Big Data and Machine Learning to Improve European Grapevine Moth (Lobesia botrana) Predictions

Area of Science:

• Agricultural Science
• Data Science
• Entomology

Background:

• Lobesia botrana is a major grapevine pest causing significant yield losses globally.
• Traditional pest modeling, like the Touzeau model, relies on limited variables such as temperature.
• Big Data and Machine Learning (ML) offer potential for enhanced agri-environmental applications.

Purpose of the Study:

• To optimize the Touzeau model for Lobesia botrana using data-driven ML approaches.
• To compare the predictive accuracy of ML models against the classical Touzeau model.
• To leverage ML for improved pest management strategies in viticulture.

Main Methods:

• Collected field observation data and 30 GB of high-resolution weather data (30-min intervals).
• Trained multiple ML models, including a four-layer artificial neural network.
• Validated model performance using F1 scores, comparing ML predictions with the Touzeau model.

Main Results:

• ML models demonstrated significantly higher accuracy (F1 score) than the Touzeau model.
• The best-performing model was an artificial neural network, utilizing multiple variables beyond just temperature.
• ML models effectively identified complex relationships in nonlinear systems for pest prediction.

Conclusions:

• ML techniques provide a powerful tool for optimizing pest management strategies in agriculture.
• Artificial intelligence-based models show substantial benefits for predicting pest behavior, like Lobesia botrana flights.
• Further development of AI in pest modeling can lead to more effective and sustainable vineyard management.