Satellite-based tillage residue measurement platform for scalable soil carbon monitoring

Overview

This solution provides scalable, satellite-based detection of tillage residue cover to support soil organic carbon (SOC) monitoring and carbon market frameworks. Tillage practice is a critical input for understanding soil carbon dynamics, yet traditional field surveys cannot scale economically. The platform addresses this gap by combining multi-source Earth observation data with deep learning to deliver field-level tillage information across North American row crop regions.

The technology serves agrifood supply chains, carbon market participants, and sustainability programs that require credible, verifiable data on regenerative farming practices such as no-till management. By enabling continental-scale tillage monitoring with quantified uncertainty, the platform supports the economic viability of SOC measurement and monetization.

Technical specifications

Key capabilities:

• Multi-source Earth observation fusion including multi-frequency radar and optical satellite data processed as virtual constellations
• Deep learning architecture using Mixture Density Networks for tillage residue cover detection
• Transfer learning and generalization techniques to scale region-specific models across continental coverage
• Per-pixel and per-field uncertainty quantification for every classification output
• Gradient boosting machine learning algorithms for generation 1 detection, achieving operational F1 scores of 0.75
• Independent accuracy tracking through windshield surveys, census statistics, and in-field residue measurements
• Field-level resolution targeting counties with at least 10,000 acres of row crops

Validation approach:

• Harmonization of ground-truth data from multiple regional sites
• Integration with process-based models for uncertainty quantification
• Application Readiness Level advancement through pilots funded by NASA and USDA as well as public-private partnerships

Technology readiness level

The platform has advanced beyond initial development through operational pilots supported by NASA, USDA, and public-private partnerships. Generation 1 satellite-derived detection algorithms have been validated across multiple regions with F1 scores of 0.75. The next phase focuses on scaling the product across North America for any county with sufficient row crop acreage, leveraging transfer learning to generalize from regional pilots to continental coverage.