An explainable Bi-LSTM model for winter wheat yield prediction.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Accurate crop yield prediction is vital for food security and agricultural management.
• Deep learning (DL) models, especially Long Short-Term Memory (LSTM) networks, show promise but lack interpretability.
• Regional-scale application and understanding of DL in crop yield prediction are underexplored.

Purpose of the Study:

• To develop and implement an explainable deep learning model for accurate crop yield prediction.
• To compare the performance of LSTM, 1D Convolutional Neural Networks (1D-CNN), and Bidirectional Long Short-Term Memory (Bi-LSTM) models.
• To utilize interpretability techniques to understand the decision-making process of these DL models.

Main Methods:

• Developed and evaluated three sequential deep learning models: LSTM, 1D-CNN, and Bi-LSTM.
• Applied three explainability techniques: Local Interpretable Model-Agnostic Explanations (LIME), Integrated Gradient (IG), and Shapley Additive Explanation (SHAP).
• Analyzed feature importance for predicting Winter wheat yield.

Main Results:

• The Bi-LSTM model achieved the highest predictive performance (R² up to 0.88) and generalizability.
• Explainability analysis identified enhanced vegetation index (EVI), temperature, and precipitation during later growth stages as critical factors for Winter wheat yield.
• Demonstrated the utility of explainable artificial intelligence (XAI) in understanding model predictions, identifying high/low yield instances, and diagnosing errors.

Conclusions:

• Explainable Bi-LSTM models offer a powerful tool for accurate and reliable crop yield prediction.
• XAI methods provide crucial insights into model behavior, enhancing trust and enabling targeted agricultural interventions.
• This approach advances the application of DL in agriculture by combining predictive accuracy with transparency.