In agricultural forecasting, the integration of high-resolution remote sensing (RS) data with advanced machine learning techniques is crucial for capturing the spatial heterogeneity and temporal variability inherent in crop growth. Traditional methods often fall short in accurately modeling these dynamics due to their inability to process complex, large-scale data effectively. The aim of this study is to develop and evaluate ST-Net, a spatio-temporal network that enhances crop yield predictions by harnessing deep learning algorithms to analyze RS data. The primary goal is to achieve unprecedented precision in yield forecasting by effectively integrating diverse datasets, including environmental variables and phenological data. ST-Net utilizes a combination of convolutional and recurrent neural networks to process multi-temporal and multi-spectral RS data. This approach allows the model to learn from both spatial patterns and temporal sequences, capturing the intricate dynamics of crop development. The model’s performance was validated using a comprehensive dataset that included ground truth yield measurements, allowing for rigorous testing against traditional yield prediction methods. ST-Net demonstrated superior predictive accuracy with a success rate of 98%, significantly outperforming existing models. This high level of precision was achieved through the model’s ability to adapt to various crop types and environmental conditions, demonstrating its robustness and scalability. The findings from this study indicate that ST-Net substantially advances the field of precision agriculture. By integrating real-time data, the model not only improves the accuracy of yield predictions but also enhances agricultural planning and resource management. These capabilities are pivotal for addressing the challenges of global food security in the context of evolving climatic conditions and increasing population demands.