices, which can be volatile and unpredictable. For the roughly 450–500 million small-scale farming households worldwide, many without access to government-provided social safety nets or insurance schemes, weather-based risks discourage farmers from investing in productive agricultural inputs and practices, ultimately contributing to a vicious “low-risk, low-return” trap.

Weather-based risk is particularly salient for farmers who depend on rainfall for their crops. Across the globe, millions of farmers tie their agricultural production cycles to the monsoon season. In recent years, improvements in AI models and statistical methods, paired with historical rainfall data, have unlocked accurate predictions up to 30 days in advance of the monsoon onset — providing enough lead time for farmers to adjust their plans for the upcoming agricultural season. 

In 2022, researchers Fiona Burlig, Amir Jina, Erin Kelly, Gregory Lane, Faraz Hayat, and Harshil Sahai, supported by CEGA and the Abdul Latif Jameel Poverty Action Lab (J-PAL), tested a long-range seasonal weather forecast with farmers in about 250 villages in Telangana, India. Using a randomized evaluation research design, the researchers found that farmers who received the forecast updated their beliefs about the predicted monsoon onset date and, in turn, adjusted their production decisions around land use, crop choice, and input use–translating to meaningful improvements in wellbeing such as per-capita food consumption.

Building on these findings, the Government of India partnered with the Human-Centered Weather Forecasts Initiative, supported by UC Berkeley atmospheric scientist Bill Boos, to refine and select a weather forecasting model for use at scale. 

Demonstrating that the long lead-time precipitation forecasts made by these AI models are of practical use in a tropical region where people live is a major step forward — no one really knew that before we did this work,” said Boos, a UC Berkeley professor of earth and planetary science.

Precision Development (PxD) — a global nonprofit that supports small-scale farmers with digital advisory services, and itself an outgrowth of two CEGA-funded randomized evaluations — led message design and testing for the scaling effort. The initiative has delivered weather forecasts via SMS to 38 million farmers, and via voice messaging to an additional 1 million farmers.

For more regarding this endeavor, read the UC Berkeley News article.