Optimising Crop Protection with AI-Driven Residue Management - The ResiYou Platform
For fruit and vegetable growers, the challenge is to protect their crops from pests and diseases while complying with government and retailer residue limits (secondary standards). ResiYou, an AI-driven cloud-based platform, was developed to address this challenge, offering advanced residue management and predictive capabilities to help growers strike a balance between crop protection and residue compliance.
Growers face a “growers dilemma”: determining the optimal amount of spraying to combat pests and diseases effectively without exceeding legal and offtakers residue limits. This involves making critical decisions on when and how much to spray to meet EU regulations, retailer’s requirements and their own crop protection needs.
The project required predicting residue profiles over time from incomplete series of samples, presenting a significant challenge due to the lack of complete data. A low signal-to-noise ratio in the data received from commercial growers made it difficult to discern meaningful patterns and required sophisticated analytical techniques. Furthermore, the complexity of chemical-environmental interactions affecting residue levels added another layer of difficulty in making accurate predictions. Making predictions with a degree of uncertainty was essential, given the variability inherent in the agricultural environment and the highly unbalanced data across various dimensions.
For this initiative, a rich database of field trials with residue measurements dating back to the 1970s was utilised, providing a historical context for the analysis. This was complemented by new data collected and processed from commercial growers and included official residue reports from regulatory authorities. Information on the chemical properties of pesticides and environmental factors, along with historical weather data, was also integrated to factor in the environmental impact on residue levels.
The project harnessed the AWS cloud infrastructure to provide the necessary scalable computing resources, ensuring that the data-intensive tasks could be managed effectively. Gradient boosted tree algorithms were employed for their proficiency in regression and classification tasks, enabling accurate predictions of residue levels. MLOps practices were integrated to streamline the machine learning lifecycle from data preparation to model deployment, ensuring that the models remained accurate and relevant over time. Additionally, CI/CD pipelines were implemented, facilitating continuous integration and deployment, which enhanced the development process and ensured robust, up-to-date application updates.
The ResiYou platform provides a user-friendly interface where growers can input their data and receive precise residue profiles for the applied ingredients at harvest. The tool allows for: Planning optimal spraying schedules. Real-time residue level predictions. Compliance checks against EU regulations and retailer standards. Decision support for contractual commitments with retailers.
