Frontiers in Marine Science (Nov 2021)
When Imagery and Physical Sampling Work Together: Toward an Integrative Methodology of Deep-Sea Image-Based Megafauna Identification
Abstract
Imagery has become a key tool for assessing deep-sea megafaunal biodiversity, historically based on physical sampling using fishing gears. Image datasets provide quantitative and repeatable estimates, small-scale spatial patterns and habitat descriptions. However, taxon identification from images is challenging and often relies on morphotypes without considering a taxonomic framework. Taxon identification is particularly challenging in regions where the fauna is poorly known and/or highly diverse. Furthermore, the efficiency of imagery and physical sampling may vary among habitat types. Here, we compared biodiversity metrics (alpha and gamma diversity, composition) based on physical sampling (dredging and trawling) and towed-camera still images (1) along the upper continental slope of Papua New Guinea (sedimented slope with wood-falls, a canyon and cold seeps), and (2) on the outer slopes of the volcanic islands of Mayotte, dominated by hard bottoms. The comparison was done on selected taxa (Pisces, Crustacea, Echinoidea, and Asteroidea), which are good candidates for identification from images. Taxonomic identification ranks obtained for the images varied among these taxa (e.g., family/order for fishes, genus for echinoderms). At these ranks, imagery provided a higher taxonomic richness for hard-bottom and complex habitats, partially explained by the poor performance of trawling on these rough substrates. For the same reason, the gamma diversity of Pisces and Crustacea was also higher from images, but no difference was observed for echinoderms. On soft bottoms, physical sampling provided higher alpha and gamma diversity for fishes and crustaceans, but these differences tended to decrease for crustaceans identified to the species/morphospecies level from images. Physical sampling and imagery were selective against some taxa (e.g., according to size or behavior), therefore providing different facets of biodiversity. In addition, specimens collected at a larger scale facilitated megafauna identification from images. Based on this complementary approach, we propose a robust methodology for image-based faunal identification relying on a taxonomic framework, from collaborative work with taxonomists. An original outcome of this collaborative work is the creation of identification keys dedicated specifically to in situ images and which take into account the state of the taxonomic knowledge for the explored sites.
Keywords
