Development Innovation Lab

AI-Powered Weather Forecasts, Built for the World’s Smallholder Farmers

As climate change makes weather more volatile and extreme, smallholder farmers need timely, reliable weather forecasts to make decisions that will protect their incomes and food security.
Dark storm clouds are visible hovering above a cluster of small brown houses. A person wearing white clothing and a white cap holds a farming tool and stands in the foreground on a small plot of farmland, with short green plants growing out of red dirt.
Storm clouds approach a farming community near Iringa, Tanzania. Credit: UN Photo/Wolff.

 

The opportunity

Worldwide, nearly 500 million small family farms support the livelihoods of more than two billion people, many with little or nothing to fall back on if a harvest fails. Unexpected weather events like late monsoon onset, dry spells, and extreme heat can cause smallholder farmers to lose crops they rely on.

Research has shown that accurate, timely, personalized weather forecasts can help farmers better prepare for these events, improving productivity and raising incomes for some of the world’s poorest people.

This effort convenes experts at the University of Chicago and the University of California, Berkeley to support national meteorological agencies around the world to tailor and operationalize AI weather models to deliver forecasts that are relevant to farmers. Initial forecasts will include monsoon onset and cessation, harvest-related rainfall, and extreme heat.

The work is conducted in collaboration with ecosystem funders and partners who are investing in complementary efforts like additional open-source forecasting capabilities, app development, SMS message delivery, and building financing mechanisms to sustain and expand these services.

The technology

For many farmers in low- and middle-income countries, high-quality forecasts have long been out of reach, in part because accurate forecasts have been prohibitively expensive to create and logistically complicated to deliver.

But AI-based models now match or exceed operational physics-based models on many tasks at a tiny fraction of the compute cost, opening the door to localized, decision-oriented forecasts of phenomena that matter to farmers: monsoon and rainy-season onset, dry spells, short-range rainfall, and heat stress.

The fastest path to getting forecasts to farmers runs through systems that already have reach. Governments and mobile network operators already collectively reach hundreds of millions of farmers through voice, SMS, WhatsApp, radio, mobile applications, and more.

AI can be used across four distinct functions:

  1. Generating the forecasts optimized for specific agricultural use cases, by combining open-access AI weather models with physics-based models;
  2. LLM-based software that automates benchmarking and AI weather model combinations, such that government offices and others can do it at much lower cost;
  3. Gathering information from farmers, through AI voice agents that collect location, crop, and feedback data at a fraction of the cost of traditional phone surveys, already tested with well over 100,000 farmers; and
  4. Disseminating forecasts and advisory content, including AI-drafted texts, AI-voiced messages, two-way AI chatbots, and AI-enabled personalization of guidance based on farmers characteristics and preferences.

The evidence

There is strong evidence that farmers change agricultural decisions in response to accurate, timely weather forecasts. They make different, better decisions—changing what and when they plant, spend, and invest—and those decisions raise their incomes and cut crop losses.

  • Rigorous research from Ghana and Pakistan indicates that farmers avoid applying pesticides or fertilizer, or irrigating, just before rain is forecast, to avoid waste (Rudder and Viviano 2024; Fosu et al. 2018). In India, farmers who were randomly assigned to receive a seasonal monsoon forecast substantially altered their planting and investment decisions to better align with the forecasted seasonal timing (Burlig et al. 2025). And a weather-based pest advisory for potato farmers in Bangladesh helped reduce the risk of crop loss (Barnett-Howell 2021).
  • Forecasts can also increase resilience to disasters. For instance, flood forecasts in Bangladesh led to timely evacuations, minimizing financial losses (Webster et al. 2010). This is corroborated by national panel data in India, where farmers increased planting-stage investments in response to accurate, favorable monsoon forecasts (Rosenzweig and Udry 2013).

Join us

The University of Chicago (DIL) and Berkeley, with partners, are actively setting up this initiative to support government partners in scaling this technology and setting up context-specific dissemination pathways.

We are facing an unprecedented, urgent opportunity to extend this same rigorous, evidence-based, government-led approach to more governments, with the potential to reach hundreds of millions more farmers. Contact us to learn more.

Leadership

William Boos
Professor of Earth and Planetary Sciences
University of California, Berkeley

Terrina Govender Heath
Director, Policy and Strategy
University of Chicago

Katie Kowal
Director, Operational Forecasting
University of Chicago

Colin Aitkin
Head of Statistics
University of Chicago

General Inquiries

Terrina Govender Heath