Satellite model maps frost damage across 700,000 hectares of Brazilian corn

Every June, crop specialists in Parana state brace for frost. The risk is baked into the calendar: second-harvest corn, planted after soybeans, hits its most vulnerable growth stages just as cold air masses push northward through southern Brazil. Most years the risk passes. In 2021, it did not.

That season, drought had already delayed soybean planting across Parana, pushing corn planting weeks behind schedule. By the time severe frost arrived in May and June, the crop was nowhere near harvest-ready. Farmers, insurers, and state officials needed to know the scale of the damage - but accurate, rapid assessment proved elusive. On-the-ground inspection by agricultural technicians, effective as it is for individual farms, cannot capture losses across hundreds of thousands of hectares in real time.

Mapping the damage from orbit

A team from Brazil's National Institute for Space Research (INPE) and Sao Paulo State University (UNESP) set out to close that gap. Working with satellite imagery from the European Space Agency's Sentinel-2 mission, they combined multispectral optical data with a Random Forest machine learning algorithm to map more than 700,000 hectares of corn in Parana's western mesoregion - the agricultural corridor centered on Toledo and Cascavel.

The results were clear. The model identified corn crops with 96% accuracy and found that roughly 70% of the mapped area sustained frost damage during that two-month window. The team named their framework GEEadas, referencing the Google Earth Engine platform on which the analysis ran. The findings appeared in the December issue of Remote Sensing Applications: Society and Environment.

"In 2021, we had a drought that disrupted soybean planting in Parana and, as a result, delayed corn planting. Then, in June, came the news of the frost," said Marcos Adami, a researcher from INPE's Earth Observation and Geoinformatics Division. "A crop failure there greatly affects the lives of the people, most of whom depend on agribusiness. Developing this study is a way to provide tools that give answers and contribute to the planning of measures that help maintain this important activity."

Why the second harvest carries higher stakes

Brazil's corn economy has transformed over the past two decades. The 2019/2020 harvest yielded around 103 million tons - double the output of the previous decade - with three-quarters of that volume coming from the second harvest cycle. This expansion was driven by shorter-cycle corn varieties planted after soybeans, but the second crop carries a structural disadvantage: it matures during cooler, drier months when frost risk peaks.

Parana ranks as Brazil's second-largest grain-producing state, trailing only Mato Grosso. The country's 2025 corn production estimate reached a record 141.6 million tons. Nationally, the cereal, legume, and oilseed harvest hit 345.6 million tons in 2025 - 18% above 2024 levels. With global grain production concentrated in just five countries (China, the United States, India, Brazil, and Argentina), harvest disruptions ripple through commodity markets worldwide.

Brazil has frost risk warning systems, but it lacks precise, scalable tools for quantifying damage after extreme events occur. Insurance claims currently depend on agricultural technicians visiting individual farms - thorough, but slow and spatially limited by what one person can walk.

"Producers still face a number of climatic uncertainties during the harvest, especially when there are extreme events, such as frost, which have social, economic, and environmental impacts," said Michel Eustaquio Dantas Chaves, a professor at UNESP Tupa and the study's first author. "This method provides accuracy, indicating the affected area and reducing uncertainties."

Cross-checking against insurance records

To test GEEadas, the researchers compared their satellite-derived damage estimates against two independent datasets: official figures from Parana's State Department of Agriculture and Supply, and loss records from insurance companies. The insurance data is particularly granular - each claim triggers a site visit by a certified agricultural specialist, producing farm-by-farm damage assessments.

The comparison held up. Frost-damaged areas identified by the satellite model closely matched the insurance and government tallies. This convergence matters not just for validating the study but for establishing the method's credibility with the banks, government agencies, and insurance companies that would need to adopt it in practice.

The GEEadas framework was designed with flexibility in mind. Users can adjust input variables to adapt the model to other crop types and growing regions, making it potentially applicable beyond corn and beyond Brazil.

Climate pressure compounds operational risk

The 2021 event was not a single anomaly - it was two anomalies stacked: drought delayed planting; frost then punished the late-stage crop. Brazilian climate scientists have documented increasing frequency and intensity of extreme weather in the country's main agricultural regions, a concern prominent enough to feature in COP30 negotiations held in Belem.

Adami is extending the work through a collaboration with Brazil's National Supply Company (CONAB), developing similar data pipelines for Rio Grande do Sul, Parana, and Sao Paulo states. The goal is practical: give farmers, lenders, and policymakers reliable numbers before the slow process of field inspection is complete.

Some caveats apply. The 96% accuracy figure comes from a single crop, a single season, and one region. Performance may vary with different cultivars, cloud cover interrupting satellite revisits, or frost events that produce more ambiguous spectral signatures than the severe 2021 damage. Broader validation across multiple seasons and geographies would sharpen confidence in the model's general applicability.

What the study demonstrates is that the information gap between an extreme weather event and a reliable damage estimate can be narrowed substantially - from weeks of field surveys to satellite passes measured in days. For an agricultural sector that moves billions of dollars and supports millions of livelihoods, that difference in response time is not trivial.