A new study in Proceedings of the National Academy of Sciences suggests that commercial satellite image archives should be opened to enhance research on the United Nations' Sustainable Development Goals (SDGs). Access to very-high-resolution (VHR) satellite imagery is crucial for monitoring smallholder farms, particularly in low- and middle-income countries where such farming is vital for food security.
"Smallholder farmers, who often cultivate less than two hectares of land, produce more than 30%of the world’s food.Yet their farming practices and productivity are poorly documented, partly due to the lack of accessible VHR satellite images, says Felicia O. Akinyemi, Associate Professor of Geomatics at Karlstad University. These images, which can cost up to 2 euros per square kilometer, are often too expensive for research institutions working in these regions."
Benefits of open satellite data
The researchers point out that initiatives such as the US Landsat, EU Copernicus, and NICFI have shown how open satellite data can drive innovation in sustainability research. They propose that similar models should be applied to make VHR data available for non-commercial research, which could significantly improve the monitoring of smallholder farming and contribute to achieving several sustainability goals, including Goal 2: Zero Hunger.
"My research focuses on Earth observation using satellite imagery and machine learning to explore how changes in land use relate to degradation processes in coupled human and natural systems, such as agroecosystems, says Felicia O. Akinyemi. My interest was sparked by the expansion of agricultural land and the simultaneous loss of high-quality farmlands to urbanization in many parts of the world."
With a particular focus on West Africa, where agricultural expansion frontiers are of globalnote, Felicia was awarded the EU Marie Skłodowska-Curie Individual Fellowship in 2022 to conduct research within the LucFRes project.
Challenges in monitoring smallholder farming
Monitoring smallholder farming systems using satellite data presents several challenges. One major difficulty is that machine learning algorithms require field-verified data, which is often lacking in regions dominated by smallholder agriculture. Without reliable training data, model predictions become weak.
"Furthermore, field sizes on smallholder farms are often extremely small — ranging from under 0.25 hectares up to 5 hectares — meaning that many publicly available satellite images have too low a resolution to accurately map crop types, particularly in systems where multiple crops are intercropped," says Felicia.
Felicia’s research contributes to development goals related to sustainable agriculture, particularly Goal 2: Zero Hunger and Land Degradation Neutrality indicator of Goal 15.3. In an ongoing project, monocultures and intercropping of maize and cassava farming systems are being mapped and predicted using satellite data from two growing seasons in the Guinea Savanna of southwestern Nigeria.
Combining data sources for improved mapping
"Due to frequent cloud cover during the growing season, we combined Sentinel-2 and Sentinel-1 radar data to analyze spectral-temporal patterns on a monthly and bi-monthly basis during the growing season. Using satellite images with sub-meter resolution could have significantly improved the mapping," she explains.
In a pilot study in southwestern Nigeria, the project investigates how changes in land use affect the resilience of farming systems in a changing climate. It combines satellite-based analyses of land use change with local stakeholders' perceptions of future land use. This approach provides a deeper understanding of how agricultural adaptability can be strengthened.