Human-Centered Forecasting: an AI-led transformation of climate adaptation for smallholder farmers

Abstract

We often take high-quality weather forecasts for granted in high-income countries, but much of the tropics still relies on low-quality forecasts—for both scientific and economic reasons—despite facing high climate risk. Striking recent advances in AI weather models—some outperforming the physics-based models that have formed the backbone of modern weather forecasting for 100 years—offer a path to transform and democratize forecasting in low- and middle-income countries. This talk focuses on three connected research papers—an RCT on forecast dissemination that explores the role played by farmer beliefs; an agricultural decision-focused approach to benchmarking AI weather models; and a decision theory-driven creation of weather forecasts for farmers in India. Together these led to efforts to scale forecast dissemination with the Indian government to 40 million farmers in 2025, the largest targeted AI-based weather outreach to date. I will also discuss the ongoing research-based efforts through AIM for Scale to expand reliable weather forecasting and delivery to an additional 100 million farmers globally.