The increasing challenges of global food security, driven by limited natural resources, climate change, and the growing demands of a growing population, require innovative solutions to agricultural decision making. Accurate pre-season crop prediction is crucial for optimizing resource allocation, minimizing risks, and fostering sustainable farming practices. This study presents a hybrid model that integrates CNN and BiLSTM networks to improve crop prediction accuracy by leveraging historical crop rotation data and synthetic field-level representations known as Crop Sequence Boundaries (CSBs). Unlike conventional pixel-based methods, which often struggle to capture complex spatial and temporal dynamics, the proposed model effectively extracts spatial features through CNNs and temporal patterns via BiLSTM, converting raw data into detailed spatial-temporal representations. Robustness and generalizability are ensured using stratified k-fold cross-validation and regularization techniques, such as dropout to prevent overfitting, making the model capable of handling large-scale datasets. Experimental results demonstrate that the hybrid model consistently outperforms existing approaches, delivering more precise crop predictions that closely align with practical agricultural needs. By offering accurate and actionable information, this framework enables better decision making for farmers and stakeholders, supports sustainable agriculture, and paves the way for resilient food systems in the face of future challenges.