Sustainable Marine Structures

Determination of Coral Distribution Using Multispectral Images from an Unmanned Aerial Vehicle in Bach Long Vy Island, Vietnam

Thao Nguyen Van

Institute of Oceanography – Haiphong Branch, Vietnam Academy of Science and Technology, Haiphong 04219, Vietnam

Hung Vu Manh

Vu Manh

Ve Nguyen Dac

Institute of Oceanography – Haiphong Branch, Vietnam Academy of Science and Technology, Haiphong 04219, Vietnam

Nhon Dang Hoai

Institute of Oceanography – Haiphong Branch, Vietnam Academy of Science and Technology, Haiphong 04219, Vietnam

Tuong Bui Manh

Institute of Oceanography – Haiphong Branch, Vietnam Academy of Science and Technology, Haiphong 04219, Vietnam

Lan Tran Dinh

Institute of Oceanography – Haiphong Branch, Vietnam Academy of Science and Technology, Haiphong 04219, Vietnam

Chris Gouramanis

Department of Climate Change, Energy, the Environment and Water, Australian Government, Canberra 2601, Australia

DOI: https://doi.org/10.36956/sms.v8i3.3360

Received: 27 May 2026; Published: 11 August 2026

Abstract

Coral reefs are among the world's most valuable marine ecosystems, providing critical ecological functions, supporting biodiversity, and protecting coastlines. Accurate mapping of coral reef distribution is essential for effective conservation and management, particularly within Marine Protected Areas (MPAs). This study mapped coral distribution in the northwestern waters of Bach Long Vy Island, Vietnam, using ultrahigh-resolution multispectral imagery acquired by a DJI Phantom 4 Multispectral Unmanned Aerial Vehicle (UAV). The imagery comprised five multispectral bands and RGB with a spatial resolution of 7 cm/pixel over an area of approximately 492,360 m2. Water-column effects were corrected using the Lyzenga method, and benthic habitats were classified using a supervised maximum likelihood algorithm. A total of 259 field survey points are used for training and validation of the classification results. Three major benthic substrate classes were identified: coral, sand, and rock. Coral reefs covered 199,410 m2 (40.50%) of the surveyed area, while sand and rock occupied 66,100 m2 (13.43%) and 226,850 m2 (46.07%), respectively. The classification achieved an overall accuracy of 86.87% with a Kappa coefficient of 0.78, demonstrating good agreement between image classification and field observations. The results confirm that UAV-based multispectral remote sensing provides an accurate and cost-effective approach for mapping shallow coral reef habitats. The proposed workflow offers an efficient tool for routine reef monitoring and supports evidence-based conservation, adaptive management of MPAs, and long-term assessment of coral reef responses to increasing environmental pressures and climate change.

Keywords: Coral Reef Mapping, UAV Multispectral Imagery, Water-Column Correction, Lyzenga Method, Maximum Likelihood Classification, Marine Protected Area

References