Deep Learning Models and Their Ensembles for Robust Agricultural Yield Prediction in Saudi Arabia

A crop yield prediction is critical to increase agricultural sustainability because it allows for the more effective use of natural resources, including water, fertilizers, and soil. Accurate yield estimates enable farmers and governments to more accurately manage resources, decreasing waste and minimizing adverse environmental effects such as the degradation of soil and water quality issues. In addition, predictive models serve to alleviate the consequences of climate change by promoting adaptable farming techniques and improving the availability of food by means of early decision-making. Thus, including a crop yield prediction into farming practices is critical for combining productivity and sustainability. In contrast to conventional machine learning models, which frequently require long feature engineering, deep learning may obtain complicated yield-related characteristics directly from initial or merely preprocessed data from different sources. This research paper aims to demonstrate the strength of deep learning models and their ensembles in agricultural yield prediction in Saudi Arabia, where agriculture faces issues such as scarce water resources and harsh climate conditions. We first define and evaluate a Multilayer Perceptron (MLP), a Gated Recurrent Unit (GRU), and a Convolutional Neural Network (CNN) as baseline deep models for the crop yield prediction. Then, we investigate combining these three models based on stacking, blending, and boosting ensemble methods. Finally, we study the uncertainty quantification for the proposed models, which involves a discussion of many enhancements’ techniques. As a result, this research shows that, by applying the right architectures with strong parametrization and optimization techniques, we obtain models that can explain 96% of the variance in the crop yield with a very low uncertainty rate (reaching an MPIW of 0.60), which proves the reliability and trustworthiness of the prediction.