Article - Mapping Coastal Marine Habitats with RGB and Multispectral UAS Imagery to Support Seaweed Aquaculture Management and Ecosystem Conservation

DATA: Othomosaic (multispectral) - mapping of the study site. 

* correspondence: isabel.urbina-barreto@univ-reunion.fr; isabelurbinab@gmail.com

Highlights of the study 

What are the main findings?

• RGB and multispectral (MS) UAS imagery, combined with a Random Forest classification model, enabled the distinction between seaweed farms and surrounding benthic habitats and achieved an increased habitat mapping accuracy of 87%, Kappa = 0.82.

• RGB imagery enabled the classification of seaweed cultivation farms (Kappaphycus alvarezii), green algal bloom, and sandy bottoms with an accuracy of over 85%. A significant difference was shown between RGB and MS imagery performance (85% vs. 89%). However, RGB imagery alone may provide a satisfactorily accurate and more cost-effective alternative for operational monitoring of shallow habitats under optimal illumination and water clarity conditions.

What are the implications of the main findings?

• This study highlights the central role of visible-spectrum information and supports the use of a cost-effective RGB approach for operational aquaculture monitoring.

• This approach supports integrated marine spatial planning by enabling regular assessment of aquaculture interactions with sensitive coastal marine habitats.

Abstract

Madagascar’s expanding blue economy is largely underpinned by seaweed aquaculture, particularly Kappaphycus alvarezii (Cottonii), which offers an alternative to declining small-scale fisheries and strengthens the resilience of coastal socio-ecosystems. Ensuring the sustainability of this economic activity requires effective ecological monitoring of aquaculture sites and surrounding habitats. This study examines and compares the performance of two imaging configurations—an RGB composite derived from a subset of multispectral images capturing red (650 nm), green (560 nm), and blue (450 nm) bands; and a five-band multispectral (MS) image encompassing blue, green, red, red-edge (730 nm), and near-infrared (840 nm) bands—combined with a Random Forest (RF) classification model, for benthic habitat mapping in a seaweed cultivation context. High-resolution orthomosaics (2 cm/pixel) enabled the discrimination of Kappaphycus cultivation plots from three shallow-water habitats: (i)‘benthic macrophytes’, which comprise: seagrass meadows and benthic macroalgal; (ii) ‘sandy bottom’ and (iii) ‘green algae’. The RF classification achieved an overall accuracy of 87% (Kappa= 0.82) across ~10 hectares. Producer’s accuracy exceeded 80% for Kappaphycus cultivation, green algae, and sandy bottom for both the RGB and MS datasets, indicating strong classification performance. However, early-stage seaweed was occasionally misclassified as benthic macrophytes, likely due to its low biomass and weak spectral signature. This UAS-based approach provided a robust and cost-effective framework for monitoring off-bottom seaweed farms and associated natural habitats. This approach supports sustainable aquaculture development and integrated coastal management in Madagascar and comparable tropical reef socio-ecosystems.

Keywords: UAS imagery; seaweed aquaculture; Kappaphycus alvarezii; Cottonii; RGB and multispectral imagery; random forest classification model; green algal blooms; benthic habitat mapping; Madagascar