University of Maryland (NASA Harvest): Satellite Earth observations for food security

Increasing food production sustainably under increasingly variable weather, a warming climate and population growth is one of the most pressing challenges of today. Big-dollar food security decisions would benefit enormously from timelier, more accurate crop information particularly in the context of smallholder farming. However, huge uncertainties remain regarding where and when food is grown as well as how much will be produced, early warning of crop shortfalls, where (and why) there are yield gaps and how conflicts around land cover will impact production. Recent advances in satellite technology – and the integration of such technology into existing monitoring systems and agricultural decision-making processes across sectors – has the potential to address most of these uncertainties.

The project started with implementation of a prototype in Kenya intended to enhance the cost-effectiveness and empirical basis of the Kenya Crop Insurance Programme, which supports 300 000 smallholder farmers. To test the generalizability of the approach to a very different context, it will then be adapted and evaluated other contexts.

A key outcome of the project was the development of an open-source, portable and cloud agnostic (containerized) machine learning-based tool - EO-FARM - for deriving key agricultural variables from satellite imagery, including cultivated area, type, condition, and yield. This tool could be used off-the-shelf for future growing seasons or fine-tuned with user-specific training data or crop types and will be refined and modified throughout the project and after.

1. To prepare for scaling and deployment of the approach elsewhere, the University of Maryland NASA Harvest project team created business case with the support of policymakers in Kenya, which will included the: Investment required for deployment
2. Estimated cost and time savings, compared to traditional methods, for assessing crop conditions and yields
3. Benefits of improved accuracy and timeliness for policymaking and policy instruments such as crop insurance programmes
4. Benefits for insurance and agricultural tech development

NASA Harvest successfully developed the Earth Observation for Field-level Agricultural Resource Mapping (EO-FARM) system, it can be found here. This innovative machine learning pipeline leverages satellite data and cloud computing to generate critical agricultural monitoring products, including cropland maps, crop type maps, and crop yield forecasts. EO-FARM has been deployed in Kenya, providing valuable insights for agricultural monitoring and policy-making. The system's scalability and flexibility have enabled its application across diverse geographies, including Tanzania, Malawi, Uganda, Zambia, Senegal, Rwanda, and even Maui County in the United States. EO-FARM's implementation has demonstrated substantial cost savings, improved data quality, and expanded capabilities, making it a valuable tool for data-driven, climate-resilient agriculture policies.

• Nasaharvest GitHub - nasaharvest/openmapflow: Rapid map creation with machine learning and earth observation data
• Scalable Cropland Mapping Workshop
• Publications - NASA Harvest

The copyright for all images displayed lies with University of Maryland (NASA Harvest)