Forest management and ecological conservation increasingly depend on precise and efficient remote sensing techniques to understand and preserve forest ecosystems. This paper introduces a novel approach that combines advanced deep learning architectures with synthetic tree point cloud datasets for automated tree part segmentation. Using SpeedTree for procedural modeling, we generate a diverse array of synthetic trees complete with detailed, per-point segmentational annotations, thus eliminating the need for labor-intensive manual labeling. These synthetic datasets are employed to train and evaluate several notable deep learning models, including PointNet, PointNet++, ShellNet, DGCNN, and PVT. We assess these models based on their ability to segment tree components—trunks, branches, and leaves—from isolated point clouds. Our results demonstrate that the trained models achieve high segmentation accuracy, showcasing the potential of synthetic datasets to enhance remote sensing tasks in forestry. This study not only highlights the benefits of integrating synthetic data with machine learning but also sets the stage for future applications, including the use of transfer learning to adapt these models to real-world forest environments.