Design and evaluation of bayesian optimized hybrid deep learning model for forecasting crop yields using climate...

Area of Science:

• Agricultural Science
• Climate Science
• Artificial Intelligence

Background:

• Accurate crop prediction is crucial due to climate change impacts on agriculture.
• Existing models struggle to capture complex climate influences on crop yields.
• Long-term climate data (1961-2021) on temperature, CO2, and precipitation in Pakistan is available.

Purpose of the Study:

• To develop an advanced AI model for precise crop yield prediction.
• To improve upon existing models in understanding intricate climate-agriculture interactions.
• To forecast future agricultural production for wheat and rice.

Main Methods:

• A hybrid deep learning model incorporating Bayesian Optimization (BO) was developed.
• The model integrates temporal and climatic patterns influencing crop production.
• Performance was evaluated against Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Vector Autoregression (VAR) models.

Main Results:

• The proposed Bayesian Optimized VAR-BiLSTM hybrid model demonstrated superior accuracy and stability.
• Achieved a coefficient of determination (R²) of 0.9611 and a Mean Absolute Percentage Error (MAPE) of 8.01%.
• The model exhibits significant forecasting power for climate-driven crop production.

Conclusions:

• The developed AI model offers a scalable solution for resilient agriculture amidst climate uncertainty.
• Provides data-driven insights for policymakers and agricultural planners to adapt strategies.
• Enhances AI applications in agriculture for improved food security and management.