SBIR Phase II: Multi-output Machine Learning Modeling Framework for a Hybridized Perennial Cover Crop for Specialty Crop Systems

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase II project is in mitigating the earth system’s temperature rise. By removing the barriers farmers face to planting permanent cover crops and generating income from the resulting soil carbon gains, this project would make our food supply chain more resilient. Today farmers leave 95% of ~400M US cropland acres bare outside of the cash crop season because the short-term costs, work and risk of annual cover crops outweigh the benefits. The soil carbon gains from cover crop adoption cannot be monetized effectively today because high measurement costs and uncertainty levels depress market prices for carbon removal credits. The innovations- a novel permanent cover crop cultivar and a novel measurement technology - commercialized during this project may contribute to solving these problems by offering a potentially 80-90% reduction in the cost of accurately measuring changes in agricultural fields compared with the status quo of collecting soil samples in every field and sending them to laboratories for analysis. Farmers would also reduce soil erosion, and improve soil health, water utilization and nutrient cycling on tens of millions of acres. More stable, profitable farms will support thriving rural communities. The intellectual merit of this project lies in quantifying changes in agricultural field parameters with precision – a challenging task due to the complex interactions of management practices, crop types, soil types, and climates that occur at specific locations and the spatial variability in parameters throughout agricultural fields. This project introduces an innovative modeling platform that delivers accurate, low-cost predictions in any location for multiple agricultural parameters powered by a single novel model architecture. The architecture leverages the latest discoveries from the deep learning field, such as attention mechanisms, self-supervised learning, and foundation models. The modeling platform will be able to predict levels and changes in Soil Organic Carbon, nitrous oxide emissions, plant water stress (Stem Water Potential) and plant nutrient stress (Soil Plant Development). Thousands of observations in permanent crop fields in California will be collected and analyzed for accurate calibration of the platform. The project will also develop an end-user software application that translates the predictions of the modeling platform into useful features for farmers and caron credit buyers. The project aims to accelerate the adoption of carbon-storing farming practices. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.