Accurate flight trajectory forecasting is crucial for air traffic management and UAV navigation. However, the task is highly challenging due to complex spatiotemporal dependencies in long sequences and computational constraints. Thus, we propose FDMamba-Net, a novel deep learning-based framework for trajectory prediction that achieves both high efficiency and precision. It combines feature decoupling, spatiotemporal attention fusion, state - space modeling, and spatial - enhanced decoding. The framework first uses a feature decoupling module to divide raw inputs into independent subspaces to eliminate feature interference. Then, spatiotemporal attention fusion is applied to model dynamic temporal relationships. Next, multi-layer Mamba blocks conduct multi-scale temporal modeling. Finally, spatial features are optimized through adaptive weight - enhanced decoding. Experiments show that FDMamba-Net surpasses baseline methods. Compared to Transformer-based approaches, it achieves better performance with 55.5% fewer parameters, making it a promising solution for efficient trajectory prediction.