<p>Accurate crop classification and crop management event prediction are crucial for food security and sustainable agriculture. However, deep learning models for crop monitoring often underutilize multimodal remote sensing data and struggle with high-resolution, multitemporal features. To address these challenges, we propose DCA-UNet3D, a deep cascade asymmetric network that enhances multiscale feature fusion and preserves temporal information via deepened architecture and asymmetric spatial–temporal processing. Evaluated on the multimodal SICKLE dataset, our model achieves an average F1-score of 90.77%, accuracy of 93.34%, and mIoU of 83.46%, significantly outperforming baseline methods. It also attains lower prediction errors across three key crop management events (sowing, transplanting, harvesting), with notable improvement in harvesting date accuracy. Ablation studies validate the contribution of each component to overall performance.</p>