Lacuna Fund aims to create the building blocks of labeled training data that allow robust applications, from personalized information on fertilizers and regenerative agriculture practices for farmers, to better information about crop yields and food security to inform decision-makers worldwide.
Agriculture is fundamental to the livelihoods of billions of people worldwide. AI-enabled tools have great promise to increase production and resilience and contribute to broader sustainable development goals, but often crucial training data is not available in the public square.
In the agricultural AI for social good domain, recent advances in the analysis of remote sensing data have enabled improved accuracy of a variety of AI tasks. Both startups and large development programs have built on this data availability to provide personalized information and predictive services for farmers.
However, there is a lack of accurate ground truth labeled datasets in a variety of geographies in the Global South. Techniques and datasets to address unique challenges in mapping intercropping and smallholder farms are out of reach. This hinders progress in building beneficial ML applications that are scalable and robust for populations worldwide.
Alongside an explosion of earth observation providers and services, nascent standards, such as the Spatio Temporal Asset Catalog (STAC) and a preliminary version of best practices for ground reference data collection and cataloguing are creating greater coherency for machine learning-ready training data across the AI for agriculture landscape.
To complement and expand these efforts, Lacuna Fund supports the creation, expansion and maintenance of labeled data in the agriculture space. Our Technical Advisory Panel, who is responsible for identifying data gaps, developing the RFP, and reviewing and selecting proposals, has identified needs for labeled datasets in the following areas. However, Lacuna Fund RFPs are intentionally open, to encourage new and innovative ideas that we may not have identified.
- Farm boundary identification, particularly for smallholder farms.
- Crop type classification, including issues related to intercropping.
- Yield estimation, including particular issues related to estimation for subsistence farmers.
- Other datasets, such as water levels, extreme weather, or agricultural practices, that could build on or complement better field boundary and crop type labels.
- Soil health data and related labels.
- Pest and disease identification training datasets, while widespread for certain staple crops, such as cassava, maize, and wheat, could be further broadened or made more context aware.
- Forage availability and depletion predictions.
- Issues related to pastoral migratory patterns and anticipated conflicts with crop farmers.
- Prediction of the best locations to establish forage banks and support infrastructure along livestock migratory routes.
- Resource availability management in managed ecosystems and carrying capacity estimations.
- Other ideas! See our Grantmaking Philosophy