Google’s Climate AI team developed a model that predicts crop yields (with 92% accuracy) using satellite imagery, soil data, and weather forecasts. Piloted in India and Kenya, it alerted farmers to a 2024 maize blight 6 months pre-harvest.
Technical Insight:
- Combines convolutional neural networks (CNNs) for image analysis with transformer models for temporal patterns.
- Trained on 40 years of agricultural data from FAO and NASA.
Impact:
- Reduced pesticide use by 30% via targeted interventions.
- Insurers use predictions to adjust premiums dynamically.
Limitations: Requires high-resolution satellite data, inaccessible in remote regions.