Remote sensing of coral reef habitats in Madagascar using Sentinel-2 satellite images

Abstract

Publicly available Sentinel-2 satellite imagery was used to map the coral reef systems of Toliara in Madagascar, to standardize methods for monitoring reef health and guiding management decisions. Fieldwork conducted between March and December 2021 used georeferenced photoquadrats to assess benthic structure. The satellite image classification was based on the Object-Based Image Analysis (OBIA) and machine learning algorithms, with k-NN achieving the highest overall accuracy at 83 %, followed by the Bayes classifier (79 %), DT (68 %), RT (67 %) and SVM (42 %). The analysis identified distinct surface areas occupied by seagrass (21 km2), sand (73 km2), rubble (21 km2), coral (10 km2) and algae (6 km2). Comparative assessment with the Allen Coral Atlas underscored the importance of aligning satellite image analysis with in-situ data. The study emphasized the role of selecting appropriate classifier algorithms for precise mapping and stressed the importance of local data collection for accurate habitat mapping. It also showcased the successful application of OBIA with satellite imagery and field data for coral reef mapping, providing insights into habitat health and spatial changes essential for effective conservation.