Location: Food and Feed Safety Research
Project Number: 6054-42000-027-018-R
Project Type: Reimbursable Cooperative Agreement
Start Date: Oct 1, 2025
End Date: Sep 14, 2027
Objective:
Major objective: Development and validation of models to predict mycotoxin outbreaks in Illinois and Texas and their implementation in IPM at region-specific scale.
Objective 1) Retrain our recently published IL models by using daily data from IL and TX to develop state-specific models that predict mycotoxin contamination levels quarterly and alert farmers of environmental risks monthly.
Objective 2) Validate and refine fungal growth and mycotoxin contamination models with empirical field data collected in IL and TX in 2023/2024.
Long-term goal: Use the state-specific models to develop a prototype web-based alert application system for use by stakeholders.This goal is beyond the time-frame of this proposal.
Approach:
Objective 1: Develop state-specific ML models to predict mycotoxin contamination levels, this objective has been an on-going collaborative effort. We have published Illinois-based ML models to predict AFL and FUM, the first of its kind in the USA, by using monthly weather data, we utilized gradient boosting machine learning (GBM) and bayesian network (BN) that can confidently predict with 93% accuracy mycotoxin contamination of corn in Illinois. In this project we propose to use daily weather data to perform weekly mycotoxin risk assessments of IL and TX states by county. We propose to develop models to generate two types of alert systems integrated with Integrated Pest Management, 1) a monthly alert system based on backcasting that detected correlations and patterns of historical weather patterns and risk index features with high and low levels of mycotoxin contamination by using Gradient Boosting models (GBM); 2) a forecasting quarterly system using GBM, Bayesian Network (BN) and Neural Network (NN).
Objective 2: Validate and refine fungal growth and mycotoxin contamination models with empirical field data. Models have been developed utilizing data generated at the county-level, but collection of field-specific data prior to and throughout the growing season will provide information that can be used to further refine models. Multiple fields in different regions varying in environmental and agronomic factors in IL and TX will be selected for soil and crop sampling.