Machine learning driven pest management optimization for tobacco and similar cropping systems

Overview

This research applies machine learning methods to a large multi-year dataset of insect abundance to identify how pesticide applications affect both target pests and beneficial non-target organisms at the whole-farm scale. By uncovering temporal relationships between early-season pesticide use and late-season insect populations, the approach enables growers and crop advisors to tailor pest management strategies that maximize beneficial outcomes and minimize unintended negative consequences. The work is particularly suited to tobacco production, where the pest complex is small and well-characterized, and is expected to extend to other cropping systems with similar defined pest profiles.

Technical specifications

• Uses machine learning analyses applied to approximately 3.5 million existing insect abundance data points collected across multiple years in tobacco fields
• Identifies similarity between nearby fields within a farm, with demonstrated results showing that insecticide application decisions made using scouting data from a single field were appropriate for all fields within 750 meters 91% of the time
• Determines whether late-season insect populations, including hornworm pests and stilt bug predators, are negatively impacted by early-season pesticide use
• Pinpoints which specific pesticide materials are associated with observed effects on temporally distant non-target populations
• Plans to expand the database with more than 10 additional years of smaller-plot on-farm and research station data to strengthen model robustness
• Designed for cropping systems with a small, well-defined key pest complex, enabling focused and interpretable analyses

Technology readiness level

The platform has reached an early validation stage. Initial machine learning methods have already demonstrated the ability to determine similarity between nearby fields within a farm and to evaluate the appropriateness of insecticide application decisions based on scouting data. The next phase of validation will apply these methods to determine whether late-season hornworm pest populations and stilt bug predator populations are negatively impacted by early-season pesticide use, and to identify which pesticide materials are responsible. Observational data suggest these effects exist, and database expansion is planned to support broader and more robust conclusions. The work is currently at the research and proof-of-concept stage, with future steps focused on expanded validation and translation into actionable pest management recommendations.