UC Launches AI Forecasting Tool for 70+ Crop Yields Across California

New model predicts yields for 70 crops with county-level precision, offering powerful insights for growers, researchers, and policymakers.

A Breakthrough in Predictive Agriculture

The University of California has released a powerful new artificial intelligence tool capable of forecasting yields for more than 70 crops across California, with a focus on county-level accuracy. Developed by researchers from UC Davis and Texas A&M University, the model integrates multiple data sources—satellite imagery, weather data, evapotranspiration, and soil characteristics—to deliver precise seasonal predictions tailored to California’s diverse growing regions.

The project, presented at the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), represents one of the most comprehensive AI tools ever created for large-scale agricultural forecasting. The research paper, titled California Crop Yield Benchmark, is publicly available on Cornell University’s Website. 

Official details and the research team’s objectives are summarized on the UC Davis Artificial Intelligence Institute for Next Generation Food Systems (AIFS) website, which highlights how the tool will improve agricultural productivity and resource management. Additionally, a broader discussion of AI’s role in agriculture, including crop yield forecasting tools, was recently featured in UC Davis’ article “Big Data Comes to Dinner.”

Integrating Satellite, Climate, Water, and Soil Data

To build this tool, the UC team pulled from multiple high-quality data repositories. Monthly satellite imagery came from the Landsat Program, while daily weather data was sourced from the Daymet Climate Database. Crop water use data was derived from OpenET, a NASA-supported evapotranspiration monitoring platform.

These data layers were processed using a multimodal Vision Transformer deep-learning model designed specifically for agricultural use. The tool was trained using publicly available yield data from the USDA National Agricultural Statistics Service (NASS), covering the years 2008 to 2022.

The result was a forecasting model with an average R² score of 0.76, demonstrating a strong correlation between predicted and actual yields across the state’s top commodity crops.

Benefits for California Growers

For growers, the ability to access reliable, localized yield forecasts months before harvest could be transformational. Early predictions allow for better decision-making around irrigation, labor planning, fertilizer application, and harvest logistics. The model’s inclusion of evapotranspiration data is especially valuable in water-constrained areas of California, helping optimize inputs and manage drought risk.

This AI forecasting system could work in tandem with existing tools such as CropManage, a UC Agriculture and Natural Resources (UC ANR) program that supports irrigation and nutrient management. By layering in forecasts from the new model, growers can fine-tune their strategies with both historical insight and predictive power.

Expert Perspective

Mason Earles, assistant professor at UC Davis and one of the lead researchers on the project, emphasized the importance of accurate yield prediction for operational planning, saying,“Predicting what yields you're going to have at the end of the season, no one is that good at it right now. But it's really important because it determines how much labor contract you're going to need and the supplies you'll need for making wine.”

Impacts Beyond the Farm

The model’s applications extend beyond production agriculture. Crop insurers, commodity traders, and government agencies can use its predictions to guide risk assessments, resource allocation, and market projections. Organizations such as the California Department of Food and Agriculture (CDFA) and USDA Risk Management Agency may benefit from such tools when issuing emergency declarations or adjusting insurance rates based on projected harvest shortfalls or surpluses.

Because the tool is open source, researchers and policy analysts can access the code and datasets directly through GitHub. This transparency invites collaboration and encourages future adaptations, including expansion into other U.S. growing regions or crop-specific refinement.

Looking Ahead

The UC research team is already exploring next steps, including incorporating pest and disease surveillance data and real-time ground observations from growers. Future upgrades could include integration with GPS-based yield monitors and variable rate application systems, tying the forecasting tool directly into the growing precision agriculture ecosystem.

As California continues to grapple with water scarcity, labor shortages, and climate volatility, tools like this will play an increasingly critical role in helping farmers plan, adapt, and remain competitive.