Deploying satellite-imagery based machine-learning models for large-scale mapping of tillage practices

State-wide mapping of soil-health indicators and related agricultural practices is key to tracking our progress towards improving sustainability of agriculture. While this information and low-cost platforms to gather this information for large areas are currently lacking, there have been some recent successes such as satelliteimagery based tillage class mapping for the United States Corn Belt. Encouraged by these recent successes, we propose to leverage advances in open-source satellite imagery and develop and evaluate a prototype machine-learning platform to classify eastern Washington State’s agricultural fields into three classes of tillage practices: “no till”, “reduced till” and “conventional till”. Given the need to undertake groundtruth data collection, our primary focus will be on wheat growing areas of eastern Washington State to keep the project tractable. However, we will evaluate the generalizability of our wheat-based machine-learning method across diverse cropping systems and take steps to maximize generalizability. The proposed project will complement ongoing efforts of the Washington State Soil Health Initiative, and provide a prototype platform that can be collaboratively expanded in the future to facilitate low-cost state-wide automated measurement and monitoring of a broad array of soil health indicators and agricultural practices that influence them. Such a platform is critical to quantify the true costs and benefits of agricultural practices, monitor changes over time, evaluate policies and programs designed to incentivize specific practices, and ultimately move the agricultural sustainability needle in the right direction. This is a research proposal involving WSU faculty in collaboration with the Washington State Department of Agriculture and Conservation Districts.