Secure Data Systems

Farmers collect valuable data about their operation but struggle to derive full value from it. A recent NASA Acres survey (>1000 U.S. farmers) found that while over 50% of respondents agreed that satellite data would be useful to their farm, only 27% agreed that the benefits of sharing their data were worth the risk, citing privacy concerns, loss of control, and financial repercussions.

Meanwhile, agritech companies and researchers develop models that pair satellite and farmer data to answer important production and environmental management questions, but often guard them for intellectual property (IP) reasons, limiting trust in the validity of these tools. This disconnect impedes progress toward productive and sustainable agriculture. A secure system that can produce the required knowledge without farm data being passed to the modelers and model IP being passed to the farmers is needed to overcome this substantial hurdle.

NASA Acres and the Harvest Sustainable and Regenerative Agriculture Initiative have partnered with Microsoft to develop and successfully test the first-of-its-kind agricultural implementation of their Confidential Clean Rooms (CCR) framework. The CCR uses access-controlled environments where data and models are encrypted and never leave their origin. A designated “Operator” creates a “Consortium”, invites “Members” (modelers and data providers), and authors a “Governance Contract” specifying who sees what and what enters and exits the CCR. Once Members consent to the Contract, the CCR is deployed via Azure Resource Manager (ARM) templates, and all parties can see ARM’s confirmation that the Contract’s policies have been enforced.

Our CCR test successfully ran both a machine learning (random forest) and a process-based (Daycent) model calibrated by and validated against real-world soil organic carbon (SOC) data from 44 farms across the U.S., with privacy and IP preserved for all. This critical milestone in collaboration between farm, agritech, research, and computer science communities provides the basis for assembling farm and other confidential environmental datasets as well as research and commercial models for validating and improving models, and for creating information systems that strengthen agricultural productivity, sustainability and resilience. 

This work is led by Guanyuan Shuai, Ritvik Sahajpal, and Alyssa Whitcraft.