Example impression of Sentinel-2. Image courtesy of ESA. Copyright ESA/ATG medialab.
Over the last year, we’ve been working on a project examining the use of satellite data in UK agriculture. So, our interest was piqued by a report on a recent paper published in the journal “Remote Sensing Applications: Society and Environment” on mapping the impact of frost on corn crops in Brazil.
The paper titled ‘GEEadas: GEE-based automatic detection of adverse-frost stress’ by Chaves and Ademi et al. was in the November 2025 edition of the journal.
In the paper, the Brazilian researchers described how Brazil’s corn production has doubled over the last decade due to an increase in second-season crops, making Brazil responsible for approximately eight percent of global output, with the country being the world’s third-largest producer and second-largest exporter. Approximately seventy-five percent of the corn is from the second season crop, however, these crops are particularly at risk of frost due to reduced water availability – during the 2020/2021 season, two frosts in Paraná state caused crop failure. Given Paraná is the second-largest grain producer in the country, with 2025 harvest estimated to be 141.6 million tons.
Traditional methods for assessing crop frost damage rely on sample surveys. However, as frost changes the leaf’s chlorophyll and β-carotene content, it also alters the plant’s spectral reflectance across optical wavelengths, allowing the use of spectral indices to identify the frost-damaged crops. The team combined Sentinel-2 optical satellite data with a Random Forest machine learning approach to map corn areas and detect frost damage.
Validation was undertaken by comparing the research team’s results with official data provided by the State Department of Agriculture and Supply, alongside information from insurance companies relating to farmers losses caused by adverse events.
The team mapped the second season corn in in Western Paraná and then identified the healthy and frost-damaged corn. The corn mapping achieved an overall accuracy 98%, covering a total of 740,007 hectares, which was 1.7 % larger than the official data which was achieved through ground surveys. While the second step identified that 70% of the corn was affected by frost, using a method the researchers named as GEEadas.
While the model was focused on crops, the researchers believe that given the customizable variables they used, the model had the potential to be used for other crops in different circumstances.
Using satellite data to map corn crops and detect frost damage, has the potential to provide better estimates of expected harvests, which can reduce uncertainty and food insecurity, which will support the food supply chain.
Summary
It is fantastic to see innovative uses for satellite data are being applied within the agricultural sector, and it would be interesting to see whether this model could be applied to other countries such as the UK, alongside other crops.
