Skip to main content

ECAAS: From Rice Yields to Business Projections

stock footage cose up of rice


Food security is one of, if not the most, pressing issues facing many African countries today. In sub-Saharan Africa, agricultural system shocks will continue to severely impact smallholder farmers' food security in the coming years. Analyzing the nature and extent of these impacts can become overwhelming tasks with only conventional capabilities like on-the-ground observations and field surveys of farmers. Satellite-based Earth observations (EO), which provide crucial information about crops in near real-time, can play a vital role in supplementing and enhancing such capabilities, enabling accurate estimates of production, earlier warnings of crop failures, and supporting response programs involving risk financing and other measures that reduce food insecurity.


Local farmers in Katavi, Tanzania, weighing final amounts of harvested rice
               Local farmers in Katavi, Tanzania, weighing final amounts of harvested rice

The Enabling Crop Analytics at Scale (ECAAS) project sought to collect rice paddy yield data in Katavia, Tanzania, through a public-private partnership between NASA Harvest, Flamingoo Foods Limited, and the Sokoine University of Agriculture. The project led a two-part data campaign across the 2022 growing season that collected field boundary data and demographic information from over 800 farmers and then precise yield data with over 600 from the original group. Rice is a critical crop in the Katavi region, and Flamingoo Foods, as a leading local rice producer, is seeking to improve operations by improving farmers’ access to storage and markets as well as better prices for their products. This unique partnership not only created workflows for public-private collaboration but aimed to demonstrate the value of machine learning models for optimizing yield data collection.


Accurately determining crop yields at the field scale can help farmers estimate their net profit and access services, including inputs, insurance, markets, and storage. This information is critical for improving service providers' operations, for example, to improve logistics (e.g., for buyers) and ascertain amounts to purchase or payout. When aggregated, crop yield estimates are critical in monitoring food security at national and regional scales. While collecting ground-truth yield data is largely cost-prohibitive, it is the most reliable way to estimate yields at the field level. 


Food insecurity is an increasingly pressing issue in many Sub-Saharan areas, and collecting crop yield data can create better opportunities to predict areas of agricultural surplus and deficit when combined with weather and satellite data. These predictions can then be used to prepare more adequately for crop shortages based on improved yield models and the scaling abilities of the project. The ultimate goal is for this approach to be applied to other crops in regions throughout the rest of the world.


This project used the surveyed yield data along with OpenDataKit (ODK) to take steps forward in machine learning processes for the future of crop analytics. ODK demonstrates the utility of machine learning models for optimizing data collection and can inform and reduce the cost of collecting yield data critical for agricultural decision-making. 




The team collected data in over 800 fields, with 100% of the farmers willing to participate in the data collection and provide consistent communication with the field team. Farmer willingness and trust were imperative to gather harvest date survey information and complete the final harvesting portion of the campaign.


The first survey on field boundaries, which reached 806 households, established rapport with the farmer and enabled the field team member to coordinate the follow-up yield measurement on the exact day of harvest. The survey helped the team anticipate which region would harvest first and when to expect the peak of rice paddy to flood the market.


The project was focused on Katavi, but the workflows can be applied and scaled to much larger regions in the future. As more data becomes available, model results can be better evaluated. Another focus has been testing deep learning approaches, including a Task-informed meta-learning (TIML) model to run a crop-type model over a larger area toward developing future models that learn efficiently from sparse field data. This would be a massive step towards filling out crop-type datasets lacking in data points and using this data to predict crop insecurity and its effects. 


Graphic highlighting different projects participating in the ECAAS initiative from the Closing Convening Event in December 2022.
        Graphic highlighting different projects participating in the ECAAS initiative from the
                              Closing Convening Event in December 2022. 


Closing Meeting


On December 6, 2022, all projects under the ECAAS umbrella participated in convening to close the work and reflect on lessons learned. The in-person event took place in Arlington, VA, with participants visiting from around the world. The “Enabling Satellite-based Crop Analytics at Scale (ECAAS) Initiative is a multiphase project funded by the Bill & Melinda Gates Foundation to unlock the promise of satellite remote sensing for smallholder agriculture. With support from TetraTech, this Initiative seeks to support innovation, community partnerships, and data-sharing infrastructure to collect better, process, and disseminate high-quality georeferenced training data for ML models. 

News Date
Jan 27, 2023
Taryn Devereux, Katie Garcia, Catherine Nakalembe