Winter wheat yield prediction using convolutional neural networks from environmental and phenological data.

Area of Science:

• Agricultural Science
• Machine Learning
• Deep Learning

Background:

• Accurate crop yield forecasting is crucial for food security and agricultural planning.
• Traditional methods often struggle to capture complex interactions between environmental factors and crop growth.
• Winter wheat yield is influenced by a dynamic interplay of genotype, weather, soil conditions, and management practices.

Purpose of the Study:

• To evaluate the performance of machine learning and deep learning models for winter wheat yield prediction.
• To develop and validate a novel Convolutional Neural Network (CNN) model for enhanced yield forecasting.
• To interpret the developed model to identify key environmental drivers of winter wheat yield.

Main Methods:

• Utilized an extensive dataset encompassing weather, soil, and crop phenology variables from 271 German counties (1999-2019).
• Developed a 1-dimensional Convolutional Neural Network (CNN) model to capture temporal dependencies in environmental data.
• Benchmarked the CNN model against eight supervised machine learning algorithms using RMSE, MAE, and correlation coefficient metrics.

Main Results:

• Nonlinear models, including the proposed CNN, Deep Neural Network (DNN), and XGBoost, demonstrated superior performance over linear models.
• The proposed CNN model achieved superior winter wheat yield prediction accuracy, showing 7-14% lower RMSE and 3-15% lower MAE compared to the best baseline.
• Model interpretation using SHAP and force plots identified DUL (Dielectric Utility), wind speed (week 10), and radiation (week 7) as critical predictors.

Conclusions:

• The developed CNN model offers a significant advancement in winter wheat yield prediction accuracy.
• Deep learning approaches are highly effective in modeling the complex relationships influencing crop yields.
• Identifying key environmental drivers provides valuable insights for optimizing agricultural management practices.