The marine defaunation began in the 19th century with industrial whaling and was accentuated byhigh seas fishing fleets which depleted populations of several marine species. Even though efforts are taken to establish no-take Marine Protected Areas (MPAs), the most critical habitats for the biodiversity of marine megafauna remain poorly known and understood. We propose to overcome this limitation by developing a novel, rapid and non-invasive environmental DNA (eDNA) metabarcoding approach specifically targeting marine vertebrates. We aim to collect many marine species to make a genetic reference database allowing to monitor in routine marine megafauna at a large spatial scale and at a high temporal frequency. Then we will apply this method to assess the biodiversity of vertebrates in 200 sites across all tropical seas in synchrony with video analyses. We will test our core hypothesis that seamounts and islands isolated from humans are the last refugia for marine megafauna.