Comparison of supervised classifications to discriminate seaweed-dominated habitats through hyperspectral imaging data
Abstract
Intertidal macroalgae define complex habitats and play a key role in structuring coastal areas. While, they are primarily studied during field campaigns, remote sensing acquisitions are becoming increasingly prevalent. However, the use of hyperspectral imagery on drones is not developed even though it allows species identification even in heterogeneous environments such as intertidal rocky shores. Based on hyperspectral drone imagery acquired in summer 2021, this study aims to identify and validate an algorithm suitable for easy integration into an operational framework for monitoring macroalgal dominated shore. The study focuses on two sites along the Brittany coast (Western France). Species identification and abundance were determined in the field. Six algorithms were tested: Mahalanobis, Minimum Distance, Maximum Likelihood, Random Forest, Spectral Angle Mapper and Support Vector Machine. Classifications showed overall accuracies ranging from 70% to 90% depending on the algorithm. The Maximum Likelihood is retained as it provides good accuracies and valuable information about the species distributions. Our analyses based on a combination of field and remote sensing data reveals globally consistent results when considering the main Phaeophyceae species but a divergence was highlighted for Rhodophyta. Despite environmental differences, the two studied sites were faithfully characterized in terms of intertidal species and habitat distribution, highlighting the potential of hyperspectral drone imagery to better understand seaweed-dominated ecosystem dynamics.