Deploying Satellite-Imagery Based Machine-Learning Models for Large-Scale Mapping of Tillage Practices: Progress Report

This BIOAg progress report documents development of a prototype satellite-imagery and machine-learning platform to classify eastern Washington agricultural fields by tillage practice. The project aims to enable scalable, low-cost monitoring of soil health–related practices by distinguishing no-till, reduced-till, and conventional till systems.

During Fall 2021, approximately 400 fields were surveyed to build a ground-truth dataset. A graduate student developed a photo-based machine-learning model capable of estimating crop residue percentage with approximately 82% accuracy, supporting improved field validation. Work is ongoing to design and refine satellite-imagery–based models for automated tillage classification across wheat-growing regions, with additional testing planned across multiple cropping systems to improve generalizability.

This project provides foundational tools for tracking agricultural practice adoption, monitoring change over time, and informing sustainability assessments and policy evaluation efforts.