Mangrove3D: Terrestrial Laser Scanning Dataset for Coastal Mangrove Forests

Mangrove3D is a terrestrial laser scanning (TLS) dataset collected in spring 2024 on Babeldaob Island, Palau (7°31′49″N, 134°33′53″E). The study sites are coastal stands of Rhizophora mangroves with dense prop-root networks and multilayered canopies. Data were acquired using a Canopy Biomass LiDAR v2.0 (CBL), based on a SICK LMS 151 scanner (905 nm), mounted on a rotating stage and inverted on a surface elevation table (SET) arm to maximize ground returns. Each scan covers 360° × 270° at 0.25° angular resolution, captured in 33 s. At each benchmark site, 8 scans were collected at 45° intervals; scans with obvious errors were removed. The dataset contains 39 scans from 7 benchmarks (31.3 M points; 0.5–0.9 M per scan). Points are labeled into six classes: Ground & Water, Stem, Canopy, Root, Object, and Void. We recommend using 30 scans (sites #1–5) for training/validation and 9 scans (sites #6–7) for testing. Mangrove3D provides one of the first open TLS datasets targeting mangrove ecosystems, offering high-density 3D geometry for segmentation, biomass estimation, and coastal blue-carbon research. Dataset Highlights High-resolution TLS dataset dedicated to Body-level semantic segmentation of mangrove forests – one of the first open-access 3D datasets capturing coastal Rhizophora stands. Points annotated into ground, stem, canopy, root, object, and void classes. Benchmark for occlusion-rich forests – inverted scanner geometry provides nadir and zenith coverage in dense, self-occluding vegetation.