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

Precise crop prediction is now of utmost significance in the era of escalating climate dynamics. In this paper, a new structured Bayesian optimized hybrid deep learning model is proposed by using existing methodologies to make accurate predictions for prominent crops wheat and rice based on long-term climate variables average maximum temperature, Co2 and precipitation from 1961 to 2021 in Pakistan. The primary objective of this research is to enhance previous models that have trouble understanding intricate climate impacts and provide reliable predictions about future agricultural production. Using a deep learning model and a classical statistical approach, we employed Bayesian Optimization to automatically identify the optimal parameters. This improved learning and prediction accuracy for our model. It incorporates both temporal patterns and climatic patterns influencing agricultural production. Forecasting is performed for the 2022-2031 decade and tested against strong metrics such as RMSE, MAE, and R². Compared to known models CNN, RNN, LSTM, GRU, BiLSTM and VAR, the proposed BO-VAR-BiLSTM hybrid model showed to have a better accuracy and stability compared to the baselines. The proposed had a R² of 0.9611 and MAPE of 8.01% show that the model had high forecasting power in predicting climate-driven production of crops. Apart from the scholarly interest, the model provides prescriptive insights to policymakers, agricultural planners, and stakeholders in climate-sensitive areas. By allowing data-driven decision-making. It facilitates adaptive crop management strategies under changing environmental conditions. This study not only pushes the frontier of AI for agriculture but also provides a scalable solution for resilient agriculture in the age of climate uncertainty.