URBANA, Ill. — A new AI-based system can generate high-resolution soybean yield maps across Brazil using only limited local data, improving yield estimates for this key agricultural region and potentially providing strategic benefits to global soybean markets.
The newly published work by researchers at the University of Illinois Urbana-Champaign demonstrates an innovative approach that enables high-performance national yield estimates for Brazilian soybeans, even in areas where directly reported local yield data are very limited.
By leveraging knowledge learned from earlier U.S.-based work through so-called "AI transfer learning," the research team was able to make detailed yield predictions at the municipal level using Brazil's state-level soybean yield data. It's one of the first successful nationwide applications of cross-scale AI yield predictions for Brazilian agriculture.
The findings are outlined in a new study published in the International Journal of Applied Earth Observations and Geoinformation .
Addressing a critical global data gap
Although Brazil is currently the world's largest soybean producer and a major global food exporter, high-resolution yield data for Brazilian soybeans remain largely unavailable. These data are essential for precision agriculture, risk management, and sustainability planning, and the data scarcity has hampered scientific understanding of this important agricultural region. Previous crop yield modeling — which relies on coarse state-level data to model finer predictions at the municipal or field level — has demonstrated limited performance nationally.
The Illinois research team developed a new framework to predict national soybean yields at a finer level by integrating satellite observations, climate data, and state-level yield statistics, leveraging AI transfer learning techniques with the knowledge learned from their U.S. based models.
Remarkably, the model for Brazilian soybean achieved strong predictive performance without using any municipal-level yield data. The explained variance (R²), a key measure of effectiveness, doubled in the new model compared to conventional cross-scale studies. When municipal data were included, performance improved further (R² of 0.57), comparable to the best existing approaches that rely on much more abundant data.
The power of transfer learning
A key innovation of the study is the use of AI transfer learning, which allows scientists to reuse existing models rather than starting from scratch in each region. This makes it possible to generate detailed agricultural information in areas where collecting large amounts of local data would be costly, slow, or impractical.
For this work, knowledge from an advanced model that was trained to predict soybean yield in the U.S. was adapted to Brazilian growing conditions. By fine-tuning the U.S. model using only state-level data or sparse municipal-level data from Brazil, the researchers were able to account for differences in climate, crop phenology, and management practices between the two countries.
Spatial maps of the yield data show the harvested-area-weighted average soybean yield across all valid years for each municipality (left); and the standard deviation across all valid years in each municipality. Credit: Paper in the International Journal of Applied Earth Observations and Geoinformation.
First author Jiaying Zhang explained, "This approach boosted the effectiveness of cross-scale yield prediction from 50 percent to 78 percent of the theoretical upper limit, which we defined as the best performance achieved by models trained with highly detailed local yield data. The results demonstrate that AI-driven transfer learning can overcome both data scarcity and scalability challenges in agricultural modeling."
Implications for yield predictions worldwide
The findings arrive at a pivotal moment for global soybean markets.
In 2018, Brazil surpassed the United States to become the world's largest soybean producer for the first time . The ability to monitor and forecast production in detail is essential for understanding global soybean supply as well as the environmental impacts of large-scale agriculture in Brazil. Enhanced predictability of soybean yield will enable more accurate assessments of supply-demand relationships, land-use change, and soil health impacts at scale for more informed decision-making.
"The ability to monitor and anticipate crop production regionally and globally with high fidelity is strategically important for market analysis, trade forecasting, and risk assessment for U.S. soybean producers," said the project lead and senior author Kaiyu Guan, Levenick Endowed Professor and Director of the Agroecosystem Sustainability Center at Illinois.
The study provides a pathway for applying advanced yield modeling in regions of the world with limited data, supporting food security planning, climate risk management, and evidence-based agricultural policy. By leveraging models trained in data-rich regions and adapting them to areas where data are scarce, the approach opens new opportunities for cost-effective, global-scale agricultural intelligence.
The study is titled "Transfer learning for improved crop yield predictions in a cross-scale pathway: a case study for Brazilian national soybean" ( DOI: 10.1016/j.jag.2025.104981 ).
The work was supported by the National Science Foundation and the U.S. Department of Agriculture.
About the Agroecosystem Sustainability Center
The Agroecosystem Sustainability Center (ASC) advances research that strengthens agricultural productivity while sustaining the ecosystems that support food systems by connecting science with real-world application. ASC is a joint initiative of the Institute for Sustainability, Energy, and Environment (iSEE), the College of Agricultural, Consumer and Environmental Sciences , and the Office of the Vice Chancellor for Research and Innovation at the University of Illinois Urbana-Champaign.
