Early warning of mycotoxins in European grain supply chain using machine learning and big data

This project aims to develop an early warning system based on forecasting modelling, using machine learning, big data and (mechanistic) models, to predict – in an early stage – mycotoxins in cereal grains at the regional level in Europe. Data to be used include mycotoxin monitoring data, weather data, agronomy data and satellite images. Models that will be used include crop phenology models, mechanistic models for crop fungal infection and mycotoxin production, and machine learning models. All data and models will be integrated in an early warning system that can be used by multiple stakeholders. Such an early warning of mycotoxin contamination could help the stakeholders to a) collect and analyse samples in a risk-based way, following EC/2017/625, b) take appropriate actions for prevention and control of mycotoxins, and c) to assign the batches as feed or food and decide upon routing and processing as early as possible.

A first attempt of integrating multiple models in a European wide system and with an user-interface has been developed in the H2020 project MyToolBox (Krska et al., 2016; van der Fels-Klerx et al., 2021). In this project, available country-specific models for specific mycotoxin-grain combinations were linked into one (MyToolBox) platform, with one user front-end. In the MytoolBox, several individual, country-specific models were linked into one tool. A tool with a more generic approach, using a European-wide weather database is currently lacking, and will be developed in this project.