Seagrass habitat mapping using UAV imagery in a port environment in Sabah, Malaysia

Seagrass habitats in port environments are exposed to high anthropogenic pressure; however, baseline spatial information is often limited at the site scale required for environmental assessment and monitoring. This study mapped benthic habitats in a working port using ultra-high-resolution UAV imagery and in situ reference observations. UAV images were processed into a georeferenced orthomosaic at 1.31 cm per pixel, covering 9.364 ha. Object-based image analysis and supervised classification were used to map six benthic habitat classes: seagrass, mixed seagrass, coral, coral rubble and coarse sand, fine sand, and deep water. Fine sand was the dominant class (3.819 ha; 41%), followed by coral rubble and coarse sand (3.597 ha; 39%). Seagrass habitats covered 0.522 ha (~6%), comprising 0.304 ha of dense seagrass meadows dominated by Enhalus acoroides and 0.218 ha of mixed seagrass assemblages dominated by Cymodocea serrulata and Halophila ovalis . Accuracy assessment yielded an overall accuracy of 88.71%, with high chance-corrected agreement (Kappa = 0.843; Tau = 0.865). Misclassifications were concentrated in the vegetated transition zones and along boundaries with unconsolidated substrates. The results demonstrate that UAV-based mapping can provide a repeatable baseline for port-scale habitat assessment and targeted monitoring, while remaining sensitive to depth, water clarity, and sun glint conditions.